data/sci_erc_test.jsonl

{"input": "Recognition of proper nouns in Japanese text has been studied as a part of the more general problem of morphological analysis in Japanese text processing -LRB- -LSB- 1 -RSB- -LSB- 2 -RSB- -RRB-. It has also been studied in the framework of Japanese information extraction -LRB- -LSB- 3 -RSB- -RRB- in recent years. Our approach to the Multi-lingual Evaluation Task -LRB- MET -RRB- for Japanese text is to consider the given task as a morphological analysis problem in Japanese. Our morphological analyzer has done all the necessary work for the recognition and classification of proper names, numerical and temporal expressions, i.e. Named Entity -LRB- NE -RRB- items in the Japanese text. The analyzer is called ``Amorph''. Amorph recognizes NE items in two stages: dictionary lookup and rule application. First, it uses several kinds of dictionaries to segment and tag Japanese character strings. Second, based on the information resulting from the dictionary lookup stage, a set of rules is applied to the segmented strings in order to identify NE items. When a segment is found to be an NE item, this information is added to the segment and it is used to generate the final output.", "output": {"entities": {"task": [{"text": "Recognition of proper nouns", "start": 0, "end": 27}, {"text": "morphological analysis", "start": 103, "end": 125}, {"text": "Japanese text processing", "start": 129, "end": 153}, {"text": "Japanese information extraction", "start": 240, "end": 271}, {"text": "Multi-lingual Evaluation Task -LRB- MET -RRB-", "start": 335, "end": 380}, {"text": "morphological analysis problem", "start": 434, "end": 464}, {"text": "recognition and classification of proper names, numerical and temporal expressions", "start": 545, "end": 627}], "other_scientific_term": [{"text": "proper nouns", "start": 15, "end": 27}, {"text": "proper names, numerical and temporal expressions", "start": 579, "end": 627}, {"text": "Named Entity -LRB- NE -RRB- items", "start": 634, "end": 667}, {"text": "NE items", "start": 743, "end": 751}, {"text": "dictionaries", "start": 839, "end": 851}, {"text": "Japanese character strings", "start": 871, "end": 897}, {"text": "rules", "start": 985, "end": 990}, {"text": "NE items", "start": 1048, "end": 1056}, {"text": "NE item", "start": 743, "end": 750}], "material": [{"text": "Japanese text", "start": 31, "end": 44}, {"text": "Japanese text", "start": 129, "end": 142}, {"text": "Japanese", "start": 31, "end": 39}, {"text": "Japanese text", "start": 385, "end": 398}], "generic": [{"text": "It", "start": 195, "end": 197}, {"text": "approach", "start": 319, "end": 327}, {"text": "task", "start": 424, "end": 428}, {"text": "analyzer", "start": 496, "end": 504}, {"text": "it", "start": 6, "end": 8}], "method": [{"text": "morphological analyzer", "start": 482, "end": 504}, {"text": "``Amorph''", "start": 713, "end": 723}, {"text": "Amorph", "start": 715, "end": 721}, {"text": "dictionary lookup", "start": 767, "end": 784}, {"text": "rule application", "start": 789, "end": 805}, {"text": "dictionary lookup stage", "start": 951, "end": 974}]}, "relations": {"part_of": [{"head": {"text": "Recognition of proper nouns", "start": 0, "end": 27}, "tail": {"text": "morphological analysis", "start": 103, "end": 125}}, {"head": {"text": "proper nouns", "start": 15, "end": 27}, "tail": {"text": "Japanese text", "start": 31, "end": 44}}, {"head": {"text": "Named Entity -LRB- NE -RRB- items", "start": 634, "end": 667}, "tail": {"text": "Japanese text", "start": 385, "end": 398}}, {"head": {"text": "dictionary lookup", "start": 767, "end": 784}, "tail": {"text": "Amorph", "start": 715, "end": 721}}, {"head": {"text": "rule application", "start": 789, "end": 805}, "tail": {"text": "Amorph", "start": 715, "end": 721}}], "used_for": [{"head": {"text": "morphological analysis", "start": 103, "end": 125}, "tail": {"text": "Japanese text processing", "start": 129, "end": 153}}, {"head": {"text": "Japanese information extraction", "start": 240, "end": 271}, "tail": {"text": "It", "start": 195, "end": 197}}, {"head": {"text": "approach", "start": 319, "end": 327}, "tail": {"text": "Multi-lingual Evaluation Task -LRB- MET -RRB-", "start": 335, "end": 380}}, {"head": {"text": "Multi-lingual Evaluation Task -LRB- MET -RRB-", "start": 335, "end": 380}, "tail": {"text": "Japanese text", "start": 129, "end": 142}}, {"head": {"text": "morphological analysis problem", "start": 434, "end": 464}, "tail": {"text": "task", "start": 424, "end": 428}}, {"head": {"text": "Japanese", "start": 31, "end": 39}, "tail": {"text": "morphological analysis problem", "start": 434, "end": 464}}, {"head": {"text": "morphological analyzer", "start": 482, "end": 504}, "tail": {"text": "recognition and classification of proper names, numerical and temporal expressions", "start": 545, "end": 627}}, {"head": {"text": "Amorph", "start": 715, "end": 721}, "tail": {"text": "NE items", "start": 743, "end": 751}}, {"head": {"text": "dictionaries", "start": 839, "end": 851}, "tail": {"text": "it", "start": 6, "end": 8}}, {"head": {"text": "dictionaries", "start": 839, "end": 851}, "tail": {"text": "Japanese character strings", "start": 871, "end": 897}}, {"head": {"text": "dictionary lookup stage", "start": 951, "end": 974}, "tail": {"text": "rules", "start": 985, "end": 990}}, {"head": {"text": "rules", "start": 985, "end": 990}, "tail": {"text": "NE items", "start": 1048, "end": 1056}}], "hyponym_of": [{"head": {"text": "Named Entity -LRB- NE -RRB- items", "start": 634, "end": 667}, "tail": {"text": "proper names, numerical and temporal expressions", "start": 579, "end": 627}}], "conjunction": [{"head": {"text": "dictionary lookup", "start": 767, "end": 784}, "tail": {"text": "rule application", "start": 789, "end": 805}}]}}, "schema": []}
{"input": "We propose to incorporate a priori geometric constraints in a 3 -- D stereo reconstruction scheme to cope with the many cases where image information alone is not sufficient to accurately recover 3 -- D shape. Our approach is based on the iterative deformation of a 3 -- D surface mesh to minimize an objective function. We show that combining anisotropic meshing with a non-quadratic approach to regularization enables us to obtain satisfactory reconstruction results using triangulations with few vertices. Structural or numerical constraints can then be added locally to the reconstruction process through a constrained optimization scheme. They improve the reconstruction results and enforce their consistency with a priori knowledge about object shape. The strong description and modeling properties of differential features make them useful tools that can be efficiently used as constraints for 3 -- D reconstruction.", "output": {"entities": {"other_scientific_term": [{"text": "priori geometric constraints", "start": 28, "end": 56}, {"text": "image information", "start": 132, "end": 149}, {"text": "3 -- D shape", "start": 196, "end": 208}, {"text": "objective function", "start": 301, "end": 319}, {"text": "regularization", "start": 397, "end": 411}, {"text": "triangulations", "start": 475, "end": 489}, {"text": "vertices", "start": 499, "end": 507}, {"text": "Structural or numerical constraints", "start": 509, "end": 544}, {"text": "priori knowledge", "start": 721, "end": 737}, {"text": "object shape", "start": 744, "end": 756}, {"text": "modeling properties", "start": 785, "end": 804}, {"text": "differential features", "start": 808, "end": 829}], "method": [{"text": "3 -- D stereo reconstruction scheme", "start": 62, "end": 97}, {"text": "iterative deformation of a 3 -- D surface mesh", "start": 239, "end": 285}, {"text": "anisotropic meshing", "start": 344, "end": 363}, {"text": "non-quadratic approach", "start": 371, "end": 393}, {"text": "reconstruction process", "start": 578, "end": 600}, {"text": "constrained optimization scheme", "start": 611, "end": 642}], "generic": [{"text": "approach", "start": 214, "end": 222}, {"text": "They", "start": 644, "end": 648}, {"text": "them", "start": 835, "end": 839}], "task": [{"text": "reconstruction", "start": 76, "end": 90}, {"text": "reconstruction", "start": 446, "end": 460}, {"text": "3 -- D reconstruction", "start": 901, "end": 922}]}, "relations": {"part_of": [{"head": {"text": "priori geometric constraints", "start": 28, "end": 56}, "tail": {"text": "3 -- D stereo reconstruction scheme", "start": 62, "end": 97}}], "used_for": [{"head": {"text": "image information", "start": 132, "end": 149}, "tail": {"text": "3 -- D shape", "start": 196, "end": 208}}, {"head": {"text": "iterative deformation of a 3 -- D surface mesh", "start": 239, "end": 285}, "tail": {"text": "approach", "start": 214, "end": 222}}, {"head": {"text": "iterative deformation of a 3 -- D surface mesh", "start": 239, "end": 285}, "tail": {"text": "objective function", "start": 301, "end": 319}}, {"head": {"text": "anisotropic meshing", "start": 344, "end": 363}, "tail": {"text": "reconstruction", "start": 76, "end": 90}}, {"head": {"text": "non-quadratic approach", "start": 371, "end": 393}, "tail": {"text": "regularization", "start": 397, "end": 411}}, {"head": {"text": "non-quadratic approach", "start": 371, "end": 393}, "tail": {"text": "reconstruction", "start": 76, "end": 90}}, {"head": {"text": "triangulations", "start": 475, "end": 489}, "tail": {"text": "reconstruction", "start": 76, "end": 90}}, {"head": {"text": "Structural or numerical constraints", "start": 509, "end": 544}, "tail": {"text": "reconstruction process", "start": 578, "end": 600}}, {"head": {"text": "constrained optimization scheme", "start": 611, "end": 642}, "tail": {"text": "Structural or numerical constraints", "start": 509, "end": 544}}, {"head": {"text": "They", "start": 644, "end": 648}, "tail": {"text": "reconstruction", "start": 446, "end": 460}}, {"head": {"text": "them", "start": 835, "end": 839}, "tail": {"text": "3 -- D reconstruction", "start": 901, "end": 922}}], "conjunction": [{"head": {"text": "anisotropic meshing", "start": 344, "end": 363}, "tail": {"text": "non-quadratic approach", "start": 371, "end": 393}}], "feature_of": [{"head": {"text": "object shape", "start": 744, "end": 756}, "tail": {"text": "priori knowledge", "start": 721, "end": 737}}]}}, "schema": []}
{"input": "This work proposes a new research direction to address the lack of structures in traditional n-gram models. It is based on a weakly supervised dependency parser that can model speech syntax without relying on any annotated training corpus. Labeled data is replaced by a few hand-crafted rules that encode basic syntactic knowledge. Bayesian inference then samples the rules, disambiguating and combining them to create complex tree structures that maximize a discriminative model's posterior on a target unlabeled corpus. This posterior encodes sparse se-lectional preferences between a head word and its dependents. The model is evaluated on English and Czech newspaper texts, and is then validated on French broadcast news transcriptions.", "output": {"entities": {"other_scientific_term": [{"text": "lack of structures", "start": 59, "end": 77}, {"text": "speech syntax", "start": 176, "end": 189}, {"text": "Labeled data", "start": 240, "end": 252}, {"text": "hand-crafted rules", "start": 274, "end": 292}, {"text": "syntactic knowledge", "start": 311, "end": 330}, {"text": "rules", "start": 287, "end": 292}, {"text": "complex tree structures", "start": 419, "end": 442}, {"text": "discriminative model's posterior", "start": 459, "end": 491}, {"text": "sparse se-lectional preferences", "start": 545, "end": 576}], "task": [{"text": "lack of structures in traditional n-gram models", "start": 59, "end": 106}, {"text": "weakly supervised dependency parser", "start": 125, "end": 160}], "method": [{"text": "n-gram models", "start": 93, "end": 106}, {"text": "Bayesian inference", "start": 332, "end": 350}], "material": [{"text": "annotated training corpus", "start": 213, "end": 238}, {"text": "unlabeled corpus", "start": 504, "end": 520}, {"text": "English and Czech newspaper texts", "start": 643, "end": 676}, {"text": "French broadcast news transcriptions", "start": 703, "end": 739}], "generic": [{"text": "them", "start": 404, "end": 408}, {"text": "posterior", "start": 482, "end": 491}, {"text": "model", "start": 100, "end": 105}]}, "relations": {"used_for": [{"head": {"text": "weakly supervised dependency parser", "start": 125, "end": 160}, "tail": {"text": "speech syntax", "start": 176, "end": 189}}, {"head": {"text": "hand-crafted rules", "start": 274, "end": 292}, "tail": {"text": "syntactic knowledge", "start": 311, "end": 330}}, {"head": {"text": "Bayesian inference", "start": 332, "end": 350}, "tail": {"text": "rules", "start": 287, "end": 292}}, {"head": {"text": "them", "start": 404, "end": 408}, "tail": {"text": "complex tree structures", "start": 419, "end": 442}}, {"head": {"text": "complex tree structures", "start": 419, "end": 442}, "tail": {"text": "discriminative model's posterior", "start": 459, "end": 491}}, {"head": {"text": "unlabeled corpus", "start": 504, "end": 520}, "tail": {"text": "discriminative model's posterior", "start": 459, "end": 491}}, {"head": {"text": "posterior", "start": 482, "end": 491}, "tail": {"text": "sparse se-lectional preferences", "start": 545, "end": 576}}], "evaluate_for": [{"head": {"text": "English and Czech newspaper texts", "start": 643, "end": 676}, "tail": {"text": "model", "start": 100, "end": 105}}, {"head": {"text": "French broadcast news transcriptions", "start": 703, "end": 739}, "tail": {"text": "model", "start": 100, "end": 105}}]}}, "schema": []}
{"input": "Listen-Communicate-Show -LRB- LCS -RRB- is a new paradigm for human interaction with data sources. We integrate a spoken language understanding system with intelligent mobile agents that mediate between users and information sources. We have built and will demonstrate an application of this approach called LCS-Marine. Using LCS-Marine, tactical personnel can converse with their logistics system to place a supply or information request. The request is passed to a mobile, intelligent agent for execution at the appropriate database. Requestors can also instruct the system to notify them when the status of a request changes or when a request is complete. We have demonstrated this capability in several field exercises with the Marines and are currently developing applications of this technology in new domains.", "output": {"entities": {"task": [{"text": "Listen-Communicate-Show -LRB- LCS -RRB-", "start": 0, "end": 39}, {"text": "human interaction with data sources", "start": 62, "end": 97}, {"text": "LCS-Marine", "start": 308, "end": 318}, {"text": "LCS-Marine", "start": 326, "end": 336}], "method": [{"text": "spoken language understanding system", "start": 114, "end": 150}, {"text": "intelligent mobile agents", "start": 156, "end": 181}, {"text": "mobile, intelligent agent", "start": 467, "end": 492}], "material": [{"text": "information sources", "start": 213, "end": 232}, {"text": "new domains", "start": 804, "end": 815}], "generic": [{"text": "approach", "start": 292, "end": 300}, {"text": "system", "start": 144, "end": 150}]}, "relations": {"used_for": [{"head": {"text": "Listen-Communicate-Show -LRB- LCS -RRB-", "start": 0, "end": 39}, "tail": {"text": "human interaction with data sources", "start": 62, "end": 97}}, {"head": {"text": "approach", "start": 292, "end": 300}, "tail": {"text": "LCS-Marine", "start": 308, "end": 318}}], "part_of": [{"head": {"text": "intelligent mobile agents", "start": 156, "end": 181}, "tail": {"text": "spoken language understanding system", "start": 114, "end": 150}}]}}, "schema": []}
{"input": "A domain independent model is proposed for the automated interpretation of nominal compounds in English. This model is meant to account for productive rules of interpretation which are inferred from the morpho-syntactic and semantic characteristics of the nominal constituents. In particular, we make extensive use of Pustejovsky's principles concerning the predicative information associated with nominals. We argue that it is necessary to draw a line between generalizable semantic principles and domain-specific semantic information. We explain this distinction and we show how this model may be applied to the interpretation of compounds in real texts, provided that complementary semantic information are retrieved.", "output": {"entities": {"method": [{"text": "domain independent model", "start": 2, "end": 26}, {"text": "Pustejovsky's principles", "start": 318, "end": 342}], "task": [{"text": "automated interpretation of nominal compounds", "start": 47, "end": 92}, {"text": "interpretation of compounds", "start": 614, "end": 641}], "other_scientific_term": [{"text": "nominal compounds", "start": 75, "end": 92}, {"text": "productive rules of interpretation", "start": 140, "end": 174}, {"text": "morpho-syntactic and semantic characteristics", "start": 203, "end": 248}, {"text": "nominal constituents", "start": 256, "end": 276}, {"text": "predicative information", "start": 358, "end": 381}, {"text": "nominals", "start": 398, "end": 406}, {"text": "generalizable semantic principles", "start": 461, "end": 494}, {"text": "domain-specific semantic information", "start": 499, "end": 535}, {"text": "semantic information", "start": 515, "end": 535}], "material": [{"text": "English", "start": 96, "end": 103}], "generic": [{"text": "model", "start": 21, "end": 26}, {"text": "model", "start": 110, "end": 115}]}, "relations": {"used_for": [{"head": {"text": "domain independent model", "start": 2, "end": 26}, "tail": {"text": "automated interpretation of nominal compounds", "start": 47, "end": 92}}, {"head": {"text": "model", "start": 21, "end": 26}, "tail": {"text": "productive rules of interpretation", "start": 140, "end": 174}}, {"head": {"text": "morpho-syntactic and semantic characteristics", "start": 203, "end": 248}, "tail": {"text": "productive rules of interpretation", "start": 140, "end": 174}}, {"head": {"text": "model", "start": 110, "end": 115}, "tail": {"text": "interpretation of compounds", "start": 614, "end": 641}}], "feature_of": [{"head": {"text": "English", "start": 96, "end": 103}, "tail": {"text": "nominal compounds", "start": 75, "end": 92}}, {"head": {"text": "morpho-syntactic and semantic characteristics", "start": 203, "end": 248}, "tail": {"text": "nominal constituents", "start": 256, "end": 276}}, {"head": {"text": "nominals", "start": 398, "end": 406}, "tail": {"text": "predicative information", "start": 358, "end": 381}}], "compare": [{"head": {"text": "generalizable semantic principles", "start": 461, "end": 494}, "tail": {"text": "domain-specific semantic information", "start": 499, "end": 535}}]}}, "schema": []}
{"input": "We present a new method for detecting interest points using histogram information. Unlike existing interest point detectors, which measure pixel-wise differences in image intensity, our detectors incorporate histogram-based representations, and thus can find image regions that present a distinct distribution in the neighborhood. The proposed detectors are able to capture large-scale structures and distinctive textured patterns, and exhibit strong invariance to rotation, illumination variation, and blur. The experimental results show that the proposed histogram-based interest point detectors perform particularly well for the tasks of matching textured scenes under blur and illumination changes, in terms of repeatability and distinctiveness. An extension of our method to space-time interest point detection for action classification is also presented.", "output": {"entities": {"generic": [{"text": "method", "start": 17, "end": 23}, {"text": "detectors", "start": 114, "end": 123}, {"text": "detectors", "start": 186, "end": 195}, {"text": "method", "start": 770, "end": 776}], "task": [{"text": "detecting interest points", "start": 28, "end": 53}, {"text": "matching textured scenes", "start": 641, "end": 665}, {"text": "space-time interest point detection", "start": 780, "end": 815}, {"text": "action classification", "start": 820, "end": 841}], "other_scientific_term": [{"text": "histogram information", "start": 60, "end": 81}, {"text": "large-scale structures", "start": 374, "end": 396}, {"text": "distinctive textured patterns", "start": 401, "end": 430}, {"text": "rotation", "start": 465, "end": 473}, {"text": "illumination variation", "start": 475, "end": 497}, {"text": "blur", "start": 503, "end": 507}, {"text": "blur and illumination changes", "start": 672, "end": 701}], "method": [{"text": "interest point detectors", "start": 99, "end": 123}, {"text": "histogram-based representations", "start": 208, "end": 239}, {"text": "histogram-based interest point detectors", "start": 557, "end": 597}], "metric": [{"text": "pixel-wise differences in image intensity", "start": 139, "end": 180}, {"text": "repeatability", "start": 715, "end": 728}, {"text": "distinctiveness", "start": 733, "end": 748}]}, "relations": {"used_for": [{"head": {"text": "method", "start": 17, "end": 23}, "tail": {"text": "detecting interest points", "start": 28, "end": 53}}, {"head": {"text": "histogram information", "start": 60, "end": 81}, "tail": {"text": "detecting interest points", "start": 28, "end": 53}}, {"head": {"text": "detectors", "start": 186, "end": 195}, "tail": {"text": "large-scale structures", "start": 374, "end": 396}}, {"head": {"text": "detectors", "start": 186, "end": 195}, "tail": {"text": "distinctive textured patterns", "start": 401, "end": 430}}, {"head": {"text": "detectors", "start": 186, "end": 195}, "tail": {"text": "rotation", "start": 465, "end": 473}}, {"head": {"text": "detectors", "start": 186, "end": 195}, "tail": {"text": "illumination variation", "start": 475, "end": 497}}, {"head": {"text": "detectors", "start": 186, "end": 195}, "tail": {"text": "blur", "start": 503, "end": 507}}, {"head": {"text": "histogram-based interest point detectors", "start": 557, "end": 597}, "tail": {"text": "matching textured scenes", "start": 641, "end": 665}}, {"head": {"text": "method", "start": 770, "end": 776}, "tail": {"text": "space-time interest point detection", "start": 780, "end": 815}}, {"head": {"text": "space-time interest point detection", "start": 780, "end": 815}, "tail": {"text": "action classification", "start": 820, "end": 841}}], "evaluate_for": [{"head": {"text": "pixel-wise differences in image intensity", "start": 139, "end": 180}, "tail": {"text": "interest point detectors", "start": 99, "end": 123}}, {"head": {"text": "repeatability", "start": 715, "end": 728}, "tail": {"text": "histogram-based interest point detectors", "start": 557, "end": 597}}, {"head": {"text": "distinctiveness", "start": 733, "end": 748}, "tail": {"text": "histogram-based interest point detectors", "start": 557, "end": 597}}], "part_of": [{"head": {"text": "histogram-based representations", "start": 208, "end": 239}, "tail": {"text": "detectors", "start": 114, "end": 123}}], "conjunction": [{"head": {"text": "large-scale structures", "start": 374, "end": 396}, "tail": {"text": "distinctive textured patterns", "start": 401, "end": 430}}, {"head": {"text": "rotation", "start": 465, "end": 473}, "tail": {"text": "illumination variation", "start": 475, "end": 497}}, {"head": {"text": "illumination variation", "start": 475, "end": 497}, "tail": {"text": "blur", "start": 503, "end": 507}}, {"head": {"text": "repeatability", "start": 715, "end": 728}, "tail": {"text": "distinctiveness", "start": 733, "end": 748}}]}}, "schema": []}
{"input": "We have implemented a restricted domain parser called Plume. Building on previous work at Carnegie-Mellon University e.g. -LSB- 4, 5, 8 -RSB-, Plume's approach to parsing is based on semantic caseframe instantiation. This has the advantages of efficiency on grammatical input, and robustness in the face of ungrammatical input. While Plume is well adapted to simple declarative and imperative utterances, it handles passives, relative clauses and interrogatives in an ad hoc manner leading to patchy syntactic coverage. This paper outlines Plume as it currently exists and describes our detailed design for extending Plume to handle passives, relative clauses, and interrogatives in a general manner.", "output": {"entities": {"method": [{"text": "restricted domain parser", "start": 22, "end": 46}, {"text": "Plume", "start": 54, "end": 59}, {"text": "Plume's approach", "start": 143, "end": 159}, {"text": "Plume", "start": 143, "end": 148}, {"text": "Plume", "start": 334, "end": 339}, {"text": "Plume", "start": 540, "end": 545}], "task": [{"text": "parsing", "start": 163, "end": 170}], "other_scientific_term": [{"text": "semantic caseframe instantiation", "start": 183, "end": 215}, {"text": "grammatical input", "start": 258, "end": 275}, {"text": "ungrammatical input", "start": 307, "end": 326}, {"text": "passives", "start": 416, "end": 424}, {"text": "relative clauses", "start": 426, "end": 442}, {"text": "interrogatives", "start": 447, "end": 461}, {"text": "patchy syntactic coverage", "start": 493, "end": 518}, {"text": "passives", "start": 633, "end": 641}, {"text": "relative clauses", "start": 643, "end": 659}, {"text": "interrogatives", "start": 665, "end": 679}], "metric": [{"text": "robustness", "start": 281, "end": 291}], "material": [{"text": "declarative and imperative utterances", "start": 366, "end": 403}], "generic": [{"text": "it", "start": 113, "end": 115}, {"text": "it", "start": 405, "end": 407}]}, "relations": {"hyponym_of": [{"head": {"text": "Plume", "start": 54, "end": 59}, "tail": {"text": "restricted domain parser", "start": 22, "end": 46}}], "used_for": [{"head": {"text": "Plume's approach", "start": 143, "end": 159}, "tail": {"text": "parsing", "start": 163, "end": 170}}, {"head": {"text": "semantic caseframe instantiation", "start": 183, "end": 215}, "tail": {"text": "Plume's approach", "start": 143, "end": 159}}, {"head": {"text": "Plume", "start": 143, "end": 148}, "tail": {"text": "declarative and imperative utterances", "start": 366, "end": 403}}, {"head": {"text": "it", "start": 113, "end": 115}, "tail": {"text": "passives", "start": 416, "end": 424}}, {"head": {"text": "it", "start": 113, "end": 115}, "tail": {"text": "relative clauses", "start": 426, "end": 442}}, {"head": {"text": "it", "start": 113, "end": 115}, "tail": {"text": "interrogatives", "start": 447, "end": 461}}, {"head": {"text": "Plume", "start": 540, "end": 545}, "tail": {"text": "passives", "start": 633, "end": 641}}, {"head": {"text": "Plume", "start": 540, "end": 545}, "tail": {"text": "relative clauses", "start": 643, "end": 659}}, {"head": {"text": "Plume", "start": 540, "end": 545}, "tail": {"text": "interrogatives", "start": 665, "end": 679}}], "feature_of": [{"head": {"text": "ungrammatical input", "start": 307, "end": 326}, "tail": {"text": "robustness", "start": 281, "end": 291}}], "conjunction": [{"head": {"text": "passives", "start": 416, "end": 424}, "tail": {"text": "relative clauses", "start": 426, "end": 442}}, {"head": {"text": "relative clauses", "start": 426, "end": 442}, "tail": {"text": "interrogatives", "start": 447, "end": 461}}, {"head": {"text": "passives", "start": 633, "end": 641}, "tail": {"text": "relative clauses", "start": 643, "end": 659}}, {"head": {"text": "relative clauses", "start": 643, "end": 659}, "tail": {"text": "interrogatives", "start": 665, "end": 679}}]}}, "schema": []}
{"input": "In this paper, we present an unlexicalized parser for German which employs smoothing and suffix analysis to achieve a labelled bracket F-score of 76.2, higher than previously reported results on the NEGRA corpus. In addition to the high accuracy of the model, the use of smoothing in an unlexicalized parser allows us to better examine the interplay between smoothing and parsing results.", "output": {"entities": {"method": [{"text": "unlexicalized parser", "start": 29, "end": 49}, {"text": "smoothing", "start": 75, "end": 84}, {"text": "suffix analysis", "start": 89, "end": 104}, {"text": "smoothing", "start": 271, "end": 280}, {"text": "unlexicalized parser", "start": 287, "end": 307}, {"text": "smoothing", "start": 358, "end": 367}, {"text": "parsing", "start": 372, "end": 379}], "material": [{"text": "German", "start": 54, "end": 60}, {"text": "NEGRA corpus", "start": 199, "end": 211}], "metric": [{"text": "labelled bracket F-score", "start": 118, "end": 142}, {"text": "accuracy", "start": 237, "end": 245}], "generic": [{"text": "model", "start": 253, "end": 258}]}, "relations": {"used_for": [{"head": {"text": "unlexicalized parser", "start": 29, "end": 49}, "tail": {"text": "German", "start": 54, "end": 60}}, {"head": {"text": "smoothing", "start": 75, "end": 84}, "tail": {"text": "unlexicalized parser", "start": 29, "end": 49}}, {"head": {"text": "suffix analysis", "start": 89, "end": 104}, "tail": {"text": "unlexicalized parser", "start": 29, "end": 49}}, {"head": {"text": "smoothing", "start": 271, "end": 280}, "tail": {"text": "unlexicalized parser", "start": 287, "end": 307}}], "conjunction": [{"head": {"text": "smoothing", "start": 75, "end": 84}, "tail": {"text": "suffix analysis", "start": 89, "end": 104}}], "evaluate_for": [{"head": {"text": "labelled bracket F-score", "start": 118, "end": 142}, "tail": {"text": "unlexicalized parser", "start": 29, "end": 49}}, {"head": {"text": "NEGRA corpus", "start": 199, "end": 211}, "tail": {"text": "unlexicalized parser", "start": 29, "end": 49}}, {"head": {"text": "accuracy", "start": 237, "end": 245}, "tail": {"text": "model", "start": 253, "end": 258}}]}}, "schema": []}
{"input": "This paper presents an unsupervised learning approach to disambiguate various relations between named entities by use of various lexical and syntactic features from the contexts. It works by calculating eigenvectors of an adjacency graph's Laplacian to recover a submanifold of data from a high dimensionality space and then performing cluster number estimation on the eigenvectors. Experiment results on ACE corpora show that this spectral clustering based approach outperforms the other clustering methods.", "output": {"entities": {"method": [{"text": "unsupervised learning approach", "start": 23, "end": 53}, {"text": "spectral clustering based approach", "start": 432, "end": 466}, {"text": "clustering methods", "start": 489, "end": 507}], "other_scientific_term": [{"text": "relations between named entities", "start": 78, "end": 110}, {"text": "lexical and syntactic features", "start": 129, "end": 159}, {"text": "eigenvectors", "start": 203, "end": 215}, {"text": "adjacency graph's Laplacian", "start": 222, "end": 249}, {"text": "submanifold", "start": 263, "end": 274}, {"text": "high dimensionality space", "start": 290, "end": 315}, {"text": "eigenvectors", "start": 369, "end": 381}], "generic": [{"text": "It", "start": 179, "end": 181}], "task": [{"text": "cluster number estimation", "start": 336, "end": 361}], "material": [{"text": "ACE corpora", "start": 405, "end": 416}]}, "relations": {"used_for": [{"head": {"text": "unsupervised learning approach", "start": 23, "end": 53}, "tail": {"text": "relations between named entities", "start": 78, "end": 110}}, {"head": {"text": "lexical and syntactic features", "start": 129, "end": 159}, "tail": {"text": "unsupervised learning approach", "start": 23, "end": 53}}, {"head": {"text": "It", "start": 179, "end": 181}, "tail": {"text": "submanifold", "start": 263, "end": 274}}, {"head": {"text": "eigenvectors", "start": 203, "end": 215}, "tail": {"text": "It", "start": 179, "end": 181}}, {"head": {"text": "high dimensionality space", "start": 290, "end": 315}, "tail": {"text": "submanifold", "start": 263, "end": 274}}, {"head": {"text": "cluster number estimation", "start": 336, "end": 361}, "tail": {"text": "It", "start": 179, "end": 181}}, {"head": {"text": "cluster number estimation", "start": 336, "end": 361}, "tail": {"text": "eigenvectors", "start": 369, "end": 381}}], "feature_of": [{"head": {"text": "adjacency graph's Laplacian", "start": 222, "end": 249}, "tail": {"text": "eigenvectors", "start": 203, "end": 215}}], "evaluate_for": [{"head": {"text": "ACE corpora", "start": 405, "end": 416}, "tail": {"text": "spectral clustering based approach", "start": 432, "end": 466}}, {"head": {"text": "ACE corpora", "start": 405, "end": 416}, "tail": {"text": "clustering methods", "start": 489, "end": 507}}], "compare": [{"head": {"text": "spectral clustering based approach", "start": 432, "end": 466}, "tail": {"text": "clustering methods", "start": 489, "end": 507}}]}}, "schema": []}
{"input": "This paper proposes a generic mathematical formalism for the combination of various structures: strings, trees, dags, graphs, and products of them. The polarization of the objects of the elementary structures controls the saturation of the final structure. This formalism is both elementary and powerful enough to strongly simulate many grammar formalisms, such as rewriting systems, dependency grammars, TAG, HPSG and LFG.", "output": {"entities": {"method": [{"text": "mathematical formalism", "start": 30, "end": 52}, {"text": "grammar formalisms", "start": 337, "end": 355}, {"text": "rewriting systems", "start": 365, "end": 382}, {"text": "dependency grammars", "start": 384, "end": 403}], "generic": [{"text": "structures", "start": 84, "end": 94}, {"text": "them", "start": 32, "end": 36}, {"text": "formalism", "start": 43, "end": 52}], "other_scientific_term": [{"text": "strings", "start": 96, "end": 103}, {"text": "trees", "start": 105, "end": 110}, {"text": "dags", "start": 112, "end": 116}, {"text": "graphs", "start": 118, "end": 124}, {"text": "polarization", "start": 152, "end": 164}, {"text": "elementary structures", "start": 187, "end": 208}, {"text": "TAG", "start": 405, "end": 408}, {"text": "HPSG", "start": 410, "end": 414}, {"text": "LFG", "start": 419, "end": 422}]}, "relations": {"hyponym_of": [{"head": {"text": "strings", "start": 96, "end": 103}, "tail": {"text": "structures", "start": 84, "end": 94}}, {"head": {"text": "trees", "start": 105, "end": 110}, "tail": {"text": "structures", "start": 84, "end": 94}}, {"head": {"text": "dags", "start": 112, "end": 116}, "tail": {"text": "structures", "start": 84, "end": 94}}, {"head": {"text": "graphs", "start": 118, "end": 124}, "tail": {"text": "structures", "start": 84, "end": 94}}, {"head": {"text": "rewriting systems", "start": 365, "end": 382}, "tail": {"text": "grammar formalisms", "start": 337, "end": 355}}, {"head": {"text": "dependency grammars", "start": 384, "end": 403}, "tail": {"text": "grammar formalisms", "start": 337, "end": 355}}, {"head": {"text": "TAG", "start": 405, "end": 408}, "tail": {"text": "grammar formalisms", "start": 337, "end": 355}}, {"head": {"text": "HPSG", "start": 410, "end": 414}, "tail": {"text": "grammar formalisms", "start": 337, "end": 355}}, {"head": {"text": "LFG", "start": 419, "end": 422}, "tail": {"text": "grammar formalisms", "start": 337, "end": 355}}], "conjunction": [{"head": {"text": "strings", "start": 96, "end": 103}, "tail": {"text": "trees", "start": 105, "end": 110}}, {"head": {"text": "trees", "start": 105, "end": 110}, "tail": {"text": "dags", "start": 112, "end": 116}}, {"head": {"text": "dags", "start": 112, "end": 116}, "tail": {"text": "graphs", "start": 118, "end": 124}}, {"head": {"text": "rewriting systems", "start": 365, "end": 382}, "tail": {"text": "dependency grammars", "start": 384, "end": 403}}, {"head": {"text": "dependency grammars", "start": 384, "end": 403}, "tail": {"text": "TAG", "start": 405, "end": 408}}, {"head": {"text": "TAG", "start": 405, "end": 408}, "tail": {"text": "HPSG", "start": 410, "end": 414}}, {"head": {"text": "HPSG", "start": 410, "end": 414}, "tail": {"text": "LFG", "start": 419, "end": 422}}], "used_for": [{"head": {"text": "formalism", "start": 43, "end": 52}, "tail": {"text": "grammar formalisms", "start": 337, "end": 355}}]}}, "schema": []}
{"input": "A mixed-signal paradigm is presented for high-resolution parallel inner-product computation in very high dimensions, suitable for efficient implementation of kernels in image processing. At the core of the externally digital architecture is a high-density, low-power analog array performing binary-binary partial matrix-vector multiplication. Full digital resolution is maintained even with low-resolution analog-to-digital conversion, owing to random statistics in the analog summation of binary products. A random modulation scheme produces near-Bernoulli statistics even for highly correlated inputs. The approach is validated with real image data, and with experimental results from a CID/DRAM analog array prototype in 0.5 cents m CMOS.", "output": {"entities": {"method": [{"text": "mixed-signal paradigm", "start": 2, "end": 23}, {"text": "kernels", "start": 158, "end": 165}, {"text": "externally digital architecture", "start": 206, "end": 237}, {"text": "binary-binary partial matrix-vector multiplication", "start": 291, "end": 341}, {"text": "low-resolution analog-to-digital conversion", "start": 391, "end": 434}, {"text": "random modulation scheme", "start": 509, "end": 533}], "task": [{"text": "high-resolution parallel inner-product computation", "start": 41, "end": 91}, {"text": "image processing", "start": 169, "end": 185}], "other_scientific_term": [{"text": "high-density, low-power analog array", "start": 243, "end": 279}, {"text": "random statistics", "start": 445, "end": 462}, {"text": "analog summation of binary products", "start": 470, "end": 505}, {"text": "near-Bernoulli statistics", "start": 543, "end": 568}, {"text": "CID/DRAM analog array prototype", "start": 689, "end": 720}], "metric": [{"text": "Full digital resolution", "start": 343, "end": 366}], "material": [{"text": "highly correlated inputs", "start": 578, "end": 602}, {"text": "real image data", "start": 635, "end": 650}], "generic": [{"text": "approach", "start": 608, "end": 616}]}, "relations": {"used_for": [{"head": {"text": "mixed-signal paradigm", "start": 2, "end": 23}, "tail": {"text": "high-resolution parallel inner-product computation", "start": 41, "end": 91}}, {"head": {"text": "mixed-signal paradigm", "start": 2, "end": 23}, "tail": {"text": "kernels", "start": 158, "end": 165}}, {"head": {"text": "kernels", "start": 158, "end": 165}, "tail": {"text": "image processing", "start": 169, "end": 185}}, {"head": {"text": "binary-binary partial matrix-vector multiplication", "start": 291, "end": 341}, "tail": {"text": "high-density, low-power analog array", "start": 243, "end": 279}}, {"head": {"text": "random modulation scheme", "start": 509, "end": 533}, "tail": {"text": "near-Bernoulli statistics", "start": 543, "end": 568}}, {"head": {"text": "highly correlated inputs", "start": 578, "end": 602}, "tail": {"text": "random modulation scheme", "start": 509, "end": 533}}], "part_of": [{"head": {"text": "high-density, low-power analog array", "start": 243, "end": 279}, "tail": {"text": "externally digital architecture", "start": 206, "end": 237}}, {"head": {"text": "random statistics", "start": 445, "end": 462}, "tail": {"text": "analog summation of binary products", "start": 470, "end": 505}}], "evaluate_for": [{"head": {"text": "real image data", "start": 635, "end": 650}, "tail": {"text": "approach", "start": 608, "end": 616}}]}}, "schema": []}
{"input": "In this paper we specialize the projective unifocal, bifo-cal, and trifocal tensors to the affine case, and show how the tensors obtained relate to the registered tensors encountered in previous work. This enables us to obtain an affine specialization of known projective relations connecting points and lines across two or three views. In the simpler case of affine cameras we give neccessary and sufficient constraints on the components of the trifocal tensor, together with a simple geometric interpretation. Finally, we show how the estimation of the tensors from point correspondences is achieved through factorization, and discuss the estimation from line correspondences.", "output": {"entities": {"method": [{"text": "projective unifocal, bifo-cal, and trifocal tensors", "start": 32, "end": 83}, {"text": "affine cameras", "start": 360, "end": 374}, {"text": "trifocal tensor", "start": 67, "end": 82}, {"text": "factorization", "start": 610, "end": 623}], "other_scientific_term": [{"text": "affine case", "start": 91, "end": 102}, {"text": "registered tensors", "start": 152, "end": 170}, {"text": "affine specialization", "start": 230, "end": 251}, {"text": "geometric interpretation", "start": 486, "end": 510}, {"text": "point correspondences", "start": 568, "end": 589}, {"text": "line correspondences", "start": 657, "end": 677}], "generic": [{"text": "tensors", "start": 76, "end": 83}, {"text": "tensors", "start": 121, "end": 128}, {"text": "estimation", "start": 537, "end": 547}], "task": [{"text": "estimation of the tensors", "start": 537, "end": 562}]}, "relations": {"used_for": [{"head": {"text": "projective unifocal, bifo-cal, and trifocal tensors", "start": 32, "end": 83}, "tail": {"text": "affine case", "start": 91, "end": 102}}, {"head": {"text": "point correspondences", "start": 568, "end": 589}, "tail": {"text": "estimation of the tensors", "start": 537, "end": 562}}, {"head": {"text": "factorization", "start": 610, "end": 623}, "tail": {"text": "tensors", "start": 121, "end": 128}}, {"head": {"text": "line correspondences", "start": 657, "end": 677}, "tail": {"text": "estimation", "start": 537, "end": 547}}]}}, "schema": []}
{"input": "This paper presents an algorithm for selecting an appropriate classifier word for a noun. In Thai language, it frequently happens that there is fluctuation in the choice of classifier for a given concrete noun, both from the point of view of the whole speech community and individual speakers. Basically, there is no exact rule for classifier selection. As far as we can do in the rule-based approach is to give a default rule to pick up a corresponding classifier of each noun. Registration of classifier for each noun is limited to the type of unit classifier because other types are open due to the meaning of representation. We propose a corpus-based method -LRB- Biber ,1993; Nagao ,1993; Smadja ,1993 -RRB- which generates Noun Classifier Associations -LRB- NCA -RRB- to overcome the problems in classifier assignment and semantic construction of noun phrase. The NCA is created statistically from a large corpus and recomposed under concept hierarchy constraints and frequency of occurrences.", "output": {"entities": {"material": [{"text": "Thai language", "start": 93, "end": 106}], "method": [{"text": "classifier", "start": 62, "end": 72}, {"text": "rule-based approach", "start": 381, "end": 400}, {"text": "classifier", "start": 173, "end": 183}, {"text": "classifier", "start": 332, "end": 342}, {"text": "corpus-based method", "start": 642, "end": 661}, {"text": "Noun Classifier Associations -LRB- NCA -RRB-", "start": 729, "end": 773}, {"text": "NCA", "start": 764, "end": 767}], "task": [{"text": "classifier selection", "start": 332, "end": 352}, {"text": "classifier assignment", "start": 802, "end": 823}, {"text": "semantic construction of noun phrase", "start": 828, "end": 864}], "other_scientific_term": [{"text": "type of unit classifier", "start": 538, "end": 561}, {"text": "concept hierarchy constraints", "start": 940, "end": 969}, {"text": "frequency of occurrences", "start": 974, "end": 998}]}, "relations": {"used_for": [{"head": {"text": "corpus-based method", "start": 642, "end": 661}, "tail": {"text": "Noun Classifier Associations -LRB- NCA -RRB-", "start": 729, "end": 773}}, {"head": {"text": "corpus-based method", "start": 642, "end": 661}, "tail": {"text": "classifier assignment", "start": 802, "end": 823}}, {"head": {"text": "corpus-based method", "start": 642, "end": 661}, "tail": {"text": "semantic construction of noun phrase", "start": 828, "end": 864}}, {"head": {"text": "Noun Classifier Associations -LRB- NCA -RRB-", "start": 729, "end": 773}, "tail": {"text": "classifier assignment", "start": 802, "end": 823}}, {"head": {"text": "Noun Classifier Associations -LRB- NCA -RRB-", "start": 729, "end": 773}, "tail": {"text": "semantic construction of noun phrase", "start": 828, "end": 864}}, {"head": {"text": "concept hierarchy constraints", "start": 940, "end": 969}, "tail": {"text": "NCA", "start": 764, "end": 767}}, {"head": {"text": "frequency of occurrences", "start": 974, "end": 998}, "tail": {"text": "NCA", "start": 764, "end": 767}}], "conjunction": [{"head": {"text": "classifier assignment", "start": 802, "end": 823}, "tail": {"text": "semantic construction of noun phrase", "start": 828, "end": 864}}]}}, "schema": []}
{"input": "The perception of transparent objects from images is known to be a very hard problem in vision. Given a single image, it is difficult to even detect the presence of transparent objects in the scene. In this paper, we explore what can be said about transparent objects by a moving observer. We show how features that are imaged through a transparent object behave differently from those that are rigidly attached to the scene. We present a novel model-based approach to recover the shapes and the poses of transparent objects from known motion. The objects can be complex in that they may be composed of multiple layers with different refractive indices. We have conducted numerous simulations to verify the practical feasibility of our algorithm. We have applied it to real scenes that include transparent objects and recovered the shapes of the objects with high accuracy.", "output": {"entities": {"task": [{"text": "perception of transparent objects", "start": 4, "end": 37}], "material": [{"text": "images", "start": 43, "end": 49}, {"text": "real scenes", "start": 769, "end": 780}], "other_scientific_term": [{"text": "transparent objects", "start": 18, "end": 37}, {"text": "transparent objects", "start": 165, "end": 184}, {"text": "features", "start": 302, "end": 310}, {"text": "transparent object", "start": 18, "end": 36}, {"text": "shapes and the poses of transparent objects", "start": 481, "end": 524}, {"text": "known motion", "start": 530, "end": 542}, {"text": "multiple layers", "start": 603, "end": 618}, {"text": "refractive indices", "start": 634, "end": 652}, {"text": "transparent objects", "start": 248, "end": 267}, {"text": "shapes of the objects", "start": 832, "end": 853}], "generic": [{"text": "those", "start": 380, "end": 385}, {"text": "objects", "start": 30, "end": 37}, {"text": "they", "start": 579, "end": 583}, {"text": "algorithm", "start": 736, "end": 745}, {"text": "it", "start": 118, "end": 120}], "method": [{"text": "model-based approach", "start": 445, "end": 465}], "metric": [{"text": "accuracy", "start": 864, "end": 872}]}, "relations": {"used_for": [{"head": {"text": "images", "start": 43, "end": 49}, "tail": {"text": "perception of transparent objects", "start": 4, "end": 37}}, {"head": {"text": "model-based approach", "start": 445, "end": 465}, "tail": {"text": "shapes and the poses of transparent objects", "start": 481, "end": 524}}, {"head": {"text": "known motion", "start": 530, "end": 542}, "tail": {"text": "shapes and the poses of transparent objects", "start": 481, "end": 524}}, {"head": {"text": "it", "start": 118, "end": 120}, "tail": {"text": "real scenes", "start": 769, "end": 780}}, {"head": {"text": "it", "start": 118, "end": 120}, "tail": {"text": "shapes of the objects", "start": 832, "end": 853}}], "compare": [{"head": {"text": "those", "start": 380, "end": 385}, "tail": {"text": "features", "start": 302, "end": 310}}], "part_of": [{"head": {"text": "multiple layers", "start": 603, "end": 618}, "tail": {"text": "they", "start": 579, "end": 583}}, {"head": {"text": "transparent objects", "start": 248, "end": 267}, "tail": {"text": "real scenes", "start": 769, "end": 780}}], "feature_of": [{"head": {"text": "refractive indices", "start": 634, "end": 652}, "tail": {"text": "multiple layers", "start": 603, "end": 618}}], "evaluate_for": [{"head": {"text": "accuracy", "start": 864, "end": 872}, "tail": {"text": "shapes of the objects", "start": 832, "end": 853}}]}}, "schema": []}
{"input": "We propose a novel probabilistic framework for learning visual models of 3D object categories by combining appearance information and geometric constraints. Objects are represented as a coherent ensemble of parts that are consistent under 3D viewpoint transformations. Each part is a collection of salient image features. A generative framework is used for learning a model that captures the relative position of parts within each of the discretized viewpoints. Contrary to most of the existing mixture of viewpoints models, our model establishes explicit correspondences of parts across different viewpoints of the object class. Given a new image, detection and classification are achieved by determining the position and viewpoint of the model that maximize recognition scores of the candidate objects. Our approach is among the first to propose a generative proba-bilistic framework for 3D object categorization. We test our algorithm on the detection task and the viewpoint classification task by using'' car'' category from both the Savarese et al. 2007 and PASCAL VOC 2006 datasets. We show promising results in both the detection and viewpoint classification tasks on these two challenging datasets.", "output": {"entities": {"method": [{"text": "probabilistic framework", "start": 19, "end": 42}, {"text": "generative framework", "start": 324, "end": 344}, {"text": "mixture of viewpoints models", "start": 495, "end": 523}, {"text": "generative proba-bilistic framework", "start": 850, "end": 885}], "task": [{"text": "visual models of 3D object categories", "start": 56, "end": 93}, {"text": "detection", "start": 649, "end": 658}, {"text": "classification", "start": 663, "end": 677}, {"text": "3D object categorization", "start": 890, "end": 914}, {"text": "detection task", "start": 945, "end": 959}, {"text": "viewpoint classification task", "start": 968, "end": 997}, {"text": "detection and viewpoint classification tasks", "start": 1127, "end": 1171}], "other_scientific_term": [{"text": "appearance information", "start": 107, "end": 129}, {"text": "geometric constraints", "start": 134, "end": 155}, {"text": "3D viewpoint transformations", "start": 239, "end": 267}, {"text": "salient image features", "start": 298, "end": 320}, {"text": "discretized viewpoints", "start": 438, "end": 460}, {"text": "viewpoints", "start": 450, "end": 460}, {"text": "position", "start": 401, "end": 409}, {"text": "viewpoint", "start": 242, "end": 251}], "generic": [{"text": "model", "start": 63, "end": 68}, {"text": "model", "start": 368, "end": 373}, {"text": "approach", "start": 809, "end": 817}, {"text": "algorithm", "start": 928, "end": 937}, {"text": "datasets", "start": 1079, "end": 1087}], "material": [{"text": "image", "start": 306, "end": 311}, {"text": "PASCAL VOC 2006 datasets", "start": 1063, "end": 1087}], "metric": [{"text": "recognition scores", "start": 760, "end": 778}]}, "relations": {"used_for": [{"head": {"text": "probabilistic framework", "start": 19, "end": 42}, "tail": {"text": "visual models of 3D object categories", "start": 56, "end": 93}}, {"head": {"text": "appearance information", "start": 107, "end": 129}, "tail": {"text": "probabilistic framework", "start": 19, "end": 42}}, {"head": {"text": "geometric constraints", "start": 134, "end": 155}, "tail": {"text": "probabilistic framework", "start": 19, "end": 42}}, {"head": {"text": "generative framework", "start": 324, "end": 344}, "tail": {"text": "model", "start": 63, "end": 68}}, {"head": {"text": "image", "start": 306, "end": 311}, "tail": {"text": "detection", "start": 649, "end": 658}}, {"head": {"text": "image", "start": 306, "end": 311}, "tail": {"text": "classification", "start": 663, "end": 677}}, {"head": {"text": "position", "start": 401, "end": 409}, "tail": {"text": "detection", "start": 649, "end": 658}}, {"head": {"text": "position", "start": 401, "end": 409}, "tail": {"text": "classification", "start": 663, "end": 677}}, {"head": {"text": "viewpoint", "start": 242, "end": 251}, "tail": {"text": "detection", "start": 649, "end": 658}}, {"head": {"text": "viewpoint", "start": 242, "end": 251}, "tail": {"text": "classification", "start": 663, "end": 677}}, {"head": {"text": "generative proba-bilistic framework", "start": 850, "end": 885}, "tail": {"text": "3D object categorization", "start": 890, "end": 914}}, {"head": {"text": "algorithm", "start": 928, "end": 937}, "tail": {"text": "detection task", "start": 945, "end": 959}}, {"head": {"text": "algorithm", "start": 928, "end": 937}, "tail": {"text": "viewpoint classification task", "start": 968, "end": 997}}], "conjunction": [{"head": {"text": "appearance information", "start": 107, "end": 129}, "tail": {"text": "geometric constraints", "start": 134, "end": 155}}, {"head": {"text": "detection", "start": 649, "end": 658}, "tail": {"text": "classification", "start": 663, "end": 677}}, {"head": {"text": "position", "start": 401, "end": 409}, "tail": {"text": "viewpoint", "start": 242, "end": 251}}, {"head": {"text": "detection task", "start": 945, "end": 959}, "tail": {"text": "viewpoint classification task", "start": 968, "end": 997}}], "compare": [{"head": {"text": "model", "start": 368, "end": 373}, "tail": {"text": "mixture of viewpoints models", "start": 495, "end": 523}}], "evaluate_for": [{"head": {"text": "PASCAL VOC 2006 datasets", "start": 1063, "end": 1087}, "tail": {"text": "algorithm", "start": 928, "end": 937}}, {"head": {"text": "datasets", "start": 1079, "end": 1087}, "tail": {"text": "detection and viewpoint classification tasks", "start": 1127, "end": 1171}}]}}, "schema": []}
{"input": "We present an application of ambiguity packing and stochastic disambiguation techniques for Lexical-Functional Grammars -LRB- LFG -RRB- to the domain of sentence condensation. Our system incorporates a linguistic parser/generator for LFG, a transfer component for parse reduction operating on packed parse forests, and a maximum-entropy model for stochastic output selection. Furthermore, we propose the use of standard parser evaluation methods for automatically evaluating the summarization quality of sentence condensation systems. An experimental evaluation of summarization quality shows a close correlation between the automatic parse-based evaluation and a manual evaluation of generated strings. Overall summarization quality of the proposed system is state-of-the-art, with guaranteed grammaticality of the system output due to the use of a constraint-based parser/generator.", "output": {"entities": {"method": [{"text": "ambiguity packing and stochastic disambiguation techniques", "start": 29, "end": 87}, {"text": "Lexical-Functional Grammars -LRB- LFG -RRB-", "start": 92, "end": 135}, {"text": "linguistic parser/generator", "start": 202, "end": 229}, {"text": "LFG", "start": 126, "end": 129}, {"text": "transfer component", "start": 241, "end": 259}, {"text": "maximum-entropy model", "start": 321, "end": 342}, {"text": "parser evaluation methods", "start": 420, "end": 445}, {"text": "sentence condensation systems", "start": 504, "end": 533}, {"text": "automatic parse-based evaluation", "start": 625, "end": 657}, {"text": "manual evaluation", "start": 664, "end": 681}, {"text": "constraint-based parser/generator", "start": 850, "end": 883}], "task": [{"text": "sentence condensation", "start": 153, "end": 174}, {"text": "parse reduction", "start": 264, "end": 279}, {"text": "stochastic output selection", "start": 347, "end": 374}], "generic": [{"text": "system", "start": 180, "end": 186}, {"text": "system", "start": 526, "end": 532}], "other_scientific_term": [{"text": "packed parse forests", "start": 293, "end": 313}], "metric": [{"text": "summarization quality", "start": 479, "end": 500}, {"text": "summarization quality", "start": 565, "end": 586}, {"text": "summarization quality", "start": 712, "end": 733}, {"text": "grammaticality", "start": 794, "end": 808}]}, "relations": {"used_for": [{"head": {"text": "ambiguity packing and stochastic disambiguation techniques", "start": 29, "end": 87}, "tail": {"text": "Lexical-Functional Grammars -LRB- LFG -RRB-", "start": 92, "end": 135}}, {"head": {"text": "ambiguity packing and stochastic disambiguation techniques", "start": 29, "end": 87}, "tail": {"text": "sentence condensation", "start": 153, "end": 174}}, {"head": {"text": "linguistic parser/generator", "start": 202, "end": 229}, "tail": {"text": "LFG", "start": 126, "end": 129}}, {"head": {"text": "transfer component", "start": 241, "end": 259}, "tail": {"text": "parse reduction", "start": 264, "end": 279}}, {"head": {"text": "packed parse forests", "start": 293, "end": 313}, "tail": {"text": "parse reduction", "start": 264, "end": 279}}, {"head": {"text": "maximum-entropy model", "start": 321, "end": 342}, "tail": {"text": "stochastic output selection", "start": 347, "end": 374}}, {"head": {"text": "constraint-based parser/generator", "start": 850, "end": 883}, "tail": {"text": "system", "start": 526, "end": 532}}], "part_of": [{"head": {"text": "linguistic parser/generator", "start": 202, "end": 229}, "tail": {"text": "system", "start": 180, "end": 186}}, {"head": {"text": "transfer component", "start": 241, "end": 259}, "tail": {"text": "system", "start": 180, "end": 186}}, {"head": {"text": "maximum-entropy model", "start": 321, "end": 342}, "tail": {"text": "system", "start": 180, "end": 186}}], "conjunction": [{"head": {"text": "linguistic parser/generator", "start": 202, "end": 229}, "tail": {"text": "transfer component", "start": 241, "end": 259}}, {"head": {"text": "transfer component", "start": 241, "end": 259}, "tail": {"text": "maximum-entropy model", "start": 321, "end": 342}}], "evaluate_for": [{"head": {"text": "parser evaluation methods", "start": 420, "end": 445}, "tail": {"text": "summarization quality", "start": 479, "end": 500}}, {"head": {"text": "summarization quality", "start": 479, "end": 500}, "tail": {"text": "sentence condensation systems", "start": 504, "end": 533}}, {"head": {"text": "summarization quality", "start": 565, "end": 586}, "tail": {"text": "automatic parse-based evaluation", "start": 625, "end": 657}}, {"head": {"text": "summarization quality", "start": 712, "end": 733}, "tail": {"text": "system", "start": 526, "end": 532}}, {"head": {"text": "grammaticality", "start": 794, "end": 808}, "tail": {"text": "system", "start": 526, "end": 532}}], "compare": [{"head": {"text": "automatic parse-based evaluation", "start": 625, "end": 657}, "tail": {"text": "manual evaluation", "start": 664, "end": 681}}]}}, "schema": []}
{"input": "The robust principal component analysis -LRB- robust PCA -RRB- problem has been considered in many machine learning applications, where the goal is to decompose the data matrix to a low rank part plus a sparse residual. While current approaches are developed by only considering the low rank plus sparse structure, in many applications, side information of row and/or column entities may also be given, and it is still unclear to what extent could such information help robust PCA. Thus, in this paper, we study the problem of robust PCA with side information, where both prior structure and features of entities are exploited for recovery. We propose a convex problem to incorporate side information in robust PCA and show that the low rank matrix can be exactly recovered via the proposed method under certain conditions. In particular, our guarantee suggests that a substantial amount of low rank matrices, which can not be recovered by standard robust PCA, become re-coverable by our proposed method. The result theoretically justifies the effectiveness of features in robust PCA. In addition, we conduct synthetic experiments as well as a real application on noisy image classification to show that our method also improves the performance in practice by exploiting side information.", "output": {"entities": {"method": [{"text": "robust principal component analysis -LRB- robust PCA -RRB- problem", "start": 4, "end": 70}, {"text": "robust PCA", "start": 46, "end": 56}, {"text": "robust PCA", "start": 470, "end": 480}, {"text": "robust PCA", "start": 527, "end": 537}, {"text": "robust PCA", "start": 704, "end": 714}, {"text": "robust PCA", "start": 949, "end": 959}], "task": [{"text": "machine learning applications", "start": 99, "end": 128}, {"text": "recovery", "start": 631, "end": 639}, {"text": "convex problem", "start": 654, "end": 668}, {"text": "noisy image classification", "start": 1164, "end": 1190}], "other_scientific_term": [{"text": "data matrix", "start": 165, "end": 176}, {"text": "low rank part", "start": 182, "end": 195}, {"text": "sparse residual", "start": 203, "end": 218}, {"text": "low rank plus sparse structure", "start": 283, "end": 313}, {"text": "side information", "start": 337, "end": 353}, {"text": "side information", "start": 543, "end": 559}, {"text": "prior structure", "start": 572, "end": 587}, {"text": "features of entities", "start": 592, "end": 612}, {"text": "side information", "start": 684, "end": 700}, {"text": "low rank matrix", "start": 733, "end": 748}, {"text": "low rank matrices", "start": 891, "end": 908}, {"text": "features", "start": 592, "end": 600}, {"text": "side information", "start": 1271, "end": 1287}], "generic": [{"text": "approaches", "start": 234, "end": 244}, {"text": "information", "start": 342, "end": 353}, {"text": "method", "start": 791, "end": 797}, {"text": "method", "start": 997, "end": 1003}, {"text": "method", "start": 1208, "end": 1214}]}, "relations": {"used_for": [{"head": {"text": "robust principal component analysis -LRB- robust PCA -RRB- problem", "start": 4, "end": 70}, "tail": {"text": "machine learning applications", "start": 99, "end": 128}}, {"head": {"text": "low rank plus sparse structure", "start": 283, "end": 313}, "tail": {"text": "approaches", "start": 234, "end": 244}}, {"head": {"text": "information", "start": 342, "end": 353}, "tail": {"text": "robust PCA", "start": 46, "end": 56}}, {"head": {"text": "side information", "start": 543, "end": 559}, "tail": {"text": "robust PCA", "start": 470, "end": 480}}, {"head": {"text": "prior structure", "start": 572, "end": 587}, "tail": {"text": "recovery", "start": 631, "end": 639}}, {"head": {"text": "features of entities", "start": 592, "end": 612}, "tail": {"text": "recovery", "start": 631, "end": 639}}, {"head": {"text": "convex problem", "start": 654, "end": 668}, "tail": {"text": "side information", "start": 684, "end": 700}}, {"head": {"text": "method", "start": 791, "end": 797}, "tail": {"text": "low rank matrix", "start": 733, "end": 748}}, {"head": {"text": "method", "start": 997, "end": 1003}, "tail": {"text": "low rank matrices", "start": 891, "end": 908}}, {"head": {"text": "side information", "start": 1271, "end": 1287}, "tail": {"text": "method", "start": 1208, "end": 1214}}], "part_of": [{"head": {"text": "low rank part", "start": 182, "end": 195}, "tail": {"text": "data matrix", "start": 165, "end": 176}}, {"head": {"text": "sparse residual", "start": 203, "end": 218}, "tail": {"text": "data matrix", "start": 165, "end": 176}}, {"head": {"text": "side information", "start": 684, "end": 700}, "tail": {"text": "robust PCA", "start": 527, "end": 537}}], "conjunction": [{"head": {"text": "low rank part", "start": 182, "end": 195}, "tail": {"text": "sparse residual", "start": 203, "end": 218}}, {"head": {"text": "prior structure", "start": 572, "end": 587}, "tail": {"text": "features of entities", "start": 592, "end": 612}}], "feature_of": [{"head": {"text": "features", "start": 592, "end": 600}, "tail": {"text": "robust PCA", "start": 949, "end": 959}}], "evaluate_for": [{"head": {"text": "noisy image classification", "start": 1164, "end": 1190}, "tail": {"text": "method", "start": 1208, "end": 1214}}]}}, "schema": []}
{"input": "This paper presents necessary and sufficient conditions for the use of demonstrative expressions in English and discusses implications for current discourse processing algorithms. We examine a broad range of texts to show how the distribution of demonstrative forms and functions is genre dependent. This research is part of a larger study of anaphoric expressions, the results of which will be incorporated into a natural language generation system.", "output": {"entities": {"other_scientific_term": [{"text": "demonstrative expressions", "start": 71, "end": 96}, {"text": "demonstrative forms and functions", "start": 246, "end": 279}, {"text": "anaphoric expressions", "start": 343, "end": 364}], "material": [{"text": "English", "start": 100, "end": 107}], "generic": [{"text": "implications", "start": 122, "end": 134}], "method": [{"text": "discourse processing algorithms", "start": 147, "end": 178}, {"text": "natural language generation system", "start": 415, "end": 449}]}, "relations": {"feature_of": [{"head": {"text": "demonstrative expressions", "start": 71, "end": 96}, "tail": {"text": "English", "start": 100, "end": 107}}], "used_for": [{"head": {"text": "implications", "start": 122, "end": 134}, "tail": {"text": "discourse processing algorithms", "start": 147, "end": 178}}, {"head": {"text": "anaphoric expressions", "start": 343, "end": 364}, "tail": {"text": "natural language generation system", "start": 415, "end": 449}}]}}, "schema": []}
{"input": "In the study of expressive speech communication, it is commonly accepted that the emotion perceived by the listener is a good approximation of the intended emotion conveyed by the speaker. This paper analyzes the validity of this assumption by comparing the mismatches between the assessments made by na ¨ ıve listeners and by the speakers that generated the data. The analysis is based on the hypothesis that people are better decoders of their own emotions. Therefore, self-assessments will be closer to the intended emotions. Using the IEMOCAP database, discrete -LRB- categorical -RRB- and continuous -LRB- attribute -RRB- emotional assessments evaluated by the actors and na ¨ ıve listeners are compared. The results indicate that there is a mismatch between the expression and perception of emotion. The speakers in the database assigned their own emotions to more specific emotional categories, which led to more extreme values in the activation-valence space.", "output": {"entities": {"task": [{"text": "expressive speech communication", "start": 16, "end": 47}, {"text": "discrete -LRB- categorical -RRB- and continuous -LRB- attribute -RRB- emotional assessments", "start": 557, "end": 648}], "material": [{"text": "IEMOCAP database", "start": 539, "end": 555}], "other_scientific_term": [{"text": "expression and perception of emotion", "start": 768, "end": 804}, {"text": "activation-valence space", "start": 942, "end": 966}], "generic": [{"text": "database", "start": 547, "end": 555}]}, "relations": {"used_for": [{"head": {"text": "IEMOCAP database", "start": 539, "end": 555}, "tail": {"text": "discrete -LRB- categorical -RRB- and continuous -LRB- attribute -RRB- emotional assessments", "start": 557, "end": 648}}]}}, "schema": []}
{"input": "The problem of blind separation of underdetermined instantaneous mixtures of independent signals is addressed through a method relying on nonstationarity of the original signals. The signals are assumed to be piecewise stationary with varying variances in different epochs. In comparison with previous works, in this paper it is assumed that the signals are not i.i.d. in each epoch, but obey a first-order autoregressive model. This model was shown to be more appropriate for blind separation of natural speech signals. A separation method is proposed that is nearly statistically efficient -LRB- approaching the corresponding Cramér-Rao lower bound -RRB-, if the separated signals obey the assumed model. In the case of natural speech signals, the method is shown to have separation accuracy better than the state-of-the-art methods.", "output": {"entities": {"task": [{"text": "blind separation of underdetermined instantaneous mixtures of independent signals", "start": 15, "end": 96}, {"text": "blind separation of natural speech signals.", "start": 477, "end": 520}], "generic": [{"text": "method", "start": 120, "end": 126}, {"text": "signals", "start": 89, "end": 96}, {"text": "signals", "start": 170, "end": 177}, {"text": "model", "start": 422, "end": 427}, {"text": "assumed model", "start": 692, "end": 705}, {"text": "method", "start": 534, "end": 540}, {"text": "methods", "start": 827, "end": 834}], "other_scientific_term": [{"text": "nonstationarity", "start": 138, "end": 153}, {"text": "original signals", "start": 161, "end": 177}, {"text": "Cramér-Rao lower bound -RRB-", "start": 628, "end": 656}, {"text": "natural speech signals", "start": 497, "end": 519}], "method": [{"text": "first-order autoregressive model", "start": 395, "end": 427}, {"text": "separation method", "start": 523, "end": 540}], "metric": [{"text": "separation accuracy", "start": 774, "end": 793}]}, "relations": {"used_for": [{"head": {"text": "method", "start": 120, "end": 126}, "tail": {"text": "blind separation of underdetermined instantaneous mixtures of independent signals", "start": 15, "end": 96}}, {"head": {"text": "nonstationarity", "start": 138, "end": 153}, "tail": {"text": "method", "start": 120, "end": 126}}, {"head": {"text": "first-order autoregressive model", "start": 395, "end": 427}, "tail": {"text": "signals", "start": 170, "end": 177}}, {"head": {"text": "model", "start": 422, "end": 427}, "tail": {"text": "blind separation of natural speech signals.", "start": 477, "end": 520}}, {"head": {"text": "method", "start": 534, "end": 540}, "tail": {"text": "natural speech signals", "start": 497, "end": 519}}, {"head": {"text": "methods", "start": 827, "end": 834}, "tail": {"text": "natural speech signals", "start": 497, "end": 519}}], "feature_of": [{"head": {"text": "Cramér-Rao lower bound -RRB-", "start": 628, "end": 656}, "tail": {"text": "separation method", "start": 523, "end": 540}}], "compare": [{"head": {"text": "method", "start": 534, "end": 540}, "tail": {"text": "methods", "start": 827, "end": 834}}], "evaluate_for": [{"head": {"text": "separation accuracy", "start": 774, "end": 793}, "tail": {"text": "method", "start": 534, "end": 540}}, {"head": {"text": "separation accuracy", "start": 774, "end": 793}, "tail": {"text": "methods", "start": 827, "end": 834}}]}}, "schema": []}
{"input": "This paper proposes to use a convolution kernel over parse trees to model syntactic structure information for relation extraction. Our study reveals that the syntactic structure features embedded in a parse tree are very effective for relation extraction and these features can be well captured by the convolution tree kernel. Evaluation on the ACE 2003 corpus shows that the convolution kernel over parse trees can achieve comparable performance with the previous best-reported feature-based methods on the 24 ACE relation subtypes. It also shows that our method significantly outperforms the previous two dependency tree kernels on the 5 ACE relation major types.", "output": {"entities": {"method": [{"text": "convolution kernel over parse trees", "start": 29, "end": 64}, {"text": "convolution tree kernel", "start": 302, "end": 325}, {"text": "convolution kernel over parse trees", "start": 376, "end": 411}, {"text": "feature-based methods", "start": 479, "end": 500}, {"text": "dependency tree kernels", "start": 607, "end": 630}], "other_scientific_term": [{"text": "syntactic structure information", "start": 74, "end": 105}, {"text": "syntactic structure features", "start": 158, "end": 186}, {"text": "parse tree", "start": 53, "end": 63}], "task": [{"text": "relation extraction", "start": 110, "end": 129}, {"text": "relation extraction", "start": 235, "end": 254}], "generic": [{"text": "features", "start": 178, "end": 186}, {"text": "method", "start": 493, "end": 499}], "material": [{"text": "ACE 2003 corpus", "start": 345, "end": 360}]}, "relations": {"used_for": [{"head": {"text": "convolution kernel over parse trees", "start": 29, "end": 64}, "tail": {"text": "syntactic structure information", "start": 74, "end": 105}}, {"head": {"text": "syntactic structure information", "start": 74, "end": 105}, "tail": {"text": "relation extraction", "start": 110, "end": 129}}, {"head": {"text": "syntactic structure features", "start": 158, "end": 186}, "tail": {"text": "relation extraction", "start": 235, "end": 254}}, {"head": {"text": "convolution tree kernel", "start": 302, "end": 325}, "tail": {"text": "features", "start": 178, "end": 186}}], "feature_of": [{"head": {"text": "syntactic structure features", "start": 158, "end": 186}, "tail": {"text": "parse tree", "start": 53, "end": 63}}], "evaluate_for": [{"head": {"text": "ACE 2003 corpus", "start": 345, "end": 360}, "tail": {"text": "convolution kernel over parse trees", "start": 376, "end": 411}}], "compare": [{"head": {"text": "feature-based methods", "start": 479, "end": 500}, "tail": {"text": "convolution kernel over parse trees", "start": 376, "end": 411}}, {"head": {"text": "method", "start": 493, "end": 499}, "tail": {"text": "dependency tree kernels", "start": 607, "end": 630}}]}}, "schema": []}
{"input": "This paper presents the results of automatically inducing a Combinatory Categorial Grammar -LRB- CCG -RRB- lexicon from a Turkish dependency treebank. The fact that Turkish is an agglutinating free word order language presents a challenge for language theories. We explored possible ways to obtain a compact lexicon, consistent with CCG principles, from a treebank which is an order of magnitude smaller than Penn WSJ.", "output": {"entities": {"task": [{"text": "automatically inducing a Combinatory Categorial Grammar -LRB- CCG -RRB- lexicon", "start": 35, "end": 114}], "other_scientific_term": [{"text": "Combinatory Categorial Grammar -LRB- CCG -RRB- lexicon", "start": 60, "end": 114}, {"text": "compact lexicon", "start": 300, "end": 315}], "material": [{"text": "Turkish dependency treebank", "start": 122, "end": 149}, {"text": "Turkish", "start": 122, "end": 129}, {"text": "agglutinating free word order language", "start": 179, "end": 217}, {"text": "Penn WSJ", "start": 409, "end": 417}], "method": [{"text": "language theories", "start": 243, "end": 260}, {"text": "CCG principles", "start": 333, "end": 347}], "generic": [{"text": "treebank", "start": 141, "end": 149}]}, "relations": {"part_of": [{"head": {"text": "Combinatory Categorial Grammar -LRB- CCG -RRB- lexicon", "start": 60, "end": 114}, "tail": {"text": "Turkish dependency treebank", "start": 122, "end": 149}}, {"head": {"text": "compact lexicon", "start": 300, "end": 315}, "tail": {"text": "treebank", "start": 141, "end": 149}}], "hyponym_of": [{"head": {"text": "Turkish", "start": 122, "end": 129}, "tail": {"text": "agglutinating free word order language", "start": 179, "end": 217}}], "compare": [{"head": {"text": "treebank", "start": 141, "end": 149}, "tail": {"text": "Penn WSJ", "start": 409, "end": 417}}]}}, "schema": []}
{"input": "While sentence extraction as an approach to summarization has been shown to work in documents of certain genres, because of the conversational nature of email communication where utterances are made in relation to one made previously, sentence extraction may not capture the necessary segments of dialogue that would make a summary coherent. In this paper, we present our work on the detection of question-answer pairs in an email conversation for the task of email summarization. We show that various features based on the structure of email-threads can be used to improve upon lexical similarity of discourse segments for question-answer pairing.", "output": {"entities": {"method": [{"text": "sentence extraction", "start": 6, "end": 25}, {"text": "sentence extraction", "start": 235, "end": 254}], "task": [{"text": "summarization", "start": 44, "end": 57}, {"text": "detection of question-answer pairs", "start": 384, "end": 418}, {"text": "email summarization", "start": 460, "end": 479}, {"text": "question-answer pairing", "start": 624, "end": 647}], "material": [{"text": "email communication", "start": 153, "end": 172}, {"text": "email conversation", "start": 425, "end": 443}], "other_scientific_term": [{"text": "features", "start": 502, "end": 510}, {"text": "structure of email-threads", "start": 524, "end": 550}, {"text": "lexical similarity", "start": 579, "end": 597}, {"text": "discourse segments", "start": 601, "end": 619}]}, "relations": {"used_for": [{"head": {"text": "sentence extraction", "start": 6, "end": 25}, "tail": {"text": "summarization", "start": 44, "end": 57}}, {"head": {"text": "detection of question-answer pairs", "start": 384, "end": 418}, "tail": {"text": "email summarization", "start": 460, "end": 479}}, {"head": {"text": "email conversation", "start": 425, "end": 443}, "tail": {"text": "detection of question-answer pairs", "start": 384, "end": 418}}, {"head": {"text": "features", "start": 502, "end": 510}, "tail": {"text": "lexical similarity", "start": 579, "end": 597}}, {"head": {"text": "features", "start": 502, "end": 510}, "tail": {"text": "question-answer pairing", "start": 624, "end": 647}}, {"head": {"text": "structure of email-threads", "start": 524, "end": 550}, "tail": {"text": "features", "start": 502, "end": 510}}], "feature_of": [{"head": {"text": "lexical similarity", "start": 579, "end": 597}, "tail": {"text": "discourse segments", "start": 601, "end": 619}}]}}, "schema": []}
{"input": "In this paper we discuss object detection when only a small number of training examples are given. Specifically, we show how to incorporate a simple prior on the distribution of natural images into support vector machines. SVMs are known to be robust to overfitting; however, a few training examples usually do not represent well the structure of the class. Thus the resulting detectors are not robust and highly depend on the choice of the training examples. We incorporate the prior on natural images by requiring that the separating hyperplane will not only yield a wide margin, but also that the corresponding positive half space will have a low probability to contain natural images -LRB- the background -RRB-. Our experiments on real data sets show that the resulting detector is more robust to the choice of training examples, and substantially improves both linear and kernel SVM when trained on 10 positive and 10 negative examples.", "output": {"entities": {"task": [{"text": "object detection", "start": 25, "end": 41}], "other_scientific_term": [{"text": "prior on the distribution of natural images", "start": 149, "end": 192}, {"text": "overfitting", "start": 254, "end": 265}, {"text": "prior on natural images", "start": 479, "end": 502}, {"text": "hyperplane", "start": 536, "end": 546}], "method": [{"text": "support vector machines", "start": 198, "end": 221}, {"text": "SVMs", "start": 223, "end": 227}, {"text": "detectors", "start": 377, "end": 386}, {"text": "detector", "start": 377, "end": 385}, {"text": "linear and kernel SVM", "start": 866, "end": 887}], "material": [{"text": "real data sets", "start": 735, "end": 749}]}, "relations": {"used_for": [{"head": {"text": "prior on the distribution of natural images", "start": 149, "end": 192}, "tail": {"text": "support vector machines", "start": 198, "end": 221}}, {"head": {"text": "SVMs", "start": 223, "end": 227}, "tail": {"text": "overfitting", "start": 254, "end": 265}}], "evaluate_for": [{"head": {"text": "real data sets", "start": 735, "end": 749}, "tail": {"text": "detector", "start": 377, "end": 385}}], "compare": [{"head": {"text": "detector", "start": 377, "end": 385}, "tail": {"text": "linear and kernel SVM", "start": 866, "end": 887}}]}}, "schema": []}
{"input": "Although the study of clustering is centered around an intuitively compelling goal, it has been very difficult to develop a unified framework for reasoning about it at a technical level, and profoundly diverse approaches to clustering abound in the research community. Here we suggest a formal perspective on the difficulty in finding such a unification, in the form of an impossibility theorem: for a set of three simple properties, we show that there is no clustering function satisfying all three. Relaxations of these properties expose some of the interesting -LRB- and unavoidable -RRB- trade-offs at work in well-studied clustering techniques such as single-linkage, sum-of-pairs, k-means, and k-median.", "output": {"entities": {"task": [{"text": "clustering", "start": 22, "end": 32}, {"text": "reasoning", "start": 146, "end": 155}], "method": [{"text": "unified framework", "start": 124, "end": 141}, {"text": "unification", "start": 342, "end": 353}, {"text": "impossibility theorem", "start": 373, "end": 394}, {"text": "well-studied clustering techniques", "start": 614, "end": 648}, {"text": "single-linkage", "start": 657, "end": 671}, {"text": "sum-of-pairs", "start": 673, "end": 685}, {"text": "k-means", "start": 687, "end": 694}, {"text": "k-median", "start": 700, "end": 708}], "other_scientific_term": [{"text": "clustering function", "start": 459, "end": 478}]}, "relations": {"used_for": [{"head": {"text": "unified framework", "start": 124, "end": 141}, "tail": {"text": "reasoning", "start": 146, "end": 155}}], "hyponym_of": [{"head": {"text": "single-linkage", "start": 657, "end": 671}, "tail": {"text": "well-studied clustering techniques", "start": 614, "end": 648}}, {"head": {"text": "sum-of-pairs", "start": 673, "end": 685}, "tail": {"text": "well-studied clustering techniques", "start": 614, "end": 648}}, {"head": {"text": "k-means", "start": 687, "end": 694}, "tail": {"text": "well-studied clustering techniques", "start": 614, "end": 648}}, {"head": {"text": "k-median", "start": 700, "end": 708}, "tail": {"text": "well-studied clustering techniques", "start": 614, "end": 648}}], "conjunction": [{"head": {"text": "single-linkage", "start": 657, "end": 671}, "tail": {"text": "sum-of-pairs", "start": 673, "end": 685}}, {"head": {"text": "sum-of-pairs", "start": 673, "end": 685}, "tail": {"text": "k-means", "start": 687, "end": 694}}, {"head": {"text": "k-means", "start": 687, "end": 694}, "tail": {"text": "k-median", "start": 700, "end": 708}}]}}, "schema": []}
{"input": "We investigate independent and relevant event-based extractive mutli-document summarization approaches. In this paper, events are defined as event terms and associated event elements. With independent approach, we identify important contents by frequency of events. With relevant approach, we identify important contents by PageRank algorithm on the event map constructed from documents. Experimental results are encouraging.", "output": {"entities": {"method": [{"text": "independent and relevant event-based extractive mutli-document summarization approaches", "start": 15, "end": 102}, {"text": "independent approach", "start": 189, "end": 209}, {"text": "relevant approach", "start": 271, "end": 288}, {"text": "PageRank algorithm", "start": 324, "end": 342}], "other_scientific_term": [{"text": "event map", "start": 350, "end": 359}], "material": [{"text": "documents", "start": 377, "end": 386}]}, "relations": {"used_for": [{"head": {"text": "PageRank algorithm", "start": 324, "end": 342}, "tail": {"text": "relevant approach", "start": 271, "end": 288}}, {"head": {"text": "event map", "start": 350, "end": 359}, "tail": {"text": "PageRank algorithm", "start": 324, "end": 342}}, {"head": {"text": "documents", "start": 377, "end": 386}, "tail": {"text": "event map", "start": 350, "end": 359}}]}}, "schema": []}
{"input": "We present a scanning method that recovers dense sub-pixel camera-projector correspondence without requiring any photometric calibration nor preliminary knowledge of their relative geometry. Subpixel accuracy is achieved by considering several zero-crossings defined by the difference between pairs of unstructured patterns. We use gray-level band-pass white noise patterns that increase robustness to indirect lighting and scene discontinuities. Simulated and experimental results show that our method recovers scene geometry with high subpixel precision, and that it can handle many challenges of active reconstruction systems. We compare our results to state of the art methods such as mi-cro phase shifting and modulated phase shifting.", "output": {"entities": {"method": [{"text": "scanning method", "start": 13, "end": 28}, {"text": "mi-cro phase shifting", "start": 689, "end": 710}, {"text": "modulated phase shifting", "start": 715, "end": 739}], "other_scientific_term": [{"text": "dense sub-pixel camera-projector correspondence", "start": 43, "end": 90}, {"text": "photometric calibration", "start": 113, "end": 136}, {"text": "relative geometry", "start": 172, "end": 189}, {"text": "zero-crossings", "start": 244, "end": 258}, {"text": "unstructured patterns", "start": 302, "end": 323}, {"text": "gray-level band-pass white noise patterns", "start": 332, "end": 373}, {"text": "indirect lighting", "start": 402, "end": 419}, {"text": "scene discontinuities", "start": 424, "end": 445}, {"text": "scene geometry", "start": 512, "end": 526}], "metric": [{"text": "Subpixel accuracy", "start": 191, "end": 208}, {"text": "robustness", "start": 388, "end": 398}, {"text": "subpixel precision", "start": 537, "end": 555}], "generic": [{"text": "method", "start": 22, "end": 28}, {"text": "it", "start": 92, "end": 94}, {"text": "state of the art methods", "start": 656, "end": 680}], "task": [{"text": "active reconstruction systems", "start": 599, "end": 628}]}, "relations": {"used_for": [{"head": {"text": "scanning method", "start": 13, "end": 28}, "tail": {"text": "dense sub-pixel camera-projector correspondence", "start": 43, "end": 90}}, {"head": {"text": "zero-crossings", "start": 244, "end": 258}, "tail": {"text": "Subpixel accuracy", "start": 191, "end": 208}}, {"head": {"text": "method", "start": 22, "end": 28}, "tail": {"text": "scene geometry", "start": 512, "end": 526}}, {"head": {"text": "it", "start": 92, "end": 94}, "tail": {"text": "active reconstruction systems", "start": 599, "end": 628}}], "evaluate_for": [{"head": {"text": "robustness", "start": 388, "end": 398}, "tail": {"text": "gray-level band-pass white noise patterns", "start": 332, "end": 373}}], "feature_of": [{"head": {"text": "indirect lighting", "start": 402, "end": 419}, "tail": {"text": "robustness", "start": 388, "end": 398}}, {"head": {"text": "scene discontinuities", "start": 424, "end": 445}, "tail": {"text": "robustness", "start": 388, "end": 398}}, {"head": {"text": "subpixel precision", "start": 537, "end": 555}, "tail": {"text": "scene geometry", "start": 512, "end": 526}}], "conjunction": [{"head": {"text": "indirect lighting", "start": 402, "end": 419}, "tail": {"text": "scene discontinuities", "start": 424, "end": 445}}, {"head": {"text": "mi-cro phase shifting", "start": 689, "end": 710}, "tail": {"text": "modulated phase shifting", "start": 715, "end": 739}}], "hyponym_of": [{"head": {"text": "mi-cro phase shifting", "start": 689, "end": 710}, "tail": {"text": "state of the art methods", "start": 656, "end": 680}}, {"head": {"text": "modulated phase shifting", "start": 715, "end": 739}, "tail": {"text": "state of the art methods", "start": 656, "end": 680}}]}}, "schema": []}
{"input": "This paper describes a novel system for acquiring adjectival subcategorization frames -LRB- scfs -RRB- and associated frequency information from English corpus data. The system incorporates a decision-tree classifier for 30 scf types which tests for the presence of grammatical relations -LRB- grs -RRB- in the output of a robust statistical parser. It uses a powerful pattern-matching language to classify grs into frames hierarchically in a way that mirrors inheritance-based lexica. The experiments show that the system is able to detect scf types with 70% precision and 66% recall rate. A new tool for linguistic annotation of scfs in corpus data is also introduced which can considerably alleviate the process of obtaining training and test data for subcategorization acquisition.", "output": {"entities": {"generic": [{"text": "system", "start": 29, "end": 35}, {"text": "system", "start": 170, "end": 176}, {"text": "It", "start": 350, "end": 352}, {"text": "system", "start": 516, "end": 522}, {"text": "tool", "start": 597, "end": 601}], "task": [{"text": "acquiring adjectival subcategorization frames -LRB- scfs -RRB-", "start": 40, "end": 102}, {"text": "linguistic annotation of scfs", "start": 606, "end": 635}, {"text": "subcategorization acquisition", "start": 755, "end": 784}], "other_scientific_term": [{"text": "subcategorization frames -LRB- scfs -RRB-", "start": 61, "end": 102}, {"text": "grammatical relations -LRB- grs -RRB-", "start": 266, "end": 303}, {"text": "pattern-matching language", "start": 369, "end": 394}, {"text": "grs", "start": 294, "end": 297}, {"text": "inheritance-based lexica", "start": 460, "end": 484}, {"text": "scfs", "start": 92, "end": 96}], "material": [{"text": "English corpus data", "start": 145, "end": 164}, {"text": "training and test data", "start": 728, "end": 750}], "method": [{"text": "decision-tree classifier", "start": 192, "end": 216}, {"text": "robust statistical parser", "start": 323, "end": 348}], "metric": [{"text": "precision", "start": 560, "end": 569}, {"text": "recall", "start": 578, "end": 584}]}, "relations": {"used_for": [{"head": {"text": "system", "start": 29, "end": 35}, "tail": {"text": "acquiring adjectival subcategorization frames -LRB- scfs -RRB-", "start": 40, "end": 102}}, {"head": {"text": "decision-tree classifier", "start": 192, "end": 216}, "tail": {"text": "grammatical relations -LRB- grs -RRB-", "start": 266, "end": 303}}, {"head": {"text": "pattern-matching language", "start": 369, "end": 394}, "tail": {"text": "It", "start": 350, "end": 352}}, {"head": {"text": "pattern-matching language", "start": 369, "end": 394}, "tail": {"text": "grs", "start": 294, "end": 297}}, {"head": {"text": "tool", "start": 597, "end": 601}, "tail": {"text": "linguistic annotation of scfs", "start": 606, "end": 635}}, {"head": {"text": "training and test data", "start": 728, "end": 750}, "tail": {"text": "subcategorization acquisition", "start": 755, "end": 784}}], "part_of": [{"head": {"text": "decision-tree classifier", "start": 192, "end": 216}, "tail": {"text": "system", "start": 170, "end": 176}}], "evaluate_for": [{"head": {"text": "precision", "start": 560, "end": 569}, "tail": {"text": "system", "start": 516, "end": 522}}, {"head": {"text": "recall", "start": 578, "end": 584}, "tail": {"text": "system", "start": 516, "end": 522}}], "conjunction": [{"head": {"text": "precision", "start": 560, "end": 569}, "tail": {"text": "recall", "start": 578, "end": 584}}]}}, "schema": []}
{"input": "Machine transliteration/back-transliteration plays an important role in many multilingual speech and language applications. In this paper, a novel framework for machine transliteration/backtransliteration that allows us to carry out direct orthographical mapping -LRB- DOM -RRB- between two different languages is presented. Under this framework, a joint source-channel transliteration model, also called n-gram transliteration model -LRB- n-gram TM -RRB-, is further proposed to model the transliteration process. We evaluate the proposed methods through several transliteration/backtransliteration experiments for English/Chinese and English/Japanese language pairs. Our study reveals that the proposed method not only reduces an extensive system development effort but also improves the transliteration accuracy significantly.", "output": {"entities": {"task": [{"text": "Machine transliteration/back-transliteration", "start": 0, "end": 44}, {"text": "multilingual speech and language applications", "start": 77, "end": 122}, {"text": "machine transliteration/backtransliteration", "start": 161, "end": 204}, {"text": "transliteration/backtransliteration", "start": 169, "end": 204}], "generic": [{"text": "framework", "start": 147, "end": 156}, {"text": "framework", "start": 336, "end": 345}, {"text": "methods", "start": 540, "end": 547}, {"text": "method", "start": 540, "end": 546}], "method": [{"text": "direct orthographical mapping -LRB- DOM -RRB-", "start": 233, "end": 278}, {"text": "joint source-channel transliteration model", "start": 349, "end": 391}, {"text": "n-gram transliteration model -LRB- n-gram TM -RRB-", "start": 405, "end": 455}, {"text": "transliteration process", "start": 490, "end": 513}], "material": [{"text": "English/Chinese and English/Japanese language pairs", "start": 616, "end": 667}], "metric": [{"text": "transliteration accuracy", "start": 790, "end": 814}]}, "relations": {"used_for": [{"head": {"text": "Machine transliteration/back-transliteration", "start": 0, "end": 44}, "tail": {"text": "multilingual speech and language applications", "start": 77, "end": 122}}, {"head": {"text": "framework", "start": 147, "end": 156}, "tail": {"text": "machine transliteration/backtransliteration", "start": 161, "end": 204}}, {"head": {"text": "machine transliteration/backtransliteration", "start": 161, "end": 204}, "tail": {"text": "direct orthographical mapping -LRB- DOM -RRB-", "start": 233, "end": 278}}, {"head": {"text": "framework", "start": 336, "end": 345}, "tail": {"text": "joint source-channel transliteration model", "start": 349, "end": 391}}, {"head": {"text": "n-gram transliteration model -LRB- n-gram TM -RRB-", "start": 405, "end": 455}, "tail": {"text": "transliteration process", "start": 490, "end": 513}}, {"head": {"text": "transliteration/backtransliteration", "start": 169, "end": 204}, "tail": {"text": "English/Chinese and English/Japanese language pairs", "start": 616, "end": 667}}], "evaluate_for": [{"head": {"text": "transliteration/backtransliteration", "start": 169, "end": 204}, "tail": {"text": "methods", "start": 540, "end": 547}}, {"head": {"text": "transliteration accuracy", "start": 790, "end": 814}, "tail": {"text": "method", "start": 540, "end": 546}}]}}, "schema": []}
{"input": "A bio-inspired model for an analog programmable array processor -LRB- APAP -RRB-, based on studies on the vertebrate retina, has permitted the realization of complex programmable spatio-temporal dynamics in VLSI. This model mimics the way in which images are processed in the visual pathway, rendering a feasible alternative for the implementation of early vision applications in standard technologies. A prototype chip has been designed and fabricated in a 0.5 µm standard CMOS process. Computing power per area and power consumption is amongst the highest reported for a single chip. Design challenges, trade-offs and some experimental results are presented in this paper.", "output": {"entities": {"method": [{"text": "bio-inspired model", "start": 2, "end": 20}, {"text": "visual pathway", "start": 276, "end": 290}], "task": [{"text": "analog programmable array processor -LRB- APAP -RRB-", "start": 28, "end": 80}, {"text": "VLSI", "start": 207, "end": 211}, {"text": "vision applications", "start": 357, "end": 376}], "other_scientific_term": [{"text": "vertebrate retina", "start": 106, "end": 123}, {"text": "complex programmable spatio-temporal dynamics", "start": 158, "end": 203}, {"text": "prototype chip", "start": 405, "end": 419}, {"text": "CMOS process", "start": 474, "end": 486}], "generic": [{"text": "model", "start": 15, "end": 20}], "material": [{"text": "images", "start": 248, "end": 254}], "metric": [{"text": "Computing power per area", "start": 488, "end": 512}, {"text": "power consumption", "start": 517, "end": 534}]}, "relations": {"used_for": [{"head": {"text": "bio-inspired model", "start": 2, "end": 20}, "tail": {"text": "analog programmable array processor -LRB- APAP -RRB-", "start": 28, "end": 80}}, {"head": {"text": "bio-inspired model", "start": 2, "end": 20}, "tail": {"text": "complex programmable spatio-temporal dynamics", "start": 158, "end": 203}}, {"head": {"text": "vertebrate retina", "start": 106, "end": 123}, "tail": {"text": "bio-inspired model", "start": 2, "end": 20}}, {"head": {"text": "visual pathway", "start": 276, "end": 290}, "tail": {"text": "images", "start": 248, "end": 254}}], "feature_of": [{"head": {"text": "complex programmable spatio-temporal dynamics", "start": 158, "end": 203}, "tail": {"text": "VLSI", "start": 207, "end": 211}}], "conjunction": [{"head": {"text": "Computing power per area", "start": 488, "end": 512}, "tail": {"text": "power consumption", "start": 517, "end": 534}}]}}, "schema": []}
{"input": "Determiners play an important role in conveying the meaning of an utterance, but they have often been disregarded, perhaps because it seemed more important to devise methods to grasp the global meaning of a sentence, even if not in a precise way. Another problem with determiners is their inherent ambiguity. In this paper we propose a logical formalism, which, among other things, is suitable for representing determiners without forcing a particular interpretation when their meaning is still not clear.", "output": {"entities": {"method": [{"text": "Determiners", "start": 0, "end": 11}, {"text": "determiners", "start": 268, "end": 279}, {"text": "logical formalism", "start": 336, "end": 353}], "other_scientific_term": [{"text": "ambiguity", "start": 298, "end": 307}], "task": [{"text": "determiners", "start": 411, "end": 422}]}, "relations": {"feature_of": [{"head": {"text": "ambiguity", "start": 298, "end": 307}, "tail": {"text": "determiners", "start": 268, "end": 279}}], "used_for": [{"head": {"text": "logical formalism", "start": 336, "end": 353}, "tail": {"text": "determiners", "start": 411, "end": 422}}]}}, "schema": []}
{"input": "We investigate the verbal and nonverbal means for grounding, and propose a design for embodied conversational agents that relies on both kinds of signals to establish common ground in human-computer interaction. We analyzed eye gaze, head nods and attentional focus in the context of a direction-giving task. The distribution of nonverbal behaviors differed depending on the type of dialogue move being grounded, and the overall pattern reflected a monitoring of lack of negative feedback. Based on these results, we present an ECA that uses verbal and nonverbal grounding acts to update dialogue state.", "output": {"entities": {"method": [{"text": "verbal and nonverbal means", "start": 19, "end": 45}, {"text": "embodied conversational agents", "start": 86, "end": 116}, {"text": "ECA", "start": 528, "end": 531}], "task": [{"text": "grounding", "start": 50, "end": 59}, {"text": "common ground", "start": 167, "end": 180}, {"text": "human-computer interaction", "start": 184, "end": 210}, {"text": "direction-giving task", "start": 286, "end": 307}], "generic": [{"text": "design", "start": 75, "end": 81}], "other_scientific_term": [{"text": "eye gaze", "start": 224, "end": 232}, {"text": "head nods", "start": 234, "end": 243}, {"text": "attentional focus", "start": 248, "end": 265}, {"text": "nonverbal behaviors", "start": 329, "end": 348}, {"text": "dialogue move", "start": 383, "end": 396}, {"text": "negative feedback", "start": 471, "end": 488}, {"text": "verbal and nonverbal grounding acts", "start": 542, "end": 577}, {"text": "dialogue state", "start": 588, "end": 602}]}, "relations": {"used_for": [{"head": {"text": "verbal and nonverbal means", "start": 19, "end": 45}, "tail": {"text": "grounding", "start": 50, "end": 59}}, {"head": {"text": "design", "start": 75, "end": 81}, "tail": {"text": "embodied conversational agents", "start": 86, "end": 116}}, {"head": {"text": "common ground", "start": 167, "end": 180}, "tail": {"text": "human-computer interaction", "start": 184, "end": 210}}, {"head": {"text": "verbal and nonverbal grounding acts", "start": 542, "end": 577}, "tail": {"text": "ECA", "start": 528, "end": 531}}, {"head": {"text": "verbal and nonverbal grounding acts", "start": 542, "end": 577}, "tail": {"text": "dialogue state", "start": 588, "end": 602}}], "conjunction": [{"head": {"text": "eye gaze", "start": 224, "end": 232}, "tail": {"text": "head nods", "start": 234, "end": 243}}, {"head": {"text": "head nods", "start": 234, "end": 243}, "tail": {"text": "attentional focus", "start": 248, "end": 265}}], "part_of": [{"head": {"text": "eye gaze", "start": 224, "end": 232}, "tail": {"text": "direction-giving task", "start": 286, "end": 307}}, {"head": {"text": "head nods", "start": 234, "end": 243}, "tail": {"text": "direction-giving task", "start": 286, "end": 307}}, {"head": {"text": "attentional focus", "start": 248, "end": 265}, "tail": {"text": "direction-giving task", "start": 286, "end": 307}}]}}, "schema": []}
{"input": "Sentence boundary detection in speech is important for enriching speech recognition output, making it easier for humans to read and downstream modules to process. In previous work, we have developed hidden Markov model -LRB- HMM -RRB- and maximum entropy -LRB- Maxent -RRB- classifiers that integrate textual and prosodic knowledge sources for detecting sentence boundaries. In this paper, we evaluate the use of a conditional random field -LRB- CRF -RRB- for this task and relate results with this model to our prior work. We evaluate across two corpora -LRB- conversational telephone speech and broadcast news speech -RRB- on both human transcriptions and speech recognition output. In general, our CRF model yields a lower error rate than the HMM and Max-ent models on the NIST sentence boundary detection task in speech, although it is interesting to note that the best results are achieved by three-way voting among the classifiers. This probably occurs because each model has different strengths and weaknesses for modeling the knowledge sources.", "output": {"entities": {"task": [{"text": "Sentence boundary detection", "start": 0, "end": 27}, {"text": "detecting sentence boundaries", "start": 344, "end": 373}], "material": [{"text": "speech", "start": 31, "end": 37}, {"text": "textual and prosodic knowledge sources", "start": 301, "end": 339}, {"text": "conversational telephone speech", "start": 561, "end": 592}, {"text": "broadcast news speech", "start": 597, "end": 618}, {"text": "NIST sentence boundary detection task", "start": 776, "end": 813}, {"text": "speech", "start": 65, "end": 71}, {"text": "knowledge sources", "start": 322, "end": 339}], "other_scientific_term": [{"text": "speech recognition output", "start": 65, "end": 90}, {"text": "human transcriptions", "start": 633, "end": 653}, {"text": "speech recognition output", "start": 658, "end": 683}], "generic": [{"text": "it", "start": 78, "end": 80}, {"text": "task", "start": 465, "end": 469}, {"text": "model", "start": 213, "end": 218}, {"text": "corpora", "start": 547, "end": 554}, {"text": "model", "start": 499, "end": 504}], "method": [{"text": "hidden Markov model -LRB- HMM -RRB- and maximum entropy -LRB- Maxent -RRB- classifiers", "start": 199, "end": 285}, {"text": "conditional random field -LRB- CRF -RRB-", "start": 415, "end": 455}, {"text": "CRF model", "start": 701, "end": 710}, {"text": "HMM and Max-ent models", "start": 746, "end": 768}, {"text": "three-way voting", "start": 898, "end": 914}, {"text": "classifiers", "start": 274, "end": 285}], "metric": [{"text": "error rate", "start": 726, "end": 736}]}, "relations": {"used_for": [{"head": {"text": "Sentence boundary detection", "start": 0, "end": 27}, "tail": {"text": "speech recognition output", "start": 65, "end": 90}}, {"head": {"text": "speech", "start": 31, "end": 37}, "tail": {"text": "Sentence boundary detection", "start": 0, "end": 27}}, {"head": {"text": "hidden Markov model -LRB- HMM -RRB- and maximum entropy -LRB- Maxent -RRB- classifiers", "start": 199, "end": 285}, "tail": {"text": "detecting sentence boundaries", "start": 344, "end": 373}}, {"head": {"text": "textual and prosodic knowledge sources", "start": 301, "end": 339}, "tail": {"text": "hidden Markov model -LRB- HMM -RRB- and maximum entropy -LRB- Maxent -RRB- classifiers", "start": 199, "end": 285}}, {"head": {"text": "conditional random field -LRB- CRF -RRB-", "start": 415, "end": 455}, "tail": {"text": "task", "start": 465, "end": 469}}, {"head": {"text": "classifiers", "start": 274, "end": 285}, "tail": {"text": "three-way voting", "start": 898, "end": 914}}, {"head": {"text": "model", "start": 499, "end": 504}, "tail": {"text": "knowledge sources", "start": 322, "end": 339}}], "evaluate_for": [{"head": {"text": "corpora", "start": 547, "end": 554}, "tail": {"text": "human transcriptions", "start": 633, "end": 653}}, {"head": {"text": "corpora", "start": 547, "end": 554}, "tail": {"text": "speech recognition output", "start": 658, "end": 683}}, {"head": {"text": "error rate", "start": 726, "end": 736}, "tail": {"text": "CRF model", "start": 701, "end": 710}}, {"head": {"text": "error rate", "start": 726, "end": 736}, "tail": {"text": "HMM and Max-ent models", "start": 746, "end": 768}}, {"head": {"text": "NIST sentence boundary detection task", "start": 776, "end": 813}, "tail": {"text": "CRF model", "start": 701, "end": 710}}, {"head": {"text": "NIST sentence boundary detection task", "start": 776, "end": 813}, "tail": {"text": "HMM and Max-ent models", "start": 746, "end": 768}}], "hyponym_of": [{"head": {"text": "conversational telephone speech", "start": 561, "end": 592}, "tail": {"text": "corpora", "start": 547, "end": 554}}, {"head": {"text": "broadcast news speech", "start": 597, "end": 618}, "tail": {"text": "corpora", "start": 547, "end": 554}}], "conjunction": [{"head": {"text": "conversational telephone speech", "start": 561, "end": 592}, "tail": {"text": "broadcast news speech", "start": 597, "end": 618}}, {"head": {"text": "human transcriptions", "start": 633, "end": 653}, "tail": {"text": "speech recognition output", "start": 658, "end": 683}}], "compare": [{"head": {"text": "CRF model", "start": 701, "end": 710}, "tail": {"text": "HMM and Max-ent models", "start": 746, "end": 768}}], "feature_of": [{"head": {"text": "speech", "start": 65, "end": 71}, "tail": {"text": "NIST sentence boundary detection task", "start": 776, "end": 813}}]}}, "schema": []}
{"input": "We propose a novel approach to associate objects across multiple PTZ cameras that can be used to perform camera handoff in wide-area surveillance scenarios. While previous approaches relied on geometric, appearance, or correlation-based information for establishing correspondences between static cameras, they each have well-known limitations and are not extendable to wide-area settings with PTZ cameras. In our approach, the slave camera only passively follows the target -LRB- by loose registration with the master -RRB- and bootstraps itself from its own incoming imagery, thus effectively circumventing the problems faced by previous approaches and avoiding the need to perform any model transfer. Towards this goal, we also propose a novel Multiple Instance Learning -LRB- MIL -RRB- formulation for the problem based on the logistic softmax function of covariance-based region features within a MAP estimation framework. We demonstrate our approach with multiple PTZ camera sequences in typical outdoor surveillance settings and show a comparison with state-of-the-art approaches.", "output": {"entities": {"generic": [{"text": "approach", "start": 19, "end": 27}, {"text": "approaches", "start": 172, "end": 182}, {"text": "approach", "start": 172, "end": 180}, {"text": "approaches", "start": 640, "end": 650}, {"text": "approach", "start": 414, "end": 422}, {"text": "state-of-the-art approaches", "start": 1059, "end": 1086}], "other_scientific_term": [{"text": "PTZ cameras", "start": 65, "end": 76}, {"text": "geometric, appearance, or correlation-based information", "start": 193, "end": 248}, {"text": "static cameras", "start": 290, "end": 304}, {"text": "wide-area settings", "start": 370, "end": 388}, {"text": "PTZ cameras", "start": 394, "end": 405}, {"text": "slave camera", "start": 428, "end": 440}, {"text": "model transfer", "start": 688, "end": 702}, {"text": "logistic softmax function of covariance-based region features", "start": 831, "end": 892}, {"text": "multiple PTZ camera sequences", "start": 961, "end": 990}, {"text": "outdoor surveillance settings", "start": 1002, "end": 1031}], "task": [{"text": "camera handoff in wide-area surveillance scenarios", "start": 105, "end": 155}], "method": [{"text": "Multiple Instance Learning -LRB- MIL -RRB- formulation", "start": 747, "end": 801}, {"text": "MAP estimation framework", "start": 902, "end": 926}]}, "relations": {"used_for": [{"head": {"text": "approach", "start": 19, "end": 27}, "tail": {"text": "camera handoff in wide-area surveillance scenarios", "start": 105, "end": 155}}, {"head": {"text": "geometric, appearance, or correlation-based information", "start": 193, "end": 248}, "tail": {"text": "approaches", "start": 172, "end": 182}}, {"head": {"text": "logistic softmax function of covariance-based region features", "start": 831, "end": 892}, "tail": {"text": "Multiple Instance Learning -LRB- MIL -RRB- formulation", "start": 747, "end": 801}}, {"head": {"text": "MAP estimation framework", "start": 902, "end": 926}, "tail": {"text": "Multiple Instance Learning -LRB- MIL -RRB- formulation", "start": 747, "end": 801}}, {"head": {"text": "approach", "start": 414, "end": 422}, "tail": {"text": "outdoor surveillance settings", "start": 1002, "end": 1031}}, {"head": {"text": "multiple PTZ camera sequences", "start": 961, "end": 990}, "tail": {"text": "approach", "start": 414, "end": 422}}], "compare": [{"head": {"text": "approach", "start": 414, "end": 422}, "tail": {"text": "state-of-the-art approaches", "start": 1059, "end": 1086}}]}}, "schema": []}
{"input": "This paper solves a specialized regression problem to obtain sampling probabilities for records in databases. The goal is to sample a small set of records over which evaluating aggregate queries can be done both efficiently and accurately. We provide a principled and provable solution for this problem; it is parameterless and requires no data insights. Unlike standard regression problems, the loss is inversely proportional to the regressed-to values. Moreover, a cost zero solution always exists and can only be excluded by hard budget constraints. A unique form of reg-ularization is also needed. We provide an efficient and simple regularized Empirical Risk Minimization -LRB- ERM -RRB- algorithm along with a theoretical generalization result. Our extensive experimental results significantly improve over both uniform sampling and standard stratified sampling which are de-facto the industry standards.", "output": {"entities": {"task": [{"text": "specialized regression problem", "start": 20, "end": 50}, {"text": "regression problems", "start": 371, "end": 390}], "other_scientific_term": [{"text": "sampling probabilities", "start": 61, "end": 83}, {"text": "aggregate queries", "start": 177, "end": 194}, {"text": "loss", "start": 396, "end": 400}, {"text": "regressed-to values", "start": 434, "end": 453}, {"text": "hard budget constraints", "start": 528, "end": 551}, {"text": "reg-ularization", "start": 570, "end": 585}], "material": [{"text": "records", "start": 88, "end": 95}, {"text": "databases", "start": 99, "end": 108}, {"text": "records", "start": 147, "end": 154}], "method": [{"text": "principled and provable solution", "start": 253, "end": 285}, {"text": "cost zero solution", "start": 467, "end": 485}, {"text": "regularized Empirical Risk Minimization -LRB- ERM -RRB- algorithm", "start": 637, "end": 702}, {"text": "uniform sampling", "start": 818, "end": 834}, {"text": "stratified sampling", "start": 848, "end": 867}], "generic": [{"text": "problem", "start": 43, "end": 50}, {"text": "it", "start": 78, "end": 80}]}, "relations": {"used_for": [{"head": {"text": "specialized regression problem", "start": 20, "end": 50}, "tail": {"text": "sampling probabilities", "start": 61, "end": 83}}, {"head": {"text": "sampling probabilities", "start": 61, "end": 83}, "tail": {"text": "records", "start": 88, "end": 95}}, {"head": {"text": "principled and provable solution", "start": 253, "end": 285}, "tail": {"text": "problem", "start": 43, "end": 50}}, {"head": {"text": "hard budget constraints", "start": 528, "end": 551}, "tail": {"text": "cost zero solution", "start": 467, "end": 485}}], "part_of": [{"head": {"text": "records", "start": 88, "end": 95}, "tail": {"text": "databases", "start": 99, "end": 108}}], "evaluate_for": [{"head": {"text": "aggregate queries", "start": 177, "end": 194}, "tail": {"text": "records", "start": 147, "end": 154}}], "conjunction": [{"head": {"text": "uniform sampling", "start": 818, "end": 834}, "tail": {"text": "stratified sampling", "start": 848, "end": 867}}]}}, "schema": []}
{"input": "We consider the problem of computing the Kullback-Leibler distance, also called the relative entropy, between a probabilistic context-free grammar and a probabilistic finite automaton. We show that there is a closed-form -LRB- analytical -RRB- solution for one part of the Kullback-Leibler distance, viz the cross-entropy. We discuss several applications of the result to the problem of distributional approximation of probabilistic context-free grammars by means of probabilistic finite automata.", "output": {"entities": {"method": [{"text": "Kullback-Leibler distance", "start": 41, "end": 66}, {"text": "relative entropy", "start": 84, "end": 100}, {"text": "probabilistic context-free grammar", "start": 112, "end": 146}, {"text": "closed-form -LRB- analytical -RRB- solution", "start": 209, "end": 252}, {"text": "Kullback-Leibler distance", "start": 273, "end": 298}, {"text": "cross-entropy", "start": 308, "end": 321}, {"text": "probabilistic context-free grammars", "start": 419, "end": 454}, {"text": "probabilistic finite automata", "start": 467, "end": 496}], "other_scientific_term": [{"text": "probabilistic finite automaton", "start": 153, "end": 183}], "task": [{"text": "distributional approximation", "start": 387, "end": 415}]}, "relations": {"compare": [{"head": {"text": "probabilistic context-free grammar", "start": 112, "end": 146}, "tail": {"text": "probabilistic finite automaton", "start": 153, "end": 183}}], "used_for": [{"head": {"text": "closed-form -LRB- analytical -RRB- solution", "start": 209, "end": 252}, "tail": {"text": "Kullback-Leibler distance", "start": 273, "end": 298}}, {"head": {"text": "closed-form -LRB- analytical -RRB- solution", "start": 209, "end": 252}, "tail": {"text": "cross-entropy", "start": 308, "end": 321}}, {"head": {"text": "probabilistic finite automata", "start": 467, "end": 496}, "tail": {"text": "distributional approximation", "start": 387, "end": 415}}], "part_of": [{"head": {"text": "cross-entropy", "start": 308, "end": 321}, "tail": {"text": "Kullback-Leibler distance", "start": 273, "end": 298}}], "feature_of": [{"head": {"text": "distributional approximation", "start": 387, "end": 415}, "tail": {"text": "probabilistic context-free grammars", "start": 419, "end": 454}}]}}, "schema": []}
{"input": "In spite of over two decades of intense research, illumination and pose invariance remain prohibitively challenging aspects of face recognition for most practical applications. The objective of this work is to recognize faces using video sequences both for training and recognition input, in a realistic, unconstrained setup in which lighting, pose and user motion pattern have a wide variability and face images are of low resolution. In particular there are three areas of novelty: -LRB- i -RRB- we show how a photometric model of image formation can be combined with a statistical model of generic face appearance variation, learnt offline, to generalize in the presence of extreme illumination changes; -LRB- ii -RRB- we use the smoothness of geodesically local appearance manifold structure and a robust same-identity likelihood to achieve invariance to unseen head poses; and -LRB- iii -RRB- we introduce an accurate video sequence'' reillumination'' algorithm to achieve robustness to face motion patterns in video. We describe a fully automatic recognition system based on the proposed method and an extensive evaluation on 171 individuals and over 1300 video sequences with extreme illumination, pose and head motion variation. On this challenging data set our system consistently demonstrated a nearly perfect recognition rate -LRB- over 99.7% on all three databases -RRB-, significantly out-performing state-of-the-art commercial software and methods from the literature.", "output": {"entities": {"task": [{"text": "illumination", "start": 50, "end": 62}, {"text": "pose invariance", "start": 67, "end": 82}, {"text": "face recognition", "start": 127, "end": 143}, {"text": "image formation", "start": 533, "end": 548}, {"text": "generic face appearance variation", "start": 593, "end": 626}], "material": [{"text": "video sequences", "start": 232, "end": 247}, {"text": "face images", "start": 401, "end": 412}, {"text": "video", "start": 232, "end": 237}, {"text": "video sequences", "start": 1162, "end": 1177}, {"text": "databases", "start": 1367, "end": 1376}], "other_scientific_term": [{"text": "lighting", "start": 334, "end": 342}, {"text": "pose", "start": 67, "end": 71}, {"text": "user motion pattern", "start": 353, "end": 372}, {"text": "extreme illumination changes", "start": 677, "end": 705}, {"text": "smoothness", "start": 733, "end": 743}, {"text": "geodesically local appearance manifold structure", "start": 747, "end": 795}, {"text": "unseen head poses", "start": 859, "end": 876}, {"text": "face motion patterns", "start": 992, "end": 1012}, {"text": "illumination", "start": 685, "end": 697}, {"text": "pose", "start": 344, "end": 348}, {"text": "head motion variation", "start": 1214, "end": 1235}], "metric": [{"text": "resolution", "start": 424, "end": 434}, {"text": "robustness", "start": 978, "end": 988}, {"text": "recognition rate", "start": 1320, "end": 1336}], "method": [{"text": "photometric model", "start": 512, "end": 529}, {"text": "statistical model", "start": 572, "end": 589}, {"text": "robust same-identity likelihood", "start": 802, "end": 833}, {"text": "video sequence'' reillumination'' algorithm", "start": 923, "end": 966}, {"text": "fully automatic recognition system", "start": 1037, "end": 1071}, {"text": "commercial software", "start": 1430, "end": 1449}], "generic": [{"text": "method", "start": 1094, "end": 1100}, {"text": "data set", "start": 1257, "end": 1265}, {"text": "system", "start": 1065, "end": 1071}, {"text": "methods", "start": 1454, "end": 1461}]}, "relations": {"conjunction": [{"head": {"text": "illumination", "start": 50, "end": 62}, "tail": {"text": "pose invariance", "start": 67, "end": 82}}, {"head": {"text": "lighting", "start": 334, "end": 342}, "tail": {"text": "pose", "start": 67, "end": 71}}, {"head": {"text": "pose", "start": 67, "end": 71}, "tail": {"text": "user motion pattern", "start": 353, "end": 372}}, {"head": {"text": "photometric model", "start": 512, "end": 529}, "tail": {"text": "statistical model", "start": 572, "end": 589}}, {"head": {"text": "geodesically local appearance manifold structure", "start": 747, "end": 795}, "tail": {"text": "robust same-identity likelihood", "start": 802, "end": 833}}, {"head": {"text": "illumination", "start": 685, "end": 697}, "tail": {"text": "pose", "start": 344, "end": 348}}, {"head": {"text": "pose", "start": 344, "end": 348}, "tail": {"text": "head motion variation", "start": 1214, "end": 1235}}, {"head": {"text": "commercial software", "start": 1430, "end": 1449}, "tail": {"text": "methods", "start": 1454, "end": 1461}}], "part_of": [{"head": {"text": "illumination", "start": 50, "end": 62}, "tail": {"text": "face recognition", "start": 127, "end": 143}}, {"head": {"text": "pose invariance", "start": 67, "end": 82}, "tail": {"text": "face recognition", "start": 127, "end": 143}}, {"head": {"text": "face motion patterns", "start": 992, "end": 1012}, "tail": {"text": "video", "start": 232, "end": 237}}], "feature_of": [{"head": {"text": "resolution", "start": 424, "end": 434}, "tail": {"text": "face images", "start": 401, "end": 412}}, {"head": {"text": "smoothness", "start": 733, "end": 743}, "tail": {"text": "geodesically local appearance manifold structure", "start": 747, "end": 795}}, {"head": {"text": "face motion patterns", "start": 992, "end": 1012}, "tail": {"text": "robustness", "start": 978, "end": 988}}, {"head": {"text": "illumination", "start": 685, "end": 697}, "tail": {"text": "video sequences", "start": 1162, "end": 1177}}, {"head": {"text": "pose", "start": 344, "end": 348}, "tail": {"text": "video sequences", "start": 1162, "end": 1177}}, {"head": {"text": "head motion variation", "start": 1214, "end": 1235}, "tail": {"text": "video sequences", "start": 1162, "end": 1177}}], "used_for": [{"head": {"text": "photometric model", "start": 512, "end": 529}, "tail": {"text": "image formation", "start": 533, "end": 548}}, {"head": {"text": "statistical model", "start": 572, "end": 589}, "tail": {"text": "generic face appearance variation", "start": 593, "end": 626}}, {"head": {"text": "statistical model", "start": 572, "end": 589}, "tail": {"text": "extreme illumination changes", "start": 677, "end": 705}}, {"head": {"text": "method", "start": 1094, "end": 1100}, "tail": {"text": "fully automatic recognition system", "start": 1037, "end": 1071}}], "evaluate_for": [{"head": {"text": "robustness", "start": 978, "end": 988}, "tail": {"text": "video sequence'' reillumination'' algorithm", "start": 923, "end": 966}}, {"head": {"text": "video sequences", "start": 1162, "end": 1177}, "tail": {"text": "fully automatic recognition system", "start": 1037, "end": 1071}}, {"head": {"text": "data set", "start": 1257, "end": 1265}, "tail": {"text": "system", "start": 1065, "end": 1071}}, {"head": {"text": "recognition rate", "start": 1320, "end": 1336}, "tail": {"text": "system", "start": 1065, "end": 1071}}], "compare": [{"head": {"text": "system", "start": 1065, "end": 1071}, "tail": {"text": "commercial software", "start": 1430, "end": 1449}}, {"head": {"text": "system", "start": 1065, "end": 1071}, "tail": {"text": "methods", "start": 1454, "end": 1461}}]}}, "schema": []}
{"input": "We present Minimum Bayes-Risk -LRB- MBR -RRB- decoding for statistical machine translation. This statistical approach aims to minimize expected loss of translation errors under loss functions that measure translation performance. We describe a hierarchy of loss functions that incorporate different levels of linguistic information from word strings, word-to-word alignments from an MT system, and syntactic structure from parse-trees of source and target language sentences. We report the performance of the MBR decoders on a Chinese-to-English translation task. Our results show that MBR decoding can be used to tune statistical MT performance for specific loss functions.", "output": {"entities": {"method": [{"text": "Minimum Bayes-Risk -LRB- MBR -RRB- decoding", "start": 11, "end": 54}, {"text": "statistical approach", "start": 97, "end": 117}, {"text": "MT system", "start": 383, "end": 392}, {"text": "MBR decoders", "start": 509, "end": 521}, {"text": "MBR decoding", "start": 586, "end": 598}, {"text": "statistical MT", "start": 619, "end": 633}], "task": [{"text": "statistical machine translation", "start": 59, "end": 90}, {"text": "translation", "start": 79, "end": 90}, {"text": "Chinese-to-English translation task", "start": 527, "end": 562}], "other_scientific_term": [{"text": "expected loss", "start": 135, "end": 148}, {"text": "translation errors", "start": 152, "end": 170}, {"text": "loss functions", "start": 177, "end": 191}, {"text": "loss functions", "start": 257, "end": 271}, {"text": "linguistic information", "start": 309, "end": 331}, {"text": "word-to-word alignments", "start": 351, "end": 374}, {"text": "syntactic structure", "start": 398, "end": 417}, {"text": "parse-trees", "start": 423, "end": 434}, {"text": "loss functions", "start": 659, "end": 673}]}, "relations": {"used_for": [{"head": {"text": "Minimum Bayes-Risk -LRB- MBR -RRB- decoding", "start": 11, "end": 54}, "tail": {"text": "statistical machine translation", "start": 59, "end": 90}}, {"head": {"text": "linguistic information", "start": 309, "end": 331}, "tail": {"text": "loss functions", "start": 257, "end": 271}}, {"head": {"text": "word-to-word alignments", "start": 351, "end": 374}, "tail": {"text": "loss functions", "start": 257, "end": 271}}, {"head": {"text": "syntactic structure", "start": 398, "end": 417}, "tail": {"text": "loss functions", "start": 257, "end": 271}}, {"head": {"text": "MBR decoders", "start": 509, "end": 521}, "tail": {"text": "Chinese-to-English translation task", "start": 527, "end": 562}}, {"head": {"text": "MBR decoding", "start": 586, "end": 598}, "tail": {"text": "statistical MT", "start": 619, "end": 633}}, {"head": {"text": "MBR decoding", "start": 586, "end": 598}, "tail": {"text": "loss functions", "start": 659, "end": 673}}], "evaluate_for": [{"head": {"text": "loss functions", "start": 177, "end": 191}, "tail": {"text": "translation", "start": 79, "end": 90}}], "part_of": [{"head": {"text": "word-to-word alignments", "start": 351, "end": 374}, "tail": {"text": "MT system", "start": 383, "end": 392}}, {"head": {"text": "parse-trees", "start": 423, "end": 434}, "tail": {"text": "syntactic structure", "start": 398, "end": 417}}]}}, "schema": []}
{"input": "This paper presents a critical discussion of the various approaches that have been used in the evaluation of Natural Language systems. We conclude that previous approaches have neglected to evaluate systems in the context of their use, e.g. solving a task requiring data retrieval. This raises questions about the validity of such approaches. In the second half of the paper, we report a laboratory study using the Wizard of Oz technique to identify NL requirements for carrying out this task. We evaluate the demands that task dialogues collected using this technique, place upon a prototype Natural Language system. We identify three important requirements which arose from the task that we gave our subjects: operators specific to the task of database access, complex contextual reference and reference to the structure of the information source. We discuss how these might be satisfied by future Natural Language systems.", "output": {"entities": {"generic": [{"text": "approaches", "start": 57, "end": 67}, {"text": "approaches", "start": 161, "end": 171}, {"text": "systems", "start": 126, "end": 133}, {"text": "task", "start": 251, "end": 255}, {"text": "approaches", "start": 331, "end": 341}, {"text": "task", "start": 488, "end": 492}, {"text": "technique", "start": 428, "end": 437}, {"text": "task", "start": 523, "end": 527}, {"text": "task", "start": 680, "end": 684}], "task": [{"text": "evaluation of Natural Language systems", "start": 95, "end": 133}, {"text": "data retrieval", "start": 266, "end": 280}, {"text": "database access", "start": 746, "end": 761}], "method": [{"text": "Natural Language systems", "start": 109, "end": 133}, {"text": "Wizard of Oz technique", "start": 415, "end": 437}, {"text": "prototype Natural Language system", "start": 583, "end": 616}, {"text": "Natural Language systems", "start": 900, "end": 924}], "other_scientific_term": [{"text": "NL requirements", "start": 450, "end": 465}, {"text": "contextual reference", "start": 771, "end": 791}], "material": [{"text": "task dialogues", "start": 523, "end": 537}]}, "relations": {"used_for": [{"head": {"text": "approaches", "start": 57, "end": 67}, "tail": {"text": "evaluation of Natural Language systems", "start": 95, "end": 133}}, {"head": {"text": "systems", "start": 126, "end": 133}, "tail": {"text": "task", "start": 251, "end": 255}}, {"head": {"text": "Wizard of Oz technique", "start": 415, "end": 437}, "tail": {"text": "NL requirements", "start": 450, "end": 465}}, {"head": {"text": "Wizard of Oz technique", "start": 415, "end": 437}, "tail": {"text": "task", "start": 488, "end": 492}}, {"head": {"text": "task dialogues", "start": 523, "end": 537}, "tail": {"text": "prototype Natural Language system", "start": 583, "end": 616}}, {"head": {"text": "technique", "start": 428, "end": 437}, "tail": {"text": "task dialogues", "start": 523, "end": 537}}], "evaluate_for": [{"head": {"text": "approaches", "start": 161, "end": 171}, "tail": {"text": "systems", "start": 126, "end": 133}}], "part_of": [{"head": {"text": "data retrieval", "start": 266, "end": 280}, "tail": {"text": "task", "start": 251, "end": 255}}]}}, "schema": []}
{"input": "We present results on addressee identification in four-participants face-to-face meetings using Bayesian Network and Naive Bayes classifiers. First, we investigate how well the addressee of a dialogue act can be predicted based on gaze, utterance and conversational context features. Then, we explore whether information about meeting context can aid classifiers' performances. Both classifiers perform the best when conversational context and utterance features are combined with speaker's gaze information. The classifiers show little gain from information about meeting context.", "output": {"entities": {"task": [{"text": "addressee identification in four-participants face-to-face meetings", "start": 22, "end": 89}], "method": [{"text": "Bayesian Network", "start": 96, "end": 112}, {"text": "Naive Bayes classifiers", "start": 117, "end": 140}, {"text": "classifiers", "start": 129, "end": 140}, {"text": "classifiers", "start": 351, "end": 362}, {"text": "classifiers", "start": 383, "end": 394}], "other_scientific_term": [{"text": "addressee of a dialogue act", "start": 177, "end": 204}, {"text": "gaze", "start": 231, "end": 235}, {"text": "utterance", "start": 237, "end": 246}, {"text": "conversational context features", "start": 251, "end": 282}, {"text": "meeting context", "start": 327, "end": 342}, {"text": "conversational context", "start": 251, "end": 273}, {"text": "utterance features", "start": 444, "end": 462}, {"text": "speaker's gaze information", "start": 481, "end": 507}, {"text": "meeting context", "start": 565, "end": 580}]}, "relations": {"used_for": [{"head": {"text": "Bayesian Network", "start": 96, "end": 112}, "tail": {"text": "addressee identification in four-participants face-to-face meetings", "start": 22, "end": 89}}, {"head": {"text": "Naive Bayes classifiers", "start": 117, "end": 140}, "tail": {"text": "addressee identification in four-participants face-to-face meetings", "start": 22, "end": 89}}, {"head": {"text": "gaze", "start": 231, "end": 235}, "tail": {"text": "addressee of a dialogue act", "start": 177, "end": 204}}, {"head": {"text": "utterance", "start": 237, "end": 246}, "tail": {"text": "addressee of a dialogue act", "start": 177, "end": 204}}, {"head": {"text": "conversational context features", "start": 251, "end": 282}, "tail": {"text": "addressee of a dialogue act", "start": 177, "end": 204}}, {"head": {"text": "conversational context", "start": 251, "end": 273}, "tail": {"text": "classifiers", "start": 351, "end": 362}}, {"head": {"text": "utterance features", "start": 444, "end": 462}, "tail": {"text": "classifiers", "start": 351, "end": 362}}, {"head": {"text": "speaker's gaze information", "start": 481, "end": 507}, "tail": {"text": "classifiers", "start": 351, "end": 362}}], "conjunction": [{"head": {"text": "Naive Bayes classifiers", "start": 117, "end": 140}, "tail": {"text": "Bayesian Network", "start": 96, "end": 112}}, {"head": {"text": "gaze", "start": 231, "end": 235}, "tail": {"text": "utterance", "start": 237, "end": 246}}, {"head": {"text": "utterance", "start": 237, "end": 246}, "tail": {"text": "conversational context features", "start": 251, "end": 282}}, {"head": {"text": "conversational context", "start": 251, "end": 273}, "tail": {"text": "utterance features", "start": 444, "end": 462}}, {"head": {"text": "speaker's gaze information", "start": 481, "end": 507}, "tail": {"text": "utterance features", "start": 444, "end": 462}}]}}, "schema": []}
{"input": "Towards deep analysis of compositional classes of paraphrases, we have examined a class-oriented framework for collecting paraphrase examples, in which sentential paraphrases are collected for each paraphrase class separately by means of automatic candidate generation and manual judgement. Our preliminary experiments on building a paraphrase corpus have so far been producing promising results, which we have evaluated according to cost-efficiency, exhaustiveness, and reliability.", "output": {"entities": {"task": [{"text": "compositional classes of paraphrases", "start": 25, "end": 61}], "method": [{"text": "class-oriented framework", "start": 82, "end": 106}, {"text": "automatic candidate generation", "start": 238, "end": 268}, {"text": "manual judgement", "start": 273, "end": 289}], "material": [{"text": "paraphrase examples", "start": 122, "end": 141}, {"text": "sentential paraphrases", "start": 152, "end": 174}, {"text": "paraphrase corpus", "start": 333, "end": 350}]}, "relations": {"used_for": [{"head": {"text": "class-oriented framework", "start": 82, "end": 106}, "tail": {"text": "compositional classes of paraphrases", "start": 25, "end": 61}}, {"head": {"text": "class-oriented framework", "start": 82, "end": 106}, "tail": {"text": "paraphrase examples", "start": 122, "end": 141}}, {"head": {"text": "automatic candidate generation", "start": 238, "end": 268}, "tail": {"text": "sentential paraphrases", "start": 152, "end": 174}}, {"head": {"text": "manual judgement", "start": 273, "end": 289}, "tail": {"text": "sentential paraphrases", "start": 152, "end": 174}}], "conjunction": [{"head": {"text": "automatic candidate generation", "start": 238, "end": 268}, "tail": {"text": "manual judgement", "start": 273, "end": 289}}]}}, "schema": []}
{"input": "The purpose of this research is to test the efficacy of applying automated evaluation techniques, originally devised for the evaluation of human language learners, to the output of machine translation -LRB- MT -RRB- systems. We believe that these evaluation techniques will provide information about both the human language learning process, the translation process and the development of machine translation systems. This, the first experiment in a series of experiments, looks at the intelligibility of MT output. A language learning experiment showed that assessors can differentiate native from non-native language essays in less than 100 words. Even more illuminating was the factors on which the assessors made their decisions. We tested this to see if similar criteria could be elicited from duplicating the experiment using machine translation output. Subjects were given a set of up to six extracts of translated newswire text. Some of the extracts were expert human translations, others were machine translation outputs. The subjects were given three minutes per extract to determine whether they believed the sample output to be an expert human translation or a machine translation. Additionally, they were asked to mark the word at which they made this decision. The results of this experiment, along with a preliminary analysis of the factors involved in the decision making process will be presented here.", "output": {"entities": {"method": [{"text": "automated evaluation techniques", "start": 65, "end": 96}, {"text": "machine translation -LRB- MT -RRB- systems", "start": 181, "end": 223}], "task": [{"text": "evaluation of human language learners", "start": 125, "end": 162}, {"text": "human language learning process", "start": 309, "end": 340}, {"text": "translation process", "start": 346, "end": 365}, {"text": "machine translation systems", "start": 389, "end": 416}, {"text": "language learning", "start": 315, "end": 332}], "generic": [{"text": "evaluation techniques", "start": 75, "end": 96}, {"text": "assessors", "start": 559, "end": 568}, {"text": "assessors", "start": 702, "end": 711}], "material": [{"text": "non-native language essays", "start": 599, "end": 625}, {"text": "translated newswire text", "start": 911, "end": 935}], "other_scientific_term": [{"text": "machine translation output", "start": 832, "end": 858}, {"text": "expert human translations", "start": 963, "end": 988}, {"text": "machine translation outputs", "start": 1002, "end": 1029}, {"text": "expert human translation", "start": 963, "end": 987}, {"text": "machine translation", "start": 181, "end": 200}]}, "relations": {"used_for": [{"head": {"text": "automated evaluation techniques", "start": 65, "end": 96}, "tail": {"text": "evaluation of human language learners", "start": 125, "end": 162}}, {"head": {"text": "evaluation techniques", "start": 75, "end": 96}, "tail": {"text": "human language learning process", "start": 309, "end": 340}}, {"head": {"text": "evaluation techniques", "start": 75, "end": 96}, "tail": {"text": "translation process", "start": 346, "end": 365}}, {"head": {"text": "evaluation techniques", "start": 75, "end": 96}, "tail": {"text": "machine translation systems", "start": 389, "end": 416}}], "conjunction": [{"head": {"text": "human language learning process", "start": 309, "end": 340}, "tail": {"text": "translation process", "start": 346, "end": 365}}, {"head": {"text": "translation process", "start": 346, "end": 365}, "tail": {"text": "machine translation systems", "start": 389, "end": 416}}, {"head": {"text": "machine translation outputs", "start": 1002, "end": 1029}, "tail": {"text": "expert human translations", "start": 963, "end": 988}}], "evaluate_for": [{"head": {"text": "language learning", "start": 315, "end": 332}, "tail": {"text": "assessors", "start": 559, "end": 568}}], "compare": [{"head": {"text": "expert human translation", "start": 963, "end": 987}, "tail": {"text": "machine translation", "start": 181, "end": 200}}]}}, "schema": []}
{"input": "This paper presents a machine learning approach to bare slice disambiguation in dialogue. We extract a set of heuristic principles from a corpus-based sample and formulate them as probabilistic Horn clauses. We then use the predicates of such clauses to create a set of domain independent features to annotate an input dataset, and run two different machine learning algorithms: SLIPPER, a rule-based learning algorithm, and TiMBL, a memory-based system. Both learners perform well, yielding similar success rates of approx 90%. The results show that the features in terms of which we formulate our heuristic principles have significant predictive power, and that rules that closely resemble our Horn clauses can be learnt automatically from these features.", "output": {"entities": {"method": [{"text": "machine learning approach", "start": 22, "end": 47}, {"text": "heuristic principles", "start": 110, "end": 130}, {"text": "machine learning algorithms", "start": 350, "end": 377}, {"text": "rule-based learning algorithm", "start": 390, "end": 419}, {"text": "memory-based system", "start": 434, "end": 453}, {"text": "heuristic principles", "start": 599, "end": 619}], "task": [{"text": "bare slice disambiguation", "start": 51, "end": 76}], "material": [{"text": "dialogue", "start": 80, "end": 88}, {"text": "corpus-based sample", "start": 138, "end": 157}], "other_scientific_term": [{"text": "probabilistic Horn clauses", "start": 180, "end": 206}, {"text": "clauses", "start": 199, "end": 206}, {"text": "domain independent features", "start": 270, "end": 297}, {"text": "features", "start": 289, "end": 297}, {"text": "rules", "start": 664, "end": 669}, {"text": "Horn clauses", "start": 194, "end": 206}, {"text": "features", "start": 555, "end": 563}], "generic": [{"text": "SLIPPER", "start": 379, "end": 386}, {"text": "TiMBL", "start": 425, "end": 430}], "metric": [{"text": "success rates", "start": 500, "end": 513}]}, "relations": {"used_for": [{"head": {"text": "machine learning approach", "start": 22, "end": 47}, "tail": {"text": "bare slice disambiguation", "start": 51, "end": 76}}, {"head": {"text": "dialogue", "start": 80, "end": 88}, "tail": {"text": "bare slice disambiguation", "start": 51, "end": 76}}, {"head": {"text": "corpus-based sample", "start": 138, "end": 157}, "tail": {"text": "heuristic principles", "start": 110, "end": 130}}], "feature_of": [{"head": {"text": "probabilistic Horn clauses", "start": 180, "end": 206}, "tail": {"text": "heuristic principles", "start": 110, "end": 130}}, {"head": {"text": "features", "start": 289, "end": 297}, "tail": {"text": "heuristic principles", "start": 599, "end": 619}}], "hyponym_of": [{"head": {"text": "SLIPPER", "start": 379, "end": 386}, "tail": {"text": "rule-based learning algorithm", "start": 390, "end": 419}}, {"head": {"text": "TiMBL", "start": 425, "end": 430}, "tail": {"text": "memory-based system", "start": 434, "end": 453}}], "part_of": [{"head": {"text": "rule-based learning algorithm", "start": 390, "end": 419}, "tail": {"text": "machine learning algorithms", "start": 350, "end": 377}}, {"head": {"text": "memory-based system", "start": 434, "end": 453}, "tail": {"text": "machine learning algorithms", "start": 350, "end": 377}}], "compare": [{"head": {"text": "rule-based learning algorithm", "start": 390, "end": 419}, "tail": {"text": "memory-based system", "start": 434, "end": 453}}]}}, "schema": []}
{"input": "We suggest a new goal and evaluation criterion for word similarity measures. The new criterion -- meaning-entailing substitutability -- fits the needs of semantic-oriented NLP applications and can be evaluated directly -LRB- independent of an application -RRB- at a good level of human agreement. Motivated by this semantic criterion we analyze the empirical quality of distributional word feature vectors and its impact on word similarity results, proposing an objective measure for evaluating feature vector quality. Finally, a novel feature weighting and selection function is presented, which yields superior feature vectors and better word similarity performance.", "output": {"entities": {"metric": [{"text": "evaluation criterion", "start": 26, "end": 46}, {"text": "word similarity measures", "start": 51, "end": 75}, {"text": "meaning-entailing substitutability", "start": 98, "end": 132}, {"text": "human agreement", "start": 280, "end": 295}, {"text": "semantic criterion", "start": 315, "end": 333}, {"text": "feature vector quality", "start": 495, "end": 517}], "generic": [{"text": "criterion", "start": 37, "end": 46}, {"text": "measure", "start": 67, "end": 74}], "task": [{"text": "semantic-oriented NLP applications", "start": 154, "end": 188}, {"text": "word similarity", "start": 51, "end": 66}, {"text": "word similarity", "start": 424, "end": 439}], "other_scientific_term": [{"text": "distributional word feature vectors", "start": 370, "end": 405}, {"text": "feature vectors", "start": 390, "end": 405}], "method": [{"text": "feature weighting and selection function", "start": 536, "end": 576}]}, "relations": {"used_for": [{"head": {"text": "evaluation criterion", "start": 26, "end": 46}, "tail": {"text": "word similarity measures", "start": 51, "end": 75}}, {"head": {"text": "meaning-entailing substitutability", "start": 98, "end": 132}, "tail": {"text": "semantic-oriented NLP applications", "start": 154, "end": 188}}, {"head": {"text": "distributional word feature vectors", "start": 370, "end": 405}, "tail": {"text": "word similarity", "start": 51, "end": 66}}, {"head": {"text": "feature weighting and selection function", "start": 536, "end": 576}, "tail": {"text": "feature vectors", "start": 390, "end": 405}}, {"head": {"text": "feature weighting and selection function", "start": 536, "end": 576}, "tail": {"text": "word similarity", "start": 424, "end": 439}}], "evaluate_for": [{"head": {"text": "human agreement", "start": 280, "end": 295}, "tail": {"text": "meaning-entailing substitutability", "start": 98, "end": 132}}, {"head": {"text": "semantic criterion", "start": 315, "end": 333}, "tail": {"text": "distributional word feature vectors", "start": 370, "end": 405}}, {"head": {"text": "measure", "start": 67, "end": 74}, "tail": {"text": "feature vector quality", "start": 495, "end": 517}}], "conjunction": [{"head": {"text": "feature vectors", "start": 390, "end": 405}, "tail": {"text": "word similarity", "start": 424, "end": 439}}]}}, "schema": []}
{"input": "Reflections in image sequences consist of several layers superimposed over each other. This phenomenon causes many image processing techniques to fail as they assume the presence of only one layer at each examined site e.g. motion estimation and object recognition. This work presents an automated technique for detecting reflections in image sequences by analyzing motion trajectories of feature points. It models reflection as regions containing two different layers moving over each other. We present a strong detector based on combining a set of weak detectors. We use novel priors, generate sparse and dense detection maps and our results show high detection rate with rejection to pathological motion and occlusion.", "output": {"entities": {"other_scientific_term": [{"text": "Reflections", "start": 0, "end": 11}, {"text": "Reflections in image sequences", "start": 0, "end": 30}, {"text": "motion trajectories", "start": 366, "end": 385}, {"text": "feature points", "start": 389, "end": 403}, {"text": "reflection", "start": 322, "end": 332}, {"text": "priors", "start": 579, "end": 585}, {"text": "sparse and dense detection maps", "start": 596, "end": 627}, {"text": "pathological motion", "start": 687, "end": 706}, {"text": "occlusion", "start": 711, "end": 720}], "material": [{"text": "image sequences", "start": 15, "end": 30}], "method": [{"text": "image processing techniques", "start": 115, "end": 142}], "task": [{"text": "motion estimation", "start": 224, "end": 241}, {"text": "object recognition", "start": 246, "end": 264}, {"text": "detecting reflections in image sequences", "start": 312, "end": 352}], "generic": [{"text": "technique", "start": 132, "end": 141}, {"text": "It", "start": 405, "end": 407}, {"text": "detector", "start": 513, "end": 521}, {"text": "detectors", "start": 555, "end": 564}], "metric": [{"text": "detection rate", "start": 654, "end": 668}]}, "relations": {"conjunction": [{"head": {"text": "motion estimation", "start": 224, "end": 241}, "tail": {"text": "object recognition", "start": 246, "end": 264}}, {"head": {"text": "pathological motion", "start": 687, "end": 706}, "tail": {"text": "occlusion", "start": 711, "end": 720}}], "used_for": [{"head": {"text": "technique", "start": 132, "end": 141}, "tail": {"text": "detecting reflections in image sequences", "start": 312, "end": 352}}, {"head": {"text": "motion trajectories", "start": 366, "end": 385}, "tail": {"text": "technique", "start": 132, "end": 141}}, {"head": {"text": "It", "start": 405, "end": 407}, "tail": {"text": "reflection", "start": 322, "end": 332}}, {"head": {"text": "detectors", "start": 555, "end": 564}, "tail": {"text": "detector", "start": 513, "end": 521}}, {"head": {"text": "priors", "start": 579, "end": 585}, "tail": {"text": "sparse and dense detection maps", "start": 596, "end": 627}}], "feature_of": [{"head": {"text": "feature points", "start": 389, "end": 403}, "tail": {"text": "motion trajectories", "start": 366, "end": 385}}]}}, "schema": []}
{"input": "This paper considers the problem of reconstructing the motion of a 3D articulated tree from 2D point correspondences subject to some temporal prior. Hitherto, smooth motion has been encouraged using a trajectory basis, yielding a hard combinatorial problem with time complexity growing exponentially in the number of frames. Branch and bound strategies have previously attempted to curb this complexity whilst maintaining global optimality. However, they provide no guarantee of being more efficient than exhaustive search. Inspired by recent work which reconstructs general trajectories using compact high-pass filters, we develop a dynamic programming approach which scales linearly in the number of frames, leveraging the intrinsically local nature of filter interactions. Extension to affine projection enables reconstruction without estimating cameras.", "output": {"entities": {"task": [{"text": "reconstructing the motion of a 3D articulated tree", "start": 36, "end": 86}, {"text": "hard combinatorial problem", "start": 230, "end": 256}, {"text": "reconstruction", "start": 815, "end": 829}], "other_scientific_term": [{"text": "2D point correspondences", "start": 92, "end": 116}, {"text": "temporal prior", "start": 133, "end": 147}, {"text": "smooth motion", "start": 159, "end": 172}, {"text": "trajectory basis", "start": 201, "end": 217}, {"text": "global optimality", "start": 422, "end": 439}, {"text": "filter interactions", "start": 755, "end": 774}], "metric": [{"text": "time complexity", "start": 262, "end": 277}], "method": [{"text": "Branch and bound strategies", "start": 325, "end": 352}, {"text": "exhaustive search", "start": 505, "end": 522}, {"text": "compact high-pass filters", "start": 594, "end": 619}, {"text": "dynamic programming approach", "start": 634, "end": 662}, {"text": "affine projection", "start": 789, "end": 806}, {"text": "estimating cameras", "start": 838, "end": 856}], "generic": [{"text": "complexity", "start": 267, "end": 277}, {"text": "they", "start": 450, "end": 454}]}, "relations": {"used_for": [{"head": {"text": "2D point correspondences", "start": 92, "end": 116}, "tail": {"text": "reconstructing the motion of a 3D articulated tree", "start": 36, "end": 86}}, {"head": {"text": "trajectory basis", "start": 201, "end": 217}, "tail": {"text": "smooth motion", "start": 159, "end": 172}}, {"head": {"text": "Branch and bound strategies", "start": 325, "end": 352}, "tail": {"text": "complexity", "start": 267, "end": 277}}, {"head": {"text": "affine projection", "start": 789, "end": 806}, "tail": {"text": "reconstruction", "start": 815, "end": 829}}], "evaluate_for": [{"head": {"text": "time complexity", "start": 262, "end": 277}, "tail": {"text": "hard combinatorial problem", "start": 230, "end": 256}}], "feature_of": [{"head": {"text": "global optimality", "start": 422, "end": 439}, "tail": {"text": "Branch and bound strategies", "start": 325, "end": 352}}], "compare": [{"head": {"text": "they", "start": 450, "end": 454}, "tail": {"text": "exhaustive search", "start": 505, "end": 522}}]}}, "schema": []}
{"input": "Topical blog post retrieval is the task of ranking blog posts with respect to their relevance for a given topic. To improve topical blog post retrieval we incorporate textual credibility indicators in the retrieval process. We consider two groups of indicators: post level -LRB- determined using information about individual blog posts only -RRB- and blog level -LRB- determined using information from the underlying blogs -RRB-. We describe how to estimate these indicators and how to integrate them into a retrieval approach based on language models. Experiments on the TREC Blog track test set show that both groups of credibility indicators significantly improve retrieval effectiveness; the best performance is achieved when combining them.", "output": {"entities": {"task": [{"text": "Topical blog post retrieval", "start": 0, "end": 27}, {"text": "ranking blog posts", "start": 43, "end": 61}, {"text": "topical blog post retrieval", "start": 124, "end": 151}], "material": [{"text": "blog posts", "start": 51, "end": 61}, {"text": "blog posts", "start": 325, "end": 335}, {"text": "blogs", "start": 417, "end": 422}, {"text": "TREC Blog track test set", "start": 572, "end": 596}], "metric": [{"text": "relevance", "start": 84, "end": 93}, {"text": "retrieval effectiveness", "start": 667, "end": 690}], "other_scientific_term": [{"text": "textual credibility indicators", "start": 167, "end": 197}, {"text": "credibility indicators", "start": 175, "end": 197}], "method": [{"text": "retrieval process", "start": 205, "end": 222}, {"text": "retrieval approach", "start": 508, "end": 526}, {"text": "language models", "start": 536, "end": 551}], "generic": [{"text": "indicators", "start": 187, "end": 197}, {"text": "indicators", "start": 250, "end": 260}, {"text": "them", "start": 496, "end": 500}]}, "relations": {"hyponym_of": [{"head": {"text": "Topical blog post retrieval", "start": 0, "end": 27}, "tail": {"text": "ranking blog posts", "start": 43, "end": 61}}], "feature_of": [{"head": {"text": "relevance", "start": 84, "end": 93}, "tail": {"text": "blog posts", "start": 51, "end": 61}}], "used_for": [{"head": {"text": "textual credibility indicators", "start": 167, "end": 197}, "tail": {"text": "topical blog post retrieval", "start": 124, "end": 151}}, {"head": {"text": "language models", "start": 536, "end": 551}, "tail": {"text": "them", "start": 496, "end": 500}}], "part_of": [{"head": {"text": "textual credibility indicators", "start": 167, "end": 197}, "tail": {"text": "retrieval process", "start": 205, "end": 222}}, {"head": {"text": "them", "start": 496, "end": 500}, "tail": {"text": "retrieval approach", "start": 508, "end": 526}}], "evaluate_for": [{"head": {"text": "TREC Blog track test set", "start": 572, "end": 596}, "tail": {"text": "credibility indicators", "start": 175, "end": 197}}, {"head": {"text": "retrieval effectiveness", "start": 667, "end": 690}, "tail": {"text": "credibility indicators", "start": 175, "end": 197}}]}}, "schema": []}
{"input": "We investigate the problem of learning to predict moves in the board game of Go from game records of expert players. In particular, we obtain a probability distribution over legal moves for professional play in a given position. This distribution has numerous applications in computer Go, including serving as an efficient stand-alone Go player. It would also be effective as a move selector and move sorter for game tree search and as a training tool for Go players. Our method has two major components: a -RRB- a pattern extraction scheme for efficiently harvesting patterns of given size and shape from expert game records and b -RRB- a Bayesian learning algorithm -LRB- in two variants -RRB- that learns a distribution over the values of a move given a board position based on the local pattern context. The system is trained on 181,000 expert games and shows excellent prediction performance as indicated by its ability to perfectly predict the moves made by professional Go players in 34% of test positions.", "output": {"entities": {"task": [{"text": "board game of Go", "start": 63, "end": 79}, {"text": "training tool", "start": 438, "end": 451}], "material": [{"text": "game records of expert players", "start": 85, "end": 115}, {"text": "expert games", "start": 841, "end": 853}], "other_scientific_term": [{"text": "probability distribution", "start": 144, "end": 168}, {"text": "computer Go", "start": 276, "end": 287}, {"text": "stand-alone Go player", "start": 323, "end": 344}, {"text": "Go players", "start": 456, "end": 466}, {"text": "local pattern context", "start": 785, "end": 806}], "generic": [{"text": "distribution", "start": 156, "end": 168}, {"text": "It", "start": 346, "end": 348}, {"text": "method", "start": 472, "end": 478}, {"text": "system", "start": 812, "end": 818}], "method": [{"text": "move selector", "start": 378, "end": 391}, {"text": "move sorter", "start": 396, "end": 407}, {"text": "game tree search", "start": 412, "end": 428}, {"text": "pattern extraction scheme", "start": 515, "end": 540}, {"text": "Bayesian learning algorithm", "start": 640, "end": 667}]}, "relations": {"used_for": [{"head": {"text": "game records of expert players", "start": 85, "end": 115}, "tail": {"text": "board game of Go", "start": 63, "end": 79}}, {"head": {"text": "distribution", "start": 156, "end": 168}, "tail": {"text": "computer Go", "start": 276, "end": 287}}, {"head": {"text": "It", "start": 346, "end": 348}, "tail": {"text": "move selector", "start": 378, "end": 391}}, {"head": {"text": "It", "start": 346, "end": 348}, "tail": {"text": "move sorter", "start": 396, "end": 407}}, {"head": {"text": "It", "start": 346, "end": 348}, "tail": {"text": "training tool", "start": 438, "end": 451}}, {"head": {"text": "move selector", "start": 378, "end": 391}, "tail": {"text": "game tree search", "start": 412, "end": 428}}, {"head": {"text": "move sorter", "start": 396, "end": 407}, "tail": {"text": "game tree search", "start": 412, "end": 428}}, {"head": {"text": "training tool", "start": 438, "end": 451}, "tail": {"text": "Go players", "start": 456, "end": 466}}, {"head": {"text": "expert games", "start": 841, "end": 853}, "tail": {"text": "system", "start": 812, "end": 818}}], "conjunction": [{"head": {"text": "move selector", "start": 378, "end": 391}, "tail": {"text": "move sorter", "start": 396, "end": 407}}, {"head": {"text": "pattern extraction scheme", "start": 515, "end": 540}, "tail": {"text": "Bayesian learning algorithm", "start": 640, "end": 667}}], "part_of": [{"head": {"text": "pattern extraction scheme", "start": 515, "end": 540}, "tail": {"text": "method", "start": 472, "end": 478}}, {"head": {"text": "Bayesian learning algorithm", "start": 640, "end": 667}, "tail": {"text": "method", "start": 472, "end": 478}}]}}, "schema": []}
{"input": "We present a novel approach for automatically acquiring English topic signatures. Given a particular concept, or word sense, a topic signature is a set of words that tend to co-occur with it. Topic signatures can be useful in a number of Natural Language Processing -LRB- NLP -RRB- applications, such as Word Sense Disambiguation -LRB- WSD -RRB- and Text Summarisation. Our method takes advantage of the different way in which word senses are lexicalised in English and Chinese, and also exploits the large amount of Chinese text available in corpora and on the Web. We evaluated the topic signatures on a WSD task, where we trained a second-order vector cooccurrence algorithm on standard WSD datasets, with promising results.", "output": {"entities": {"generic": [{"text": "approach", "start": 19, "end": 27}, {"text": "method", "start": 374, "end": 380}, {"text": "corpora", "start": 543, "end": 550}], "task": [{"text": "automatically acquiring English topic signatures", "start": 32, "end": 80}, {"text": "Natural Language Processing -LRB- NLP -RRB- applications", "start": 238, "end": 294}, {"text": "Word Sense Disambiguation -LRB- WSD -RRB-", "start": 304, "end": 345}, {"text": "Text Summarisation", "start": 350, "end": 368}, {"text": "WSD task", "start": 606, "end": 614}], "other_scientific_term": [{"text": "concept", "start": 101, "end": 108}, {"text": "word sense", "start": 113, "end": 123}, {"text": "topic signature", "start": 64, "end": 79}, {"text": "Topic signatures", "start": 192, "end": 208}, {"text": "word senses", "start": 427, "end": 438}, {"text": "topic signatures", "start": 64, "end": 80}], "material": [{"text": "English", "start": 56, "end": 63}, {"text": "Chinese", "start": 470, "end": 477}, {"text": "Chinese text", "start": 517, "end": 529}, {"text": "Web", "start": 562, "end": 565}, {"text": "WSD datasets", "start": 690, "end": 702}], "method": [{"text": "second-order vector cooccurrence algorithm", "start": 635, "end": 677}]}, "relations": {"used_for": [{"head": {"text": "approach", "start": 19, "end": 27}, "tail": {"text": "automatically acquiring English topic signatures", "start": 32, "end": 80}}, {"head": {"text": "Topic signatures", "start": 192, "end": 208}, "tail": {"text": "Natural Language Processing -LRB- NLP -RRB- applications", "start": 238, "end": 294}}, {"head": {"text": "Topic signatures", "start": 192, "end": 208}, "tail": {"text": "Word Sense Disambiguation -LRB- WSD -RRB-", "start": 304, "end": 345}}, {"head": {"text": "Topic signatures", "start": 192, "end": 208}, "tail": {"text": "Text Summarisation", "start": 350, "end": 368}}, {"head": {"text": "WSD datasets", "start": 690, "end": 702}, "tail": {"text": "second-order vector cooccurrence algorithm", "start": 635, "end": 677}}], "hyponym_of": [{"head": {"text": "Word Sense Disambiguation -LRB- WSD -RRB-", "start": 304, "end": 345}, "tail": {"text": "Natural Language Processing -LRB- NLP -RRB- applications", "start": 238, "end": 294}}, {"head": {"text": "Text Summarisation", "start": 350, "end": 368}, "tail": {"text": "Natural Language Processing -LRB- NLP -RRB- applications", "start": 238, "end": 294}}], "conjunction": [{"head": {"text": "Word Sense Disambiguation -LRB- WSD -RRB-", "start": 304, "end": 345}, "tail": {"text": "Text Summarisation", "start": 350, "end": 368}}, {"head": {"text": "corpora", "start": 543, "end": 550}, "tail": {"text": "Web", "start": 562, "end": 565}}], "part_of": [{"head": {"text": "Chinese text", "start": 517, "end": 529}, "tail": {"text": "corpora", "start": 543, "end": 550}}, {"head": {"text": "Chinese text", "start": 517, "end": 529}, "tail": {"text": "Web", "start": 562, "end": 565}}], "evaluate_for": [{"head": {"text": "WSD task", "start": 606, "end": 614}, "tail": {"text": "topic signatures", "start": 64, "end": 80}}]}}, "schema": []}
{"input": "Joint matrix triangularization is often used for estimating the joint eigenstructure of a set M of matrices, with applications in signal processing and machine learning. We consider the problem of approximate joint matrix triangularization when the matrices in M are jointly diagonalizable and real, but we only observe a set M' of noise perturbed versions of the matrices in M. Our main result is a first-order upper bound on the distance between any approximate joint triangularizer of the matrices in M' and any exact joint triangularizer of the matrices in M. The bound depends only on the observable matrices in M' and the noise level. In particular, it does not depend on optimization specific properties of the triangularizer, such as its proximity to critical points, that are typical of existing bounds in the literature. To our knowledge, this is the first a posteriori bound for joint matrix decomposition. We demonstrate the bound on synthetic data for which the ground truth is known.", "output": {"entities": {"task": [{"text": "Joint matrix triangularization", "start": 0, "end": 30}, {"text": "signal processing", "start": 130, "end": 147}, {"text": "machine learning", "start": 152, "end": 168}, {"text": "approximate joint matrix triangularization", "start": 197, "end": 239}, {"text": "joint matrix decomposition", "start": 890, "end": 916}], "other_scientific_term": [{"text": "joint eigenstructure", "start": 64, "end": 84}, {"text": "first-order upper bound", "start": 400, "end": 423}, {"text": "posteriori bound", "start": 869, "end": 885}], "method": [{"text": "approximate joint triangularizer", "start": 452, "end": 484}, {"text": "exact joint triangularizer", "start": 515, "end": 541}, {"text": "triangularizer", "start": 470, "end": 484}], "generic": [{"text": "bound", "start": 418, "end": 423}], "material": [{"text": "synthetic data", "start": 946, "end": 960}]}, "relations": {"used_for": [{"head": {"text": "Joint matrix triangularization", "start": 0, "end": 30}, "tail": {"text": "joint eigenstructure", "start": 64, "end": 84}}, {"head": {"text": "joint eigenstructure", "start": 64, "end": 84}, "tail": {"text": "signal processing", "start": 130, "end": 147}}, {"head": {"text": "joint eigenstructure", "start": 64, "end": 84}, "tail": {"text": "machine learning", "start": 152, "end": 168}}, {"head": {"text": "posteriori bound", "start": 869, "end": 885}, "tail": {"text": "joint matrix decomposition", "start": 890, "end": 916}}], "conjunction": [{"head": {"text": "signal processing", "start": 130, "end": 147}, "tail": {"text": "machine learning", "start": 152, "end": 168}}, {"head": {"text": "approximate joint triangularizer", "start": 452, "end": 484}, "tail": {"text": "exact joint triangularizer", "start": 515, "end": 541}}]}}, "schema": []}
{"input": "The psycholinguistic literature provides evidence for syntactic priming, i.e., the tendency to repeat structures. This paper describes a method for incorporating priming into an incremental probabilistic parser. Three models are compared, which involve priming of rules between sentences, within sentences, and within coordinate structures. These models simulate the reading time advantage for parallel structures found in human data, and also yield a small increase in overall parsing accuracy.", "output": {"entities": {"other_scientific_term": [{"text": "psycholinguistic literature", "start": 4, "end": 31}, {"text": "syntactic priming", "start": 54, "end": 71}, {"text": "priming", "start": 64, "end": 71}, {"text": "priming", "start": 162, "end": 169}, {"text": "rules", "start": 264, "end": 269}, {"text": "coordinate structures", "start": 318, "end": 339}, {"text": "parallel structures", "start": 394, "end": 413}], "generic": [{"text": "method", "start": 137, "end": 143}], "method": [{"text": "incremental probabilistic parser", "start": 178, "end": 210}], "material": [{"text": "human data", "start": 423, "end": 433}], "metric": [{"text": "parsing accuracy", "start": 478, "end": 494}]}, "relations": {"used_for": [{"head": {"text": "psycholinguistic literature", "start": 4, "end": 31}, "tail": {"text": "syntactic priming", "start": 54, "end": 71}}, {"head": {"text": "method", "start": 137, "end": 143}, "tail": {"text": "priming", "start": 64, "end": 71}}, {"head": {"text": "priming", "start": 64, "end": 71}, "tail": {"text": "incremental probabilistic parser", "start": 178, "end": 210}}], "part_of": [{"head": {"text": "parallel structures", "start": 394, "end": 413}, "tail": {"text": "human data", "start": 423, "end": 433}}]}}, "schema": []}
{"input": "Learned confidence measures gain increasing importance for outlier removal and quality improvement in stereo vision. However, acquiring the necessary training data is typically a tedious and time consuming task that involves manual interaction, active sensing devices and/or synthetic scenes. To overcome this problem, we propose a new, flexible, and scalable way for generating training data that only requires a set of stereo images as input. The key idea of our approach is to use different view points for reasoning about contradictions and consistencies between multiple depth maps generated with the same stereo algorithm. This enables us to generate a huge amount of training data in a fully automated manner. Among other experiments, we demonstrate the potential of our approach by boosting the performance of three learned confidence measures on the KITTI2012 dataset by simply training them on a vast amount of automatically generated training data rather than a limited amount of laser ground truth data.", "output": {"entities": {"method": [{"text": "Learned confidence measures", "start": 0, "end": 27}, {"text": "stereo algorithm", "start": 611, "end": 627}, {"text": "learned confidence measures", "start": 824, "end": 851}], "task": [{"text": "outlier removal", "start": 59, "end": 74}, {"text": "quality improvement", "start": 79, "end": 98}, {"text": "stereo vision", "start": 102, "end": 115}], "generic": [{"text": "task", "start": 206, "end": 210}, {"text": "problem", "start": 310, "end": 317}, {"text": "approach", "start": 465, "end": 473}, {"text": "approach", "start": 778, "end": 786}, {"text": "them", "start": 896, "end": 900}], "other_scientific_term": [{"text": "manual interaction", "start": 225, "end": 243}, {"text": "active sensing devices", "start": 245, "end": 267}, {"text": "synthetic scenes", "start": 275, "end": 291}, {"text": "view points", "start": 494, "end": 505}, {"text": "multiple depth maps", "start": 567, "end": 586}], "material": [{"text": "stereo images", "start": 421, "end": 434}, {"text": "KITTI2012 dataset", "start": 859, "end": 876}, {"text": "automatically generated training data", "start": 921, "end": 958}, {"text": "laser ground truth data", "start": 991, "end": 1014}]}, "relations": {"used_for": [{"head": {"text": "Learned confidence measures", "start": 0, "end": 27}, "tail": {"text": "outlier removal", "start": 59, "end": 74}}, {"head": {"text": "Learned confidence measures", "start": 0, "end": 27}, "tail": {"text": "quality improvement", "start": 79, "end": 98}}, {"head": {"text": "manual interaction", "start": 225, "end": 243}, "tail": {"text": "task", "start": 206, "end": 210}}, {"head": {"text": "active sensing devices", "start": 245, "end": 267}, "tail": {"text": "task", "start": 206, "end": 210}}, {"head": {"text": "synthetic scenes", "start": 275, "end": 291}, "tail": {"text": "task", "start": 206, "end": 210}}, {"head": {"text": "view points", "start": 494, "end": 505}, "tail": {"text": "approach", "start": 465, "end": 473}}, {"head": {"text": "approach", "start": 778, "end": 786}, "tail": {"text": "learned confidence measures", "start": 824, "end": 851}}, {"head": {"text": "automatically generated training data", "start": 921, "end": 958}, "tail": {"text": "them", "start": 896, "end": 900}}], "conjunction": [{"head": {"text": "outlier removal", "start": 59, "end": 74}, "tail": {"text": "quality improvement", "start": 79, "end": 98}}, {"head": {"text": "manual interaction", "start": 225, "end": 243}, "tail": {"text": "active sensing devices", "start": 245, "end": 267}}, {"head": {"text": "active sensing devices", "start": 245, "end": 267}, "tail": {"text": "synthetic scenes", "start": 275, "end": 291}}], "part_of": [{"head": {"text": "outlier removal", "start": 59, "end": 74}, "tail": {"text": "stereo vision", "start": 102, "end": 115}}, {"head": {"text": "quality improvement", "start": 79, "end": 98}, "tail": {"text": "stereo vision", "start": 102, "end": 115}}], "evaluate_for": [{"head": {"text": "KITTI2012 dataset", "start": 859, "end": 876}, "tail": {"text": "learned confidence measures", "start": 824, "end": 851}}], "compare": [{"head": {"text": "laser ground truth data", "start": 991, "end": 1014}, "tail": {"text": "automatically generated training data", "start": 921, "end": 958}}]}}, "schema": []}
{"input": "An important area of learning in autonomous agents is the ability to learn domain-speciic models of actions to be used by planning systems. In this paper, we present methods by which an agent learns action models from its own experience and from its observation of a domain expert. These methods diier from previous work in the area in two ways: the use of an action model formalism which is better suited to the needs of a re-active agent, and successful implementation of noise-handling mechanisms. Training instances are generated from experience and observation, and a variant of GOLEM is used to learn action models from these instances. The integrated learning system has been experimentally validated in simulated construction and ooce domains.", "output": {"entities": {"task": [{"text": "learning in autonomous agents", "start": 21, "end": 50}, {"text": "planning systems", "start": 122, "end": 138}, {"text": "simulated construction", "start": 711, "end": 733}, {"text": "ooce domains", "start": 738, "end": 750}], "method": [{"text": "domain-speciic models of actions", "start": 75, "end": 107}, {"text": "action models", "start": 199, "end": 212}, {"text": "action model formalism", "start": 360, "end": 382}, {"text": "re-active agent", "start": 424, "end": 439}, {"text": "noise-handling mechanisms", "start": 474, "end": 499}, {"text": "GOLEM", "start": 584, "end": 589}, {"text": "action models", "start": 607, "end": 620}, {"text": "integrated learning system", "start": 647, "end": 673}], "generic": [{"text": "methods", "start": 166, "end": 173}], "other_scientific_term": [{"text": "domain expert", "start": 267, "end": 280}], "metric": [{"text": "methods", "start": 288, "end": 295}]}, "relations": {"used_for": [{"head": {"text": "learning in autonomous agents", "start": 21, "end": 50}, "tail": {"text": "domain-speciic models of actions", "start": 75, "end": 107}}, {"head": {"text": "planning systems", "start": 122, "end": 138}, "tail": {"text": "domain-speciic models of actions", "start": 75, "end": 107}}, {"head": {"text": "action model formalism", "start": 360, "end": 382}, "tail": {"text": "methods", "start": 288, "end": 295}}, {"head": {"text": "action model formalism", "start": 360, "end": 382}, "tail": {"text": "re-active agent", "start": 424, "end": 439}}, {"head": {"text": "noise-handling mechanisms", "start": 474, "end": 499}, "tail": {"text": "methods", "start": 288, "end": 295}}, {"head": {"text": "GOLEM", "start": 584, "end": 589}, "tail": {"text": "action models", "start": 607, "end": 620}}], "evaluate_for": [{"head": {"text": "simulated construction", "start": 711, "end": 733}, "tail": {"text": "integrated learning system", "start": 647, "end": 673}}, {"head": {"text": "ooce domains", "start": 738, "end": 750}, "tail": {"text": "integrated learning system", "start": 647, "end": 673}}], "conjunction": [{"head": {"text": "simulated construction", "start": 711, "end": 733}, "tail": {"text": "ooce domains", "start": 738, "end": 750}}]}}, "schema": []}
{"input": "This paper describes FERRET, an interactive question-answering -LRB- Q/A -RRB- system designed to address the challenges of integrating automatic Q/A applications into real-world environments. FERRET utilizes a novel approach to Q/A known as predictive questioning which attempts to identify the questions -LRB- and answers -RRB- that users need by analyzing how a user interacts with a system while gathering information related to a particular scenario.", "output": {"entities": {"method": [{"text": "FERRET", "start": 21, "end": 27}, {"text": "interactive question-answering -LRB- Q/A -RRB- system", "start": 32, "end": 85}, {"text": "FERRET", "start": 193, "end": 199}, {"text": "Q/A", "start": 69, "end": 72}, {"text": "predictive questioning", "start": 242, "end": 264}], "task": [{"text": "integrating automatic Q/A applications into real-world environments", "start": 124, "end": 191}], "generic": [{"text": "approach", "start": 217, "end": 225}]}, "relations": {"hyponym_of": [{"head": {"text": "FERRET", "start": 21, "end": 27}, "tail": {"text": "interactive question-answering -LRB- Q/A -RRB- system", "start": 32, "end": 85}}], "used_for": [{"head": {"text": "FERRET", "start": 21, "end": 27}, "tail": {"text": "integrating automatic Q/A applications into real-world environments", "start": 124, "end": 191}}, {"head": {"text": "approach", "start": 217, "end": 225}, "tail": {"text": "FERRET", "start": 193, "end": 199}}, {"head": {"text": "approach", "start": 217, "end": 225}, "tail": {"text": "Q/A", "start": 69, "end": 72}}]}}, "schema": []}
{"input": "In order to build robust automatic abstracting systems, there is a need for better training resources than are currently available. In this paper, we introduce an annotation scheme for scientific articles which can be used to build such a resource in a consistent way. The seven categories of the scheme are based on rhetorical moves of argumentation. Our experimental results show that the scheme is stable, reproducible and intuitive to use.", "output": {"entities": {"task": [{"text": "automatic abstracting systems", "start": 25, "end": 54}], "material": [{"text": "training resources", "start": 83, "end": 101}, {"text": "scientific articles", "start": 185, "end": 204}], "method": [{"text": "annotation scheme", "start": 163, "end": 180}, {"text": "rhetorical moves of argumentation", "start": 317, "end": 350}], "generic": [{"text": "resource", "start": 92, "end": 100}, {"text": "scheme", "start": 174, "end": 180}, {"text": "scheme", "start": 297, "end": 303}]}, "relations": {"used_for": [{"head": {"text": "training resources", "start": 83, "end": 101}, "tail": {"text": "automatic abstracting systems", "start": 25, "end": 54}}, {"head": {"text": "annotation scheme", "start": 163, "end": 180}, "tail": {"text": "scientific articles", "start": 185, "end": 204}}, {"head": {"text": "annotation scheme", "start": 163, "end": 180}, "tail": {"text": "resource", "start": 92, "end": 100}}, {"head": {"text": "rhetorical moves of argumentation", "start": 317, "end": 350}, "tail": {"text": "scheme", "start": 174, "end": 180}}]}}, "schema": []}
{"input": "The automated segmentation of images into semantically meaningful parts requires shape information since low-level feature analysis alone often fails to reach this goal. We introduce a novel method of shape constrained image segmentation which is based on mixtures of feature distributions for color and texture as well as probabilistic shape knowledge. The combined approach is formulated in the framework of Bayesian statistics to account for the robust-ness requirement in image understanding. Experimental evidence shows that semantically meaningful segments are inferred, even when image data alone gives rise to ambiguous segmentations.", "output": {"entities": {"task": [{"text": "automated segmentation", "start": 4, "end": 26}, {"text": "shape constrained image segmentation", "start": 201, "end": 237}, {"text": "robust-ness requirement in image understanding", "start": 449, "end": 495}], "material": [{"text": "images", "start": 30, "end": 36}, {"text": "image data", "start": 587, "end": 597}], "other_scientific_term": [{"text": "shape information", "start": 81, "end": 98}, {"text": "mixtures of feature distributions", "start": 256, "end": 289}, {"text": "color", "start": 294, "end": 299}, {"text": "texture", "start": 304, "end": 311}, {"text": "probabilistic shape knowledge", "start": 323, "end": 352}], "method": [{"text": "low-level feature analysis", "start": 105, "end": 131}, {"text": "method", "start": 191, "end": 197}, {"text": "Bayesian statistics", "start": 410, "end": 429}], "generic": [{"text": "approach", "start": 367, "end": 375}]}, "relations": {"used_for": [{"head": {"text": "images", "start": 30, "end": 36}, "tail": {"text": "automated segmentation", "start": 4, "end": 26}}, {"head": {"text": "method", "start": 191, "end": 197}, "tail": {"text": "shape constrained image segmentation", "start": 201, "end": 237}}, {"head": {"text": "mixtures of feature distributions", "start": 256, "end": 289}, "tail": {"text": "method", "start": 191, "end": 197}}, {"head": {"text": "mixtures of feature distributions", "start": 256, "end": 289}, "tail": {"text": "color", "start": 294, "end": 299}}, {"head": {"text": "mixtures of feature distributions", "start": 256, "end": 289}, "tail": {"text": "texture", "start": 304, "end": 311}}, {"head": {"text": "mixtures of feature distributions", "start": 256, "end": 289}, "tail": {"text": "probabilistic shape knowledge", "start": 323, "end": 352}}, {"head": {"text": "approach", "start": 367, "end": 375}, "tail": {"text": "robust-ness requirement in image understanding", "start": 449, "end": 495}}, {"head": {"text": "Bayesian statistics", "start": 410, "end": 429}, "tail": {"text": "approach", "start": 367, "end": 375}}], "conjunction": [{"head": {"text": "color", "start": 294, "end": 299}, "tail": {"text": "texture", "start": 304, "end": 311}}, {"head": {"text": "texture", "start": 304, "end": 311}, "tail": {"text": "probabilistic shape knowledge", "start": 323, "end": 352}}]}}, "schema": []}
{"input": "The goal of this work is the enrichment of human-machine interactions in a natural language environment. Because a speaker and listener can not be assured to have the same beliefs, contexts, perceptions, backgrounds, or goals, at each point in a conversation, difficulties and mistakes arise when a listener interprets a speaker's utterance. These mistakes can lead to various kinds of misunderstandings between speaker and listener, including reference failures or failure to understand the speaker's intention. We call these misunderstandings miscommunication. Such mistakes can slow, and possibly break down, communication. Our goal is to recognize and isolate such miscommunications and circumvent them. This paper highlights a particular class of miscommunication -- reference problems -- by describing a case study and techniques for avoiding failures of reference. We want to illustrate a framework less restrictive than earlier ones by allowing a speaker leeway in forming an utterance about a task and in determining the conversational vehicle to deliver it. The paper also promotes a new view for extensional reference.", "output": {"entities": {"task": [{"text": "human-machine interactions", "start": 43, "end": 69}, {"text": "miscommunication", "start": 545, "end": 561}, {"text": "miscommunications", "start": 669, "end": 686}, {"text": "miscommunication", "start": 669, "end": 685}, {"text": "reference problems", "start": 772, "end": 790}, {"text": "failures of reference", "start": 849, "end": 870}, {"text": "extensional reference", "start": 1107, "end": 1128}], "other_scientific_term": [{"text": "natural language environment", "start": 75, "end": 103}, {"text": "reference failures", "start": 444, "end": 462}, {"text": "speaker's intention", "start": 492, "end": 511}], "generic": [{"text": "them", "start": 702, "end": 706}, {"text": "techniques", "start": 825, "end": 835}]}, "relations": {"feature_of": [{"head": {"text": "natural language environment", "start": 75, "end": 103}, "tail": {"text": "human-machine interactions", "start": 43, "end": 69}}], "hyponym_of": [{"head": {"text": "reference problems", "start": 772, "end": 790}, "tail": {"text": "miscommunication", "start": 669, "end": 685}}], "used_for": [{"head": {"text": "techniques", "start": 825, "end": 835}, "tail": {"text": "failures of reference", "start": 849, "end": 870}}]}}, "schema": []}
{"input": "Combination methods are an effective way of improving system performance. This paper examines the benefits of system combination for unsupervised WSD. We investigate several voting-and arbiter-based combination strategies over a diverse pool of unsupervised WSD systems. Our combination methods rely on predominant senses which are derived automatically from raw text. Experiments using the SemCor and Senseval-3 data sets demonstrate that our ensembles yield significantly better results when compared with state-of-the-art.", "output": {"entities": {"method": [{"text": "Combination methods", "start": 0, "end": 19}, {"text": "system combination", "start": 110, "end": 128}, {"text": "voting-and arbiter-based combination strategies", "start": 174, "end": 221}, {"text": "unsupervised WSD systems", "start": 245, "end": 269}, {"text": "combination methods", "start": 275, "end": 294}], "task": [{"text": "unsupervised WSD", "start": 133, "end": 149}], "other_scientific_term": [{"text": "predominant senses", "start": 303, "end": 321}], "material": [{"text": "raw text", "start": 359, "end": 367}, {"text": "SemCor and Senseval-3 data sets", "start": 391, "end": 422}], "generic": [{"text": "ensembles", "start": 444, "end": 453}, {"text": "state-of-the-art", "start": 508, "end": 524}]}, "relations": {"used_for": [{"head": {"text": "system combination", "start": 110, "end": 128}, "tail": {"text": "unsupervised WSD", "start": 133, "end": 149}}, {"head": {"text": "voting-and arbiter-based combination strategies", "start": 174, "end": 221}, "tail": {"text": "unsupervised WSD systems", "start": 245, "end": 269}}, {"head": {"text": "predominant senses", "start": 303, "end": 321}, "tail": {"text": "combination methods", "start": 275, "end": 294}}, {"head": {"text": "raw text", "start": 359, "end": 367}, "tail": {"text": "predominant senses", "start": 303, "end": 321}}], "evaluate_for": [{"head": {"text": "SemCor and Senseval-3 data sets", "start": 391, "end": 422}, "tail": {"text": "ensembles", "start": 444, "end": 453}}, {"head": {"text": "SemCor and Senseval-3 data sets", "start": 391, "end": 422}, "tail": {"text": "state-of-the-art", "start": 508, "end": 524}}]}}, "schema": []}
{"input": "The applicability of many current information extraction techniques is severely limited by the need for supervised training data. We demonstrate that for certain field structured extraction tasks, such as classified advertisements and bibliographic citations, small amounts of prior knowledge can be used to learn effective models in a primarily unsupervised fashion. Although hidden Markov models -LRB- HMMs -RRB- provide a suitable generative model for field structured text, general unsupervised HMM learning fails to learn useful structure in either of our domains. However, one can dramatically improve the quality of the learned structure by exploiting simple prior knowledge of the desired solutions. In both domains, we found that unsupervised methods can attain accuracies with 400 unlabeled examples comparable to those attained by supervised methods on 50 labeled examples, and that semi-supervised methods can make good use of small amounts of labeled data.", "output": {"entities": {"method": [{"text": "information extraction techniques", "start": 34, "end": 67}, {"text": "hidden Markov models -LRB- HMMs -RRB-", "start": 377, "end": 414}, {"text": "generative model", "start": 434, "end": 450}, {"text": "unsupervised HMM learning", "start": 486, "end": 511}, {"text": "unsupervised methods", "start": 739, "end": 759}, {"text": "supervised methods", "start": 741, "end": 759}, {"text": "semi-supervised methods", "start": 894, "end": 917}], "material": [{"text": "supervised training data", "start": 104, "end": 128}, {"text": "classified advertisements", "start": 205, "end": 230}, {"text": "bibliographic citations", "start": 235, "end": 258}, {"text": "field structured text", "start": 455, "end": 476}, {"text": "unlabeled examples", "start": 791, "end": 809}, {"text": "labeled examples", "start": 793, "end": 809}, {"text": "labeled data", "start": 956, "end": 968}], "task": [{"text": "field structured extraction tasks", "start": 162, "end": 195}], "other_scientific_term": [{"text": "prior knowledge", "start": 277, "end": 292}, {"text": "prior knowledge", "start": 666, "end": 681}], "metric": [{"text": "accuracies", "start": 771, "end": 781}]}, "relations": {"used_for": [{"head": {"text": "supervised training data", "start": 104, "end": 128}, "tail": {"text": "information extraction techniques", "start": 34, "end": 67}}, {"head": {"text": "prior knowledge", "start": 277, "end": 292}, "tail": {"text": "field structured extraction tasks", "start": 162, "end": 195}}, {"head": {"text": "hidden Markov models -LRB- HMMs -RRB-", "start": 377, "end": 414}, "tail": {"text": "generative model", "start": 434, "end": 450}}, {"head": {"text": "generative model", "start": 434, "end": 450}, "tail": {"text": "field structured text", "start": 455, "end": 476}}, {"head": {"text": "unlabeled examples", "start": 791, "end": 809}, "tail": {"text": "unsupervised methods", "start": 739, "end": 759}}, {"head": {"text": "labeled examples", "start": 793, "end": 809}, "tail": {"text": "supervised methods", "start": 741, "end": 759}}, {"head": {"text": "labeled data", "start": 956, "end": 968}, "tail": {"text": "semi-supervised methods", "start": 894, "end": 917}}], "hyponym_of": [{"head": {"text": "classified advertisements", "start": 205, "end": 230}, "tail": {"text": "field structured extraction tasks", "start": 162, "end": 195}}, {"head": {"text": "bibliographic citations", "start": 235, "end": 258}, "tail": {"text": "field structured extraction tasks", "start": 162, "end": 195}}], "conjunction": [{"head": {"text": "classified advertisements", "start": 205, "end": 230}, "tail": {"text": "bibliographic citations", "start": 235, "end": 258}}], "compare": [{"head": {"text": "unsupervised methods", "start": 739, "end": 759}, "tail": {"text": "supervised methods", "start": 741, "end": 759}}], "evaluate_for": [{"head": {"text": "accuracies", "start": 771, "end": 781}, "tail": {"text": "unsupervised methods", "start": 739, "end": 759}}, {"head": {"text": "accuracies", "start": 771, "end": 781}, "tail": {"text": "supervised methods", "start": 741, "end": 759}}]}}, "schema": []}
{"input": "This paper gives an overall account of a prototype natural language question answering system, called Chat-80. Chat-80 has been designed to be both efficient and easily adaptable to a variety of applications. The system is implemented entirely in Prolog, a programming language based on logic. With the aid of a logic-based grammar formalism called extraposition grammars, Chat-80 translates English questions into the Prolog subset of logic. The resulting logical expression is then transformed by a planning algorithm into efficient Prolog, cf. query optimisation in a relational database. Finally, the Prolog form is executed to yield the answer.", "output": {"entities": {"method": [{"text": "natural language question answering system", "start": 51, "end": 93}, {"text": "Chat-80", "start": 102, "end": 109}, {"text": "Chat-80", "start": 111, "end": 118}, {"text": "logic-based grammar formalism", "start": 312, "end": 341}, {"text": "extraposition grammars", "start": 349, "end": 371}, {"text": "Chat-80", "start": 373, "end": 380}, {"text": "planning algorithm", "start": 501, "end": 519}, {"text": "query optimisation", "start": 547, "end": 565}], "generic": [{"text": "system", "start": 87, "end": 93}], "other_scientific_term": [{"text": "Prolog", "start": 247, "end": 253}, {"text": "programming language", "start": 257, "end": 277}, {"text": "logic", "start": 287, "end": 292}, {"text": "Prolog subset of logic", "start": 419, "end": 441}, {"text": "logical expression", "start": 457, "end": 475}, {"text": "Prolog", "start": 419, "end": 425}, {"text": "Prolog form", "start": 605, "end": 616}], "material": [{"text": "relational database", "start": 571, "end": 590}]}, "relations": {"hyponym_of": [{"head": {"text": "Chat-80", "start": 102, "end": 109}, "tail": {"text": "natural language question answering system", "start": 51, "end": 93}}, {"head": {"text": "Prolog", "start": 247, "end": 253}, "tail": {"text": "programming language", "start": 257, "end": 277}}, {"head": {"text": "extraposition grammars", "start": 349, "end": 371}, "tail": {"text": "logic-based grammar formalism", "start": 312, "end": 341}}], "used_for": [{"head": {"text": "Prolog", "start": 247, "end": 253}, "tail": {"text": "system", "start": 87, "end": 93}}, {"head": {"text": "logic", "start": 287, "end": 292}, "tail": {"text": "programming language", "start": 257, "end": 277}}, {"head": {"text": "extraposition grammars", "start": 349, "end": 371}, "tail": {"text": "Chat-80", "start": 373, "end": 380}}, {"head": {"text": "planning algorithm", "start": 501, "end": 519}, "tail": {"text": "logical expression", "start": 457, "end": 475}}, {"head": {"text": "relational database", "start": 571, "end": 590}, "tail": {"text": "query optimisation", "start": 547, "end": 565}}]}}, "schema": []}
{"input": "Human action recognition from well-segmented 3D skeleton data has been intensively studied and attracting an increasing attention. Online action detection goes one step further and is more challenging, which identifies the action type and localizes the action positions on the fly from the untrimmed stream. In this paper, we study the problem of online action detection from the streaming skeleton data. We propose a multi-task end-to-end Joint Classification-Regression Recurrent Neural Network to better explore the action type and temporal localiza-tion information. By employing a joint classification and regression optimization objective, this network is capable of automatically localizing the start and end points of actions more accurately. Specifically, by leveraging the merits of the deep Long Short-Term Memory -LRB- LSTM -RRB- subnetwork, the proposed model automatically captures the complex long-range temporal dynamics, which naturally avoids the typical sliding window design and thus ensures high computational efficiency. Furthermore, the subtask of regression optimization provides the ability to forecast the action prior to its occurrence. To evaluate our proposed model, we build a large streaming video dataset with annotations. Experimental results on our dataset and the public G3D dataset both demonstrate very promising performance of our scheme.", "output": {"entities": {"task": [{"text": "Human action recognition", "start": 0, "end": 24}, {"text": "Online action detection", "start": 131, "end": 154}, {"text": "online action detection", "start": 347, "end": 370}, {"text": "regression optimization", "start": 611, "end": 634}], "material": [{"text": "well-segmented 3D skeleton data", "start": 30, "end": 61}, {"text": "untrimmed stream", "start": 290, "end": 306}, {"text": "streaming skeleton data", "start": 380, "end": 403}, {"text": "streaming video dataset", "start": 1213, "end": 1236}, {"text": "G3D dataset", "start": 1306, "end": 1317}], "other_scientific_term": [{"text": "action type", "start": 223, "end": 234}, {"text": "action positions", "start": 253, "end": 269}, {"text": "action type", "start": 519, "end": 530}, {"text": "temporal localiza-tion information", "start": 535, "end": 569}, {"text": "joint classification and regression optimization objective", "start": 586, "end": 644}, {"text": "long-range temporal dynamics", "start": 908, "end": 936}, {"text": "computational efficiency", "start": 1017, "end": 1041}], "method": [{"text": "multi-task end-to-end Joint Classification-Regression Recurrent Neural Network", "start": 418, "end": 496}, {"text": "deep Long Short-Term Memory -LRB- LSTM -RRB- subnetwork", "start": 797, "end": 852}, {"text": "sliding window design", "start": 973, "end": 994}], "generic": [{"text": "network", "start": 651, "end": 658}, {"text": "model", "start": 867, "end": 872}, {"text": "model", "start": 1189, "end": 1194}, {"text": "dataset", "start": 1229, "end": 1236}]}, "relations": {"used_for": [{"head": {"text": "well-segmented 3D skeleton data", "start": 30, "end": 61}, "tail": {"text": "Human action recognition", "start": 0, "end": 24}}, {"head": {"text": "Online action detection", "start": 131, "end": 154}, "tail": {"text": "action type", "start": 223, "end": 234}}, {"head": {"text": "Online action detection", "start": 131, "end": 154}, "tail": {"text": "action positions", "start": 253, "end": 269}}, {"head": {"text": "untrimmed stream", "start": 290, "end": 306}, "tail": {"text": "Online action detection", "start": 131, "end": 154}}, {"head": {"text": "streaming skeleton data", "start": 380, "end": 403}, "tail": {"text": "online action detection", "start": 347, "end": 370}}, {"head": {"text": "multi-task end-to-end Joint Classification-Regression Recurrent Neural Network", "start": 418, "end": 496}, "tail": {"text": "action type", "start": 519, "end": 530}}, {"head": {"text": "multi-task end-to-end Joint Classification-Regression Recurrent Neural Network", "start": 418, "end": 496}, "tail": {"text": "temporal localiza-tion information", "start": 535, "end": 569}}, {"head": {"text": "joint classification and regression optimization objective", "start": 586, "end": 644}, "tail": {"text": "network", "start": 651, "end": 658}}, {"head": {"text": "deep Long Short-Term Memory -LRB- LSTM -RRB- subnetwork", "start": 797, "end": 852}, "tail": {"text": "model", "start": 867, "end": 872}}], "conjunction": [{"head": {"text": "action type", "start": 223, "end": 234}, "tail": {"text": "action positions", "start": 253, "end": 269}}, {"head": {"text": "action type", "start": 519, "end": 530}, "tail": {"text": "temporal localiza-tion information", "start": 535, "end": 569}}, {"head": {"text": "dataset", "start": 1229, "end": 1236}, "tail": {"text": "G3D dataset", "start": 1306, "end": 1317}}], "feature_of": [{"head": {"text": "long-range temporal dynamics", "start": 908, "end": 936}, "tail": {"text": "model", "start": 867, "end": 872}}], "evaluate_for": [{"head": {"text": "streaming video dataset", "start": 1213, "end": 1236}, "tail": {"text": "model", "start": 1189, "end": 1194}}]}}, "schema": []}
{"input": "The task of machine translation -LRB- MT -RRB- evaluation is closely related to the task of sentence-level semantic equivalence classification. This paper investigates the utility of applying standard MT evaluation methods -LRB- BLEU, NIST, WER and PER -RRB- to building classifiers to predict semantic equivalence and entailment. We also introduce a novel classification method based on PER which leverages part of speech information of the words contributing to the word matches and non-matches in the sentence. Our results show that MT evaluation techniques are able to produce useful features for paraphrase classification and to a lesser extent entailment. Our technique gives a substantial improvement in paraphrase classification accuracy over all of the other models used in the experiments.", "output": {"entities": {"task": [{"text": "machine translation -LRB- MT -RRB- evaluation", "start": 12, "end": 57}, {"text": "sentence-level semantic equivalence classification", "start": 92, "end": 142}, {"text": "semantic equivalence", "start": 107, "end": 127}, {"text": "entailment", "start": 319, "end": 329}, {"text": "word matches and non-matches", "start": 468, "end": 496}, {"text": "paraphrase classification", "start": 601, "end": 626}, {"text": "entailment", "start": 650, "end": 660}, {"text": "paraphrase classification", "start": 711, "end": 736}], "metric": [{"text": "MT evaluation methods", "start": 201, "end": 222}, {"text": "BLEU", "start": 229, "end": 233}, {"text": "NIST", "start": 235, "end": 239}, {"text": "WER", "start": 241, "end": 244}, {"text": "PER", "start": 249, "end": 252}, {"text": "PER", "start": 388, "end": 391}, {"text": "paraphrase classification accuracy", "start": 711, "end": 745}], "method": [{"text": "classifiers", "start": 271, "end": 282}, {"text": "classification method", "start": 357, "end": 378}, {"text": "MT evaluation techniques", "start": 536, "end": 560}], "other_scientific_term": [{"text": "part of speech information", "start": 408, "end": 434}, {"text": "features", "start": 588, "end": 596}], "generic": [{"text": "technique", "start": 550, "end": 559}, {"text": "models", "start": 768, "end": 774}]}, "relations": {"conjunction": [{"head": {"text": "machine translation -LRB- MT -RRB- evaluation", "start": 12, "end": 57}, "tail": {"text": "sentence-level semantic equivalence classification", "start": 92, "end": 142}}, {"head": {"text": "BLEU", "start": 229, "end": 233}, "tail": {"text": "NIST", "start": 235, "end": 239}}, {"head": {"text": "NIST", "start": 235, "end": 239}, "tail": {"text": "WER", "start": 241, "end": 244}}, {"head": {"text": "WER", "start": 241, "end": 244}, "tail": {"text": "PER", "start": 249, "end": 252}}, {"head": {"text": "semantic equivalence", "start": 107, "end": 127}, "tail": {"text": "entailment", "start": 319, "end": 329}}, {"head": {"text": "paraphrase classification", "start": 601, "end": 626}, "tail": {"text": "entailment", "start": 650, "end": 660}}], "used_for": [{"head": {"text": "MT evaluation methods", "start": 201, "end": 222}, "tail": {"text": "classifiers", "start": 271, "end": 282}}, {"head": {"text": "classifiers", "start": 271, "end": 282}, "tail": {"text": "semantic equivalence", "start": 107, "end": 127}}, {"head": {"text": "classifiers", "start": 271, "end": 282}, "tail": {"text": "entailment", "start": 319, "end": 329}}, {"head": {"text": "PER", "start": 388, "end": 391}, "tail": {"text": "classification method", "start": 357, "end": 378}}, {"head": {"text": "PER", "start": 388, "end": 391}, "tail": {"text": "part of speech information", "start": 408, "end": 434}}, {"head": {"text": "part of speech information", "start": 408, "end": 434}, "tail": {"text": "word matches and non-matches", "start": 468, "end": 496}}, {"head": {"text": "MT evaluation techniques", "start": 536, "end": 560}, "tail": {"text": "features", "start": 588, "end": 596}}, {"head": {"text": "MT evaluation techniques", "start": 536, "end": 560}, "tail": {"text": "paraphrase classification", "start": 601, "end": 626}}, {"head": {"text": "MT evaluation techniques", "start": 536, "end": 560}, "tail": {"text": "entailment", "start": 650, "end": 660}}], "hyponym_of": [{"head": {"text": "BLEU", "start": 229, "end": 233}, "tail": {"text": "MT evaluation methods", "start": 201, "end": 222}}, {"head": {"text": "NIST", "start": 235, "end": 239}, "tail": {"text": "MT evaluation methods", "start": 201, "end": 222}}, {"head": {"text": "WER", "start": 241, "end": 244}, "tail": {"text": "MT evaluation methods", "start": 201, "end": 222}}, {"head": {"text": "PER", "start": 249, "end": 252}, "tail": {"text": "MT evaluation methods", "start": 201, "end": 222}}], "compare": [{"head": {"text": "technique", "start": 550, "end": 559}, "tail": {"text": "models", "start": 768, "end": 774}}], "evaluate_for": [{"head": {"text": "paraphrase classification accuracy", "start": 711, "end": 745}, "tail": {"text": "technique", "start": 550, "end": 559}}, {"head": {"text": "paraphrase classification accuracy", "start": 711, "end": 745}, "tail": {"text": "models", "start": 768, "end": 774}}]}}, "schema": []}
{"input": "Given an object model and a black-box measure of similarity between the model and candidate targets, we consider visual object tracking as a numerical optimization problem. During normal tracking conditions when the object is visible from frame to frame, local optimization is used to track the local mode of the similarity measure in a parameter space of translation, rotation and scale. However, when the object becomes partially or totally occluded, such local tracking is prone to failure, especially when common prediction techniques like the Kalman filter do not provide a good estimate of object parameters in future frames. To recover from these inevitable tracking failures, we consider object detection as a global optimization problem and solve it via Adaptive Simulated Annealing -LRB- ASA -RRB-, a method that avoids becoming trapped at local modes and is much faster than exhaustive search. As a Monte Carlo approach, ASA stochastically samples the parameter space, in contrast to local deterministic search. We apply cluster analysis on the sampled parameter space to redetect the object and renew the local tracker. Our numerical hybrid local and global mode-seeking tracker is validated on challenging airborne videos with heavy occlusion and large camera motions. Our approach outperforms state-of-the-art trackers on the VIVID benchmark datasets.", "output": {"entities": {"method": [{"text": "object model", "start": 9, "end": 21}, {"text": "local optimization", "start": 255, "end": 273}, {"text": "local tracking", "start": 458, "end": 472}, {"text": "prediction techniques", "start": 517, "end": 538}, {"text": "Kalman filter", "start": 548, "end": 561}, {"text": "Adaptive Simulated Annealing -LRB- ASA -RRB-", "start": 763, "end": 807}, {"text": "exhaustive search", "start": 886, "end": 903}, {"text": "Monte Carlo approach", "start": 910, "end": 930}, {"text": "ASA", "start": 798, "end": 801}, {"text": "local deterministic search", "start": 995, "end": 1021}, {"text": "cluster analysis", "start": 1032, "end": 1048}, {"text": "local tracker", "start": 1117, "end": 1130}, {"text": "numerical hybrid local and global mode-seeking tracker", "start": 1136, "end": 1190}], "metric": [{"text": "black-box measure of similarity", "start": 28, "end": 59}], "generic": [{"text": "model", "start": 16, "end": 21}, {"text": "it", "start": 56, "end": 58}, {"text": "method", "start": 811, "end": 817}, {"text": "approach", "start": 922, "end": 930}, {"text": "state-of-the-art trackers", "start": 1307, "end": 1332}], "task": [{"text": "visual object tracking", "start": 113, "end": 135}, {"text": "numerical optimization problem", "start": 141, "end": 171}, {"text": "object detection", "start": 696, "end": 712}, {"text": "global optimization problem", "start": 718, "end": 745}], "other_scientific_term": [{"text": "local mode of the similarity measure", "start": 295, "end": 331}, {"text": "parameter space of translation, rotation and scale", "start": 337, "end": 387}, {"text": "parameter space", "start": 337, "end": 352}, {"text": "sampled parameter space", "start": 1056, "end": 1079}, {"text": "heavy occlusion", "start": 1240, "end": 1255}, {"text": "camera motions", "start": 1266, "end": 1280}], "material": [{"text": "airborne videos", "start": 1219, "end": 1234}, {"text": "VIVID benchmark datasets", "start": 1340, "end": 1364}]}, "relations": {"used_for": [{"head": {"text": "numerical optimization problem", "start": 141, "end": 171}, "tail": {"text": "visual object tracking", "start": 113, "end": 135}}, {"head": {"text": "local optimization", "start": 255, "end": 273}, "tail": {"text": "local mode of the similarity measure", "start": 295, "end": 331}}, {"head": {"text": "parameter space of translation, rotation and scale", "start": 337, "end": 387}, "tail": {"text": "local mode of the similarity measure", "start": 295, "end": 331}}, {"head": {"text": "global optimization problem", "start": 718, "end": 745}, "tail": {"text": "object detection", "start": 696, "end": 712}}, {"head": {"text": "Adaptive Simulated Annealing -LRB- ASA -RRB-", "start": 763, "end": 807}, "tail": {"text": "it", "start": 56, "end": 58}}, {"head": {"text": "cluster analysis", "start": 1032, "end": 1048}, "tail": {"text": "sampled parameter space", "start": 1056, "end": 1079}}, {"head": {"text": "cluster analysis", "start": 1032, "end": 1048}, "tail": {"text": "local tracker", "start": 1117, "end": 1130}}], "part_of": [{"head": {"text": "Kalman filter", "start": 548, "end": 561}, "tail": {"text": "prediction techniques", "start": 517, "end": 538}}], "compare": [{"head": {"text": "method", "start": 811, "end": 817}, "tail": {"text": "exhaustive search", "start": 886, "end": 903}}, {"head": {"text": "ASA", "start": 798, "end": 801}, "tail": {"text": "local deterministic search", "start": 995, "end": 1021}}, {"head": {"text": "state-of-the-art trackers", "start": 1307, "end": 1332}, "tail": {"text": "approach", "start": 922, "end": 930}}], "hyponym_of": [{"head": {"text": "ASA", "start": 798, "end": 801}, "tail": {"text": "Monte Carlo approach", "start": 910, "end": 930}}], "evaluate_for": [{"head": {"text": "airborne videos", "start": 1219, "end": 1234}, "tail": {"text": "numerical hybrid local and global mode-seeking tracker", "start": 1136, "end": 1190}}, {"head": {"text": "VIVID benchmark datasets", "start": 1340, "end": 1364}, "tail": {"text": "approach", "start": 922, "end": 930}}, {"head": {"text": "VIVID benchmark datasets", "start": 1340, "end": 1364}, "tail": {"text": "state-of-the-art trackers", "start": 1307, "end": 1332}}], "feature_of": [{"head": {"text": "heavy occlusion", "start": 1240, "end": 1255}, "tail": {"text": "airborne videos", "start": 1219, "end": 1234}}, {"head": {"text": "camera motions", "start": 1266, "end": 1280}, "tail": {"text": "airborne videos", "start": 1219, "end": 1234}}], "conjunction": [{"head": {"text": "heavy occlusion", "start": 1240, "end": 1255}, "tail": {"text": "camera motions", "start": 1266, "end": 1280}}]}}, "schema": []}
{"input": "Techniques for automatically training modules of a natural language generator have recently been proposed, but a fundamental concern is whether the quality of utterances produced with trainable components can compete with hand-crafted template-based or rule-based approaches. In this paper We experimentally evaluate a trainable sentence planner for a spoken dialogue system by eliciting subjective human judgments. In order to perform an exhaustive comparison, we also evaluate a hand-crafted template-based generation component, two rule-based sentence planners, and two baseline sentence planners. We show that the trainable sentence planner performs better than the rule-based systems and the baselines, and as well as the hand-crafted system.", "output": {"entities": {"generic": [{"text": "Techniques", "start": 0, "end": 10}, {"text": "baselines", "start": 697, "end": 706}], "method": [{"text": "automatically training modules", "start": 15, "end": 45}, {"text": "natural language generator", "start": 51, "end": 77}, {"text": "trainable components", "start": 184, "end": 204}, {"text": "hand-crafted template-based or rule-based approaches", "start": 222, "end": 274}, {"text": "trainable sentence planner", "start": 319, "end": 345}, {"text": "spoken dialogue system", "start": 352, "end": 374}, {"text": "hand-crafted template-based generation component", "start": 481, "end": 529}, {"text": "rule-based sentence planners", "start": 535, "end": 563}, {"text": "baseline sentence planners", "start": 573, "end": 599}, {"text": "trainable sentence planner", "start": 618, "end": 644}, {"text": "rule-based systems", "start": 670, "end": 688}, {"text": "hand-crafted system", "start": 727, "end": 746}], "metric": [{"text": "quality of utterances", "start": 148, "end": 169}], "other_scientific_term": [{"text": "utterances", "start": 159, "end": 169}, {"text": "subjective human judgments", "start": 388, "end": 414}]}, "relations": {"used_for": [{"head": {"text": "Techniques", "start": 0, "end": 10}, "tail": {"text": "automatically training modules", "start": 15, "end": 45}}, {"head": {"text": "trainable sentence planner", "start": 319, "end": 345}, "tail": {"text": "spoken dialogue system", "start": 352, "end": 374}}], "part_of": [{"head": {"text": "automatically training modules", "start": 15, "end": 45}, "tail": {"text": "natural language generator", "start": 51, "end": 77}}], "evaluate_for": [{"head": {"text": "utterances", "start": 159, "end": 169}, "tail": {"text": "trainable components", "start": 184, "end": 204}}, {"head": {"text": "utterances", "start": 159, "end": 169}, "tail": {"text": "hand-crafted template-based or rule-based approaches", "start": 222, "end": 274}}, {"head": {"text": "subjective human judgments", "start": 388, "end": 414}, "tail": {"text": "trainable sentence planner", "start": 319, "end": 345}}], "compare": [{"head": {"text": "trainable components", "start": 184, "end": 204}, "tail": {"text": "hand-crafted template-based or rule-based approaches", "start": 222, "end": 274}}, {"head": {"text": "trainable sentence planner", "start": 618, "end": 644}, "tail": {"text": "rule-based systems", "start": 670, "end": 688}}, {"head": {"text": "trainable sentence planner", "start": 618, "end": 644}, "tail": {"text": "baselines", "start": 697, "end": 706}}, {"head": {"text": "trainable sentence planner", "start": 618, "end": 644}, "tail": {"text": "hand-crafted system", "start": 727, "end": 746}}], "conjunction": [{"head": {"text": "hand-crafted template-based generation component", "start": 481, "end": 529}, "tail": {"text": "rule-based sentence planners", "start": 535, "end": 563}}, {"head": {"text": "rule-based sentence planners", "start": 535, "end": 563}, "tail": {"text": "baseline sentence planners", "start": 573, "end": 599}}, {"head": {"text": "rule-based systems", "start": 670, "end": 688}, "tail": {"text": "baselines", "start": 697, "end": 706}}, {"head": {"text": "baselines", "start": 697, "end": 706}, "tail": {"text": "hand-crafted system", "start": 727, "end": 746}}]}}, "schema": []}
{"input": "A new algorithm is proposed for novel view generation in one-to-one teleconferencing applications. Given the video streams acquired by two cameras placed on either side of a computer monitor, the proposed algorithm synthesises images from a virtual camera in arbitrary position -LRB- typically located within the monitor -RRB- to facilitate eye contact. Our technique is based on an improved, dynamic-programming, stereo algorithm for efficient novel-view generation. The two main contributions of this paper are: i -RRB- a new type of three-plane graph for dense-stereo dynamic-programming, that encourages correct occlusion labeling; ii -RRB- a compact geometric derivation for novel-view synthesis by direct projection of the minimum-cost surface. Furthermore, this paper presents a novel algorithm for the temporal maintenance of a background model to enhance the rendering of occlusions and reduce temporal artefacts -LRB- flicker -RRB-; and a cost aggregation algorithm that acts directly on our three-dimensional matching cost space. Examples are given that demonstrate the robustness of the new algorithm to spatial and temporal artefacts for long stereo video streams. These include demonstrations of synthesis of cyclopean views of extended conversational sequences. We further demonstrate synthesis from a freely translating virtual camera.", "output": {"entities": {"generic": [{"text": "algorithm", "start": 6, "end": 15}, {"text": "algorithm", "start": 205, "end": 214}, {"text": "technique", "start": 358, "end": 367}, {"text": "algorithm", "start": 421, "end": 430}, {"text": "algorithm", "start": 792, "end": 801}], "task": [{"text": "novel view generation", "start": 32, "end": 53}, {"text": "one-to-one teleconferencing applications", "start": 57, "end": 97}, {"text": "novel-view generation", "start": 445, "end": 466}, {"text": "occlusion labeling", "start": 616, "end": 634}, {"text": "novel-view synthesis", "start": 680, "end": 700}, {"text": "temporal maintenance of a background model", "start": 810, "end": 852}, {"text": "rendering of occlusions", "start": 868, "end": 891}, {"text": "synthesis of cyclopean views of extended conversational sequences", "start": 1210, "end": 1275}, {"text": "synthesis", "start": 215, "end": 224}], "material": [{"text": "video streams", "start": 109, "end": 122}, {"text": "images", "start": 227, "end": 233}, {"text": "long stereo video streams", "start": 1151, "end": 1176}, {"text": "extended conversational sequences", "start": 1242, "end": 1275}], "other_scientific_term": [{"text": "cameras", "start": 139, "end": 146}, {"text": "computer monitor", "start": 174, "end": 190}, {"text": "virtual camera", "start": 241, "end": 255}, {"text": "arbitrary position", "start": 259, "end": 277}, {"text": "eye contact", "start": 341, "end": 352}, {"text": "compact geometric derivation", "start": 647, "end": 675}, {"text": "temporal artefacts -LRB- flicker -RRB-", "start": 903, "end": 941}, {"text": "three-dimensional matching cost space", "start": 1002, "end": 1039}, {"text": "spatial and temporal artefacts", "start": 1116, "end": 1146}, {"text": "cyclopean views", "start": 1223, "end": 1238}, {"text": "translating virtual camera", "start": 1324, "end": 1350}], "method": [{"text": "dynamic-programming, stereo algorithm", "start": 393, "end": 430}, {"text": "three-plane graph", "start": 536, "end": 553}, {"text": "dense-stereo dynamic-programming", "start": 558, "end": 590}, {"text": "direct projection of the minimum-cost surface", "start": 704, "end": 749}, {"text": "cost aggregation algorithm", "start": 949, "end": 975}], "metric": [{"text": "robustness", "start": 1081, "end": 1091}]}, "relations": {"used_for": [{"head": {"text": "algorithm", "start": 6, "end": 15}, "tail": {"text": "novel view generation", "start": 32, "end": 53}}, {"head": {"text": "novel view generation", "start": 32, "end": 53}, "tail": {"text": "one-to-one teleconferencing applications", "start": 57, "end": 97}}, {"head": {"text": "cameras", "start": 139, "end": 146}, "tail": {"text": "video streams", "start": 109, "end": 122}}, {"head": {"text": "algorithm", "start": 205, "end": 214}, "tail": {"text": "eye contact", "start": 341, "end": 352}}, {"head": {"text": "virtual camera", "start": 241, "end": 255}, "tail": {"text": "images", "start": 227, "end": 233}}, {"head": {"text": "technique", "start": 358, "end": 367}, "tail": {"text": "novel-view generation", "start": 445, "end": 466}}, {"head": {"text": "dynamic-programming, stereo algorithm", "start": 393, "end": 430}, "tail": {"text": "technique", "start": 358, "end": 367}}, {"head": {"text": "three-plane graph", "start": 536, "end": 553}, "tail": {"text": "dense-stereo dynamic-programming", "start": 558, "end": 590}}, {"head": {"text": "dense-stereo dynamic-programming", "start": 558, "end": 590}, "tail": {"text": "occlusion labeling", "start": 616, "end": 634}}, {"head": {"text": "compact geometric derivation", "start": 647, "end": 675}, "tail": {"text": "novel-view synthesis", "start": 680, "end": 700}}, {"head": {"text": "direct projection of the minimum-cost surface", "start": 704, "end": 749}, "tail": {"text": "compact geometric derivation", "start": 647, "end": 675}}, {"head": {"text": "algorithm", "start": 421, "end": 430}, "tail": {"text": "temporal maintenance of a background model", "start": 810, "end": 852}}, {"head": {"text": "algorithm", "start": 421, "end": 430}, "tail": {"text": "rendering of occlusions", "start": 868, "end": 891}}, {"head": {"text": "algorithm", "start": 421, "end": 430}, "tail": {"text": "temporal artefacts -LRB- flicker -RRB-", "start": 903, "end": 941}}, {"head": {"text": "cost aggregation algorithm", "start": 949, "end": 975}, "tail": {"text": "three-dimensional matching cost space", "start": 1002, "end": 1039}}, {"head": {"text": "algorithm", "start": 792, "end": 801}, "tail": {"text": "spatial and temporal artefacts", "start": 1116, "end": 1146}}, {"head": {"text": "spatial and temporal artefacts", "start": 1116, "end": 1146}, "tail": {"text": "long stereo video streams", "start": 1151, "end": 1176}}, {"head": {"text": "translating virtual camera", "start": 1324, "end": 1350}, "tail": {"text": "synthesis", "start": 215, "end": 224}}], "feature_of": [{"head": {"text": "arbitrary position", "start": 259, "end": 277}, "tail": {"text": "virtual camera", "start": 241, "end": 255}}], "conjunction": [{"head": {"text": "cost aggregation algorithm", "start": 949, "end": 975}, "tail": {"text": "algorithm", "start": 421, "end": 430}}], "evaluate_for": [{"head": {"text": "robustness", "start": 1081, "end": 1091}, "tail": {"text": "algorithm", "start": 792, "end": 801}}]}}, "schema": []}
{"input": "Manual acquisition of semantic constraints in broad domains is very expensive. This paper presents an automatic scheme for collecting statistics on cooccurrence patterns in a large corpus. To a large extent, these statistics reflect semantic constraints and thus are used to disambiguate anaphora references and syntactic ambiguities. The scheme was implemented by gathering statistics on the output of other linguistic tools. An experiment was performed to resolve references of the pronoun ``it'' in sentences that were randomly selected from the corpus. The results of the experiment show that in most of the cases the cooccurrence statistics indeed reflect the semantic constraints and thus provide a basis for a useful disambiguation tool.", "output": {"entities": {"task": [{"text": "Manual acquisition of semantic constraints", "start": 0, "end": 42}], "other_scientific_term": [{"text": "cooccurrence patterns", "start": 148, "end": 169}, {"text": "semantic constraints", "start": 22, "end": 42}, {"text": "anaphora references", "start": 288, "end": 307}, {"text": "syntactic ambiguities", "start": 312, "end": 333}, {"text": "cooccurrence statistics", "start": 622, "end": 645}, {"text": "semantic constraints", "start": 233, "end": 253}], "method": [{"text": "linguistic tools", "start": 409, "end": 425}, {"text": "disambiguation tool", "start": 724, "end": 743}]}, "relations": {"used_for": [{"head": {"text": "semantic constraints", "start": 22, "end": 42}, "tail": {"text": "anaphora references", "start": 288, "end": 307}}, {"head": {"text": "semantic constraints", "start": 22, "end": 42}, "tail": {"text": "syntactic ambiguities", "start": 312, "end": 333}}, {"head": {"text": "cooccurrence statistics", "start": 622, "end": 645}, "tail": {"text": "disambiguation tool", "start": 724, "end": 743}}], "conjunction": [{"head": {"text": "anaphora references", "start": 288, "end": 307}, "tail": {"text": "syntactic ambiguities", "start": 312, "end": 333}}]}}, "schema": []}
{"input": "We present a novel method for discovering parallel sentences in comparable, non-parallel corpora. We train a maximum entropy classifier that, given a pair of sentences, can reliably determine whether or not they are translations of each other. Using this approach, we extract parallel data from large Chinese, Arabic, and English non-parallel newspaper corpora. We evaluate the quality of the extracted data by showing that it improves the performance of a state-of-the-art statistical machine translation system. We also show that a good-quality MT system can be built from scratch by starting with a very small parallel corpus -LRB- 100,000 words -RRB- and exploiting a large non-parallel corpus. Thus, our method can be applied with great benefit to language pairs for which only scarce resources are available.", "output": {"entities": {"generic": [{"text": "method", "start": 19, "end": 25}, {"text": "approach", "start": 255, "end": 263}, {"text": "it", "start": 382, "end": 384}, {"text": "method", "start": 709, "end": 715}], "task": [{"text": "discovering parallel sentences", "start": 30, "end": 60}], "material": [{"text": "comparable, non-parallel corpora", "start": 64, "end": 96}, {"text": "parallel data", "start": 276, "end": 289}, {"text": "Chinese, Arabic, and English non-parallel newspaper corpora", "start": 301, "end": 360}, {"text": "parallel corpus", "start": 613, "end": 628}, {"text": "non-parallel corpus", "start": 678, "end": 697}, {"text": "scarce resources", "start": 783, "end": 799}], "method": [{"text": "maximum entropy classifier", "start": 109, "end": 135}, {"text": "statistical machine translation system", "start": 474, "end": 512}, {"text": "MT system", "start": 547, "end": 556}]}, "relations": {"used_for": [{"head": {"text": "method", "start": 19, "end": 25}, "tail": {"text": "discovering parallel sentences", "start": 30, "end": 60}}, {"head": {"text": "comparable, non-parallel corpora", "start": 64, "end": 96}, "tail": {"text": "discovering parallel sentences", "start": 30, "end": 60}}, {"head": {"text": "approach", "start": 255, "end": 263}, "tail": {"text": "parallel data", "start": 276, "end": 289}}, {"head": {"text": "it", "start": 382, "end": 384}, "tail": {"text": "statistical machine translation system", "start": 474, "end": 512}}, {"head": {"text": "parallel corpus", "start": 613, "end": 628}, "tail": {"text": "MT system", "start": 547, "end": 556}}, {"head": {"text": "non-parallel corpus", "start": 678, "end": 697}, "tail": {"text": "MT system", "start": 547, "end": 556}}, {"head": {"text": "scarce resources", "start": 783, "end": 799}, "tail": {"text": "method", "start": 709, "end": 715}}], "part_of": [{"head": {"text": "parallel data", "start": 276, "end": 289}, "tail": {"text": "Chinese, Arabic, and English non-parallel newspaper corpora", "start": 301, "end": 360}}], "conjunction": [{"head": {"text": "parallel corpus", "start": 613, "end": 628}, "tail": {"text": "non-parallel corpus", "start": 678, "end": 697}}]}}, "schema": []}
{"input": "In this paper, we describe a search procedure for statistical machine translation -LRB- MT -RRB- based on dynamic programming -LRB- DP -RRB-. Starting from a DP-based solution to the traveling salesman problem, we present a novel technique to restrict the possible word reordering between source and target language in order to achieve an efficient search algorithm. A search restriction especially useful for the translation direction from German to English is presented. The experimental tests are carried out on the Verbmobil task -LRB- German-English, 8000-word vocabulary -RRB-, which is a limited-domain spoken-language task.", "output": {"entities": {"generic": [{"text": "search procedure", "start": 29, "end": 45}, {"text": "technique", "start": 230, "end": 239}], "task": [{"text": "statistical machine translation -LRB- MT -RRB-", "start": 50, "end": 96}, {"text": "limited-domain spoken-language task", "start": 595, "end": 630}], "method": [{"text": "dynamic programming -LRB- DP -RRB-", "start": 106, "end": 140}, {"text": "DP-based solution", "start": 158, "end": 175}, {"text": "traveling salesman problem", "start": 183, "end": 209}, {"text": "search algorithm", "start": 349, "end": 365}], "other_scientific_term": [{"text": "word reordering", "start": 265, "end": 280}], "material": [{"text": "Verbmobil task", "start": 519, "end": 533}, {"text": "German-English", "start": 540, "end": 554}]}, "relations": {"used_for": [{"head": {"text": "search procedure", "start": 29, "end": 45}, "tail": {"text": "statistical machine translation -LRB- MT -RRB-", "start": 50, "end": 96}}, {"head": {"text": "dynamic programming -LRB- DP -RRB-", "start": 106, "end": 140}, "tail": {"text": "statistical machine translation -LRB- MT -RRB-", "start": 50, "end": 96}}, {"head": {"text": "technique", "start": 230, "end": 239}, "tail": {"text": "search algorithm", "start": 349, "end": 365}}], "hyponym_of": [{"head": {"text": "Verbmobil task", "start": 519, "end": 533}, "tail": {"text": "limited-domain spoken-language task", "start": 595, "end": 630}}]}}, "schema": []}
{"input": "A purely functional implementation of LR-parsers is given, together with a simple correctness proof. It is presented as a generalization of the recursive descent parser. For non-LR grammars the time-complexity of our parser is cubic if the functions that constitute the parser are implemented as memo-functions, i.e. functions that memorize the results of previous invocations. Memo-functions also facilitate a simple way to construct a very compact representation of the parse forest. For LR -LRB- 0 -RRB- grammars, our algorithm is closely related to the recursive ascent parsers recently discovered by Kruse-man Aretz -LSB- 1 -RSB- and Roberts -LSB- 2 -RSB-. Extended CF grammars -LRB- grammars with regular expressions at the right hand side -RRB- can be parsed with a simple modification of the LR-parser for normal CF grammars.", "output": {"entities": {"method": [{"text": "LR-parsers", "start": 38, "end": 48}, {"text": "correctness proof", "start": 82, "end": 99}, {"text": "recursive descent parser", "start": 144, "end": 168}, {"text": "non-LR grammars", "start": 174, "end": 189}, {"text": "parser", "start": 41, "end": 47}, {"text": "parser", "start": 162, "end": 168}, {"text": "memo-functions", "start": 296, "end": 310}, {"text": "Memo-functions", "start": 378, "end": 392}, {"text": "LR -LRB- 0 -RRB- grammars", "start": 490, "end": 515}, {"text": "Extended CF grammars", "start": 662, "end": 682}, {"text": "grammars", "start": 181, "end": 189}, {"text": "LR-parser", "start": 38, "end": 47}, {"text": "CF grammars", "start": 671, "end": 682}], "generic": [{"text": "It", "start": 101, "end": 103}, {"text": "algorithm", "start": 521, "end": 530}], "metric": [{"text": "time-complexity", "start": 194, "end": 209}], "other_scientific_term": [{"text": "parse forest", "start": 472, "end": 484}, {"text": "recursive ascent parsers", "start": 557, "end": 581}, {"text": "regular expressions", "start": 703, "end": 722}]}, "relations": {"conjunction": [{"head": {"text": "correctness proof", "start": 82, "end": 99}, "tail": {"text": "LR-parsers", "start": 38, "end": 48}}, {"head": {"text": "algorithm", "start": 521, "end": 530}, "tail": {"text": "recursive ascent parsers", "start": 557, "end": 581}}], "used_for": [{"head": {"text": "recursive descent parser", "start": 144, "end": 168}, "tail": {"text": "It", "start": 101, "end": 103}}, {"head": {"text": "parser", "start": 41, "end": 47}, "tail": {"text": "non-LR grammars", "start": 174, "end": 189}}, {"head": {"text": "memo-functions", "start": 296, "end": 310}, "tail": {"text": "parser", "start": 162, "end": 168}}, {"head": {"text": "Memo-functions", "start": 378, "end": 392}, "tail": {"text": "parse forest", "start": 472, "end": 484}}, {"head": {"text": "algorithm", "start": 521, "end": 530}, "tail": {"text": "LR -LRB- 0 -RRB- grammars", "start": 490, "end": 515}}, {"head": {"text": "LR-parser", "start": 38, "end": 47}, "tail": {"text": "Extended CF grammars", "start": 662, "end": 682}}, {"head": {"text": "LR-parser", "start": 38, "end": 47}, "tail": {"text": "CF grammars", "start": 671, "end": 682}}], "evaluate_for": [{"head": {"text": "time-complexity", "start": 194, "end": 209}, "tail": {"text": "parser", "start": 41, "end": 47}}], "feature_of": [{"head": {"text": "regular expressions", "start": 703, "end": 722}, "tail": {"text": "grammars", "start": 181, "end": 189}}]}}, "schema": []}
{"input": "In this paper we explore a new theory of discourse structure that stresses the role of purpose and processing in discourse. In this theory, discourse structure is composed of three separate but interrelated components: the structure of the sequence of utterances -LRB- called the linguistic structure -RRB-, a structure of purposes -LRB- called the intentional structure -RRB-, and the state of focus of attention -LRB- called the attentional state -RRB-. The linguistic structure consists of segments of the discourse into which the utterances naturally aggregate. The intentional structure captures the discourse-relevant purposes, expressed in each of the linguistic segments as well as relationships among them. The attentional state is an abstraction of the focus of attention of the participants as the discourse unfolds. The attentional state, being dynamic, records the objects, properties, and relations that are salient at each point of the discourse. The distinction among these components is essential to provide an adequate explanation of such discourse phenomena as cue phrases, referring expressions, and interruptions. The theory of attention, intention, and aggregation of utterances is illustrated in the paper with a number of example discourses. Various properties of discourse are described, and explanations for the behaviour of cue phrases, referring expressions, and interruptions are explored. This theory provides a framework for describing the processing of utterances in a discourse. Discourse processing requires recognizing how the utterances of the discourse aggregate into segments, recognizing the intentions expressed in the discourse and the relationships among intentions, and tracking the discourse through the operation of the mechanisms associated with attentional state. This processing description specifies in these recognition tasks the role of information from the discourse and from the participants' knowledge of the domain.", "output": {"entities": {"method": [{"text": "theory of discourse structure", "start": 31, "end": 60}, {"text": "theory of attention, intention, and aggregation of utterances", "start": 1139, "end": 1200}, {"text": "Discourse processing", "start": 1512, "end": 1532}], "other_scientific_term": [{"text": "discourse", "start": 41, "end": 50}, {"text": "discourse structure", "start": 41, "end": 60}, {"text": "linguistic structure", "start": 280, "end": 300}, {"text": "intentional structure", "start": 349, "end": 370}, {"text": "attentional state", "start": 431, "end": 448}, {"text": "linguistic structure", "start": 460, "end": 480}, {"text": "discourse", "start": 113, "end": 122}, {"text": "intentional structure", "start": 570, "end": 591}, {"text": "discourse-relevant purposes", "start": 605, "end": 632}, {"text": "attentional state", "start": 720, "end": 737}, {"text": "discourse", "start": 140, "end": 149}, {"text": "attentional state", "start": 832, "end": 849}, {"text": "discourse", "start": 509, "end": 518}, {"text": "discourse phenomena", "start": 1057, "end": 1076}, {"text": "cue phrases", "start": 1080, "end": 1091}, {"text": "referring expressions", "start": 1093, "end": 1114}, {"text": "interruptions", "start": 1120, "end": 1133}, {"text": "discourses", "start": 1254, "end": 1264}, {"text": "discourse", "start": 605, "end": 614}, {"text": "cue phrases", "start": 1351, "end": 1362}, {"text": "referring expressions", "start": 1364, "end": 1385}, {"text": "interruptions", "start": 1391, "end": 1404}, {"text": "discourse", "start": 809, "end": 818}, {"text": "discourse", "start": 951, "end": 960}, {"text": "discourse", "start": 1057, "end": 1066}, {"text": "discourse", "start": 1254, "end": 1263}, {"text": "attentional state", "start": 1792, "end": 1809}, {"text": "discourse", "start": 1288, "end": 1297}], "generic": [{"text": "theory", "start": 31, "end": 37}, {"text": "components", "start": 207, "end": 217}, {"text": "theory", "start": 132, "end": 138}, {"text": "processing", "start": 99, "end": 109}], "task": [{"text": "recognition tasks", "start": 1858, "end": 1875}]}, "relations": {"part_of": [{"head": {"text": "components", "start": 207, "end": 217}, "tail": {"text": "discourse structure", "start": 41, "end": 60}}, {"head": {"text": "linguistic structure", "start": 280, "end": 300}, "tail": {"text": "components", "start": 207, "end": 217}}, {"head": {"text": "intentional structure", "start": 349, "end": 370}, "tail": {"text": "components", "start": 207, "end": 217}}, {"head": {"text": "attentional state", "start": 431, "end": 448}, "tail": {"text": "components", "start": 207, "end": 217}}], "conjunction": [{"head": {"text": "linguistic structure", "start": 280, "end": 300}, "tail": {"text": "intentional structure", "start": 349, "end": 370}}, {"head": {"text": "intentional structure", "start": 349, "end": 370}, "tail": {"text": "attentional state", "start": 431, "end": 448}}, {"head": {"text": "cue phrases", "start": 1080, "end": 1091}, "tail": {"text": "referring expressions", "start": 1093, "end": 1114}}, {"head": {"text": "referring expressions", "start": 1093, "end": 1114}, "tail": {"text": "interruptions", "start": 1120, "end": 1133}}], "used_for": [{"head": {"text": "intentional structure", "start": 570, "end": 591}, "tail": {"text": "discourse-relevant purposes", "start": 605, "end": 632}}], "hyponym_of": [{"head": {"text": "cue phrases", "start": 1080, "end": 1091}, "tail": {"text": "discourse phenomena", "start": 1057, "end": 1076}}, {"head": {"text": "referring expressions", "start": 1093, "end": 1114}, "tail": {"text": "discourse phenomena", "start": 1057, "end": 1076}}, {"head": {"text": "interruptions", "start": 1120, "end": 1133}, "tail": {"text": "discourse phenomena", "start": 1057, "end": 1076}}]}}, "schema": []}
{"input": "We examine the relationship between the two grammatical formalisms: Tree Adjoining Grammars and Head Grammars. We briefly investigate the weak equivalence of the two formalisms. We then turn to a discussion comparing the linguistic expressiveness of the two formalisms.", "output": {"entities": {"method": [{"text": "grammatical formalisms", "start": 44, "end": 66}, {"text": "Tree Adjoining Grammars", "start": 68, "end": 91}, {"text": "Head Grammars", "start": 96, "end": 109}], "generic": [{"text": "formalisms", "start": 56, "end": 66}, {"text": "formalisms", "start": 166, "end": 176}], "other_scientific_term": [{"text": "linguistic expressiveness", "start": 221, "end": 246}]}, "relations": {"hyponym_of": [{"head": {"text": "Tree Adjoining Grammars", "start": 68, "end": 91}, "tail": {"text": "grammatical formalisms", "start": 44, "end": 66}}, {"head": {"text": "Head Grammars", "start": 96, "end": 109}, "tail": {"text": "grammatical formalisms", "start": 44, "end": 66}}], "compare": [{"head": {"text": "Tree Adjoining Grammars", "start": 68, "end": 91}, "tail": {"text": "Head Grammars", "start": 96, "end": 109}}], "feature_of": [{"head": {"text": "linguistic expressiveness", "start": 221, "end": 246}, "tail": {"text": "formalisms", "start": 166, "end": 176}}]}}, "schema": []}
{"input": "We provide a unified account of sentence-level and text-level anaphora within the framework of a dependency-based grammar model. Criteria for anaphora resolution within sentence boundaries rephrase major concepts from GB's binding theory, while those for text-level anaphora incorporate an adapted version of a Grosz-Sidner-style focus model.", "output": {"entities": {"other_scientific_term": [{"text": "sentence-level and text-level anaphora", "start": 32, "end": 70}, {"text": "text-level anaphora", "start": 51, "end": 70}], "method": [{"text": "dependency-based grammar model", "start": 97, "end": 127}, {"text": "GB's binding theory", "start": 218, "end": 237}, {"text": "Grosz-Sidner-style focus model", "start": 311, "end": 341}], "generic": [{"text": "Criteria", "start": 129, "end": 137}, {"text": "those", "start": 245, "end": 250}], "task": [{"text": "anaphora resolution within sentence boundaries", "start": 142, "end": 188}]}, "relations": {"used_for": [{"head": {"text": "dependency-based grammar model", "start": 97, "end": 127}, "tail": {"text": "sentence-level and text-level anaphora", "start": 32, "end": 70}}, {"head": {"text": "Criteria", "start": 129, "end": 137}, "tail": {"text": "anaphora resolution within sentence boundaries", "start": 142, "end": 188}}, {"head": {"text": "GB's binding theory", "start": 218, "end": 237}, "tail": {"text": "Criteria", "start": 129, "end": 137}}, {"head": {"text": "those", "start": 245, "end": 250}, "tail": {"text": "text-level anaphora", "start": 51, "end": 70}}], "part_of": [{"head": {"text": "Grosz-Sidner-style focus model", "start": 311, "end": 341}, "tail": {"text": "those", "start": 245, "end": 250}}]}}, "schema": []}
{"input": "Coedition of a natural language text and its representation in some interlingual form seems the best and simplest way to share text revision across languages. For various reasons, UNL graphs are the best candidates in this context. We are developing a prototype where, in the simplest sharing scenario, naive users interact directly with the text in their language -LRB- L0 -RRB-, and indirectly with the associated graph. The modified graph is then sent to the UNL-L0 deconverter and the result shown. If is is satisfactory, the errors were probably due to the graph, not to the deconverter, and the graph is sent to deconverters in other languages. Versions in some other languages known by the user may be displayed, so that improvement sharing is visible and encouraging. As new versions are added with appropriate tags and attributes in the original multilingual document, nothing is ever lost, and cooperative working on a document is rendered feasible. On the internal side, liaisons are established between elements of the text and the graph by using broadly available resources such as a LO-English or better a L0-UNL dictionary, a morphosyntactic parser of L0, and a canonical graph2tree transformation. Establishing a ``best'' correspondence between the'' UNL-tree + L0'' and the'' MS-L0 structure'', a lattice, may be done using the dictionary and trying to align the tree and the selected trajectory with as few crossing liaisons as possible. A central goal of this research is to merge approaches from pivot MT, interactive MT, and multilingual text authoring.", "output": {"entities": {"task": [{"text": "Coedition", "start": 0, "end": 9}, {"text": "pivot MT", "start": 1516, "end": 1524}, {"text": "interactive MT", "start": 1526, "end": 1540}, {"text": "multilingual text authoring", "start": 1546, "end": 1573}], "material": [{"text": "natural language text", "start": 15, "end": 36}, {"text": "languages", "start": 148, "end": 157}, {"text": "original multilingual document", "start": 846, "end": 876}, {"text": "LO-English or better a L0-UNL dictionary", "start": 1097, "end": 1137}, {"text": "dictionary", "start": 1127, "end": 1137}], "other_scientific_term": [{"text": "text revision", "start": 127, "end": 140}, {"text": "UNL graphs", "start": 180, "end": 190}, {"text": "graph", "start": 184, "end": 189}, {"text": "graph", "start": 416, "end": 421}, {"text": "graph", "start": 436, "end": 441}, {"text": "graph", "start": 562, "end": 567}, {"text": "liaisons", "start": 982, "end": 990}, {"text": "canonical graph2tree transformation", "start": 1177, "end": 1212}, {"text": "UNL-tree + L0", "start": 1267, "end": 1280}, {"text": "MS-L0 structure", "start": 1293, "end": 1308}, {"text": "lattice", "start": 1314, "end": 1321}, {"text": "crossing liaisons", "start": 1425, "end": 1442}], "method": [{"text": "UNL-L0 deconverter", "start": 462, "end": 480}, {"text": "deconverter", "start": 469, "end": 480}, {"text": "deconverters", "start": 618, "end": 630}, {"text": "morphosyntactic parser of L0", "start": 1141, "end": 1169}], "generic": [{"text": "resources", "start": 1077, "end": 1086}]}, "relations": {"used_for": [{"head": {"text": "Coedition", "start": 0, "end": 9}, "tail": {"text": "text revision", "start": 127, "end": 140}}, {"head": {"text": "natural language text", "start": 15, "end": 36}, "tail": {"text": "Coedition", "start": 0, "end": 9}}, {"head": {"text": "graph", "start": 416, "end": 421}, "tail": {"text": "UNL-L0 deconverter", "start": 462, "end": 480}}, {"head": {"text": "resources", "start": 1077, "end": 1086}, "tail": {"text": "liaisons", "start": 982, "end": 990}}, {"head": {"text": "dictionary", "start": 1127, "end": 1137}, "tail": {"text": "lattice", "start": 1314, "end": 1321}}], "hyponym_of": [{"head": {"text": "LO-English or better a L0-UNL dictionary", "start": 1097, "end": 1137}, "tail": {"text": "resources", "start": 1077, "end": 1086}}, {"head": {"text": "morphosyntactic parser of L0", "start": 1141, "end": 1169}, "tail": {"text": "resources", "start": 1077, "end": 1086}}, {"head": {"text": "canonical graph2tree transformation", "start": 1177, "end": 1212}, "tail": {"text": "resources", "start": 1077, "end": 1086}}], "conjunction": [{"head": {"text": "LO-English or better a L0-UNL dictionary", "start": 1097, "end": 1137}, "tail": {"text": "morphosyntactic parser of L0", "start": 1141, "end": 1169}}, {"head": {"text": "morphosyntactic parser of L0", "start": 1141, "end": 1169}, "tail": {"text": "canonical graph2tree transformation", "start": 1177, "end": 1212}}, {"head": {"text": "UNL-tree + L0", "start": 1267, "end": 1280}, "tail": {"text": "MS-L0 structure", "start": 1293, "end": 1308}}, {"head": {"text": "pivot MT", "start": 1516, "end": 1524}, "tail": {"text": "interactive MT", "start": 1526, "end": 1540}}, {"head": {"text": "interactive MT", "start": 1526, "end": 1540}, "tail": {"text": "multilingual text authoring", "start": 1546, "end": 1573}}]}}, "schema": []}
{"input": "The reality of analogies between words is refuted by noone -LRB- e.g., I walked is to to walk as I laughed is to to laugh, noted I walked: to walk:: I laughed: to laugh -RRB-. But computational linguists seem to be quite dubious about analogies between sentences: they would not be enough numerous to be of any use. We report experiments conducted on a multilingual corpus to estimate the number of analogies among the sentences that it contains. We give two estimates, a lower one and a higher one. As an analogy must be valid on the level of form as well as on the level of meaning, we relied on the idea that translation should preserve meaning to test for similar meanings.", "output": {"entities": {"task": [{"text": "analogies between words", "start": 15, "end": 38}, {"text": "analogies between sentences", "start": 235, "end": 262}], "material": [{"text": "multilingual corpus", "start": 353, "end": 372}], "other_scientific_term": [{"text": "analogies", "start": 15, "end": 24}, {"text": "analogy", "start": 506, "end": 513}]}, "relations": {"evaluate_for": [{"head": {"text": "multilingual corpus", "start": 353, "end": 372}, "tail": {"text": "analogies", "start": 15, "end": 24}}]}}, "schema": []}
{"input": "Our goal is to learn a Mahalanobis distance by minimizing a loss defined on the weighted sum of the precision at different ranks. Our core motivation is that minimizing a weighted rank loss is a natural criterion for many problems in computer vision such as person re-identification. We propose a novel metric learning formulation called Weighted Approximate Rank Component Analysis -LRB- WARCA -RRB-. We then derive a scalable stochastic gradient descent algorithm for the resulting learning problem. We also derive an efficient non-linear extension of WARCA by using the kernel trick. Kernel space embedding decouples the training and prediction costs from the data dimension and enables us to plug inarbitrary distance measures which are more natural for the features. We also address a more general problem of matrix rank degeneration & non-isolated minima in the low-rank matrix optimization by using new type of regularizer which approximately enforces the or-thonormality of the learned matrix very efficiently. We validate this new method on nine standard person re-identification datasets including two large scale Market-1501 and CUHK03 datasets and show that we improve upon the current state-of-the-art methods on all of them.", "output": {"entities": {"task": [{"text": "Mahalanobis distance", "start": 23, "end": 43}, {"text": "computer vision", "start": 234, "end": 249}, {"text": "person re-identification", "start": 258, "end": 282}, {"text": "low-rank matrix optimization", "start": 868, "end": 896}], "other_scientific_term": [{"text": "loss", "start": 60, "end": 64}, {"text": "weighted sum", "start": 80, "end": 92}, {"text": "ranks", "start": 123, "end": 128}, {"text": "weighted rank loss", "start": 171, "end": 189}, {"text": "Kernel space embedding", "start": 587, "end": 609}, {"text": "training and prediction costs", "start": 624, "end": 653}, {"text": "data dimension", "start": 663, "end": 677}, {"text": "features", "start": 762, "end": 770}, {"text": "matrix rank degeneration", "start": 814, "end": 838}, {"text": "non-isolated minima", "start": 841, "end": 860}, {"text": "or-thonormality", "start": 963, "end": 978}, {"text": "learned matrix", "start": 986, "end": 1000}], "metric": [{"text": "precision", "start": 100, "end": 109}], "method": [{"text": "metric learning formulation", "start": 303, "end": 330}, {"text": "Weighted Approximate Rank Component Analysis -LRB- WARCA -RRB-", "start": 338, "end": 400}, {"text": "stochastic gradient descent algorithm", "start": 428, "end": 465}, {"text": "non-linear extension of WARCA", "start": 530, "end": 559}, {"text": "WARCA", "start": 389, "end": 394}, {"text": "kernel trick", "start": 573, "end": 585}, {"text": "inarbitrary distance measures", "start": 701, "end": 730}, {"text": "regularizer", "start": 918, "end": 929}], "generic": [{"text": "learning problem", "start": 484, "end": 500}, {"text": "method", "start": 1040, "end": 1046}, {"text": "them", "start": 1233, "end": 1237}], "material": [{"text": "person re-identification datasets", "start": 1064, "end": 1097}, {"text": "scale Market-1501", "start": 1118, "end": 1135}, {"text": "CUHK03 datasets", "start": 1140, "end": 1155}]}, "relations": {"used_for": [{"head": {"text": "loss", "start": 60, "end": 64}, "tail": {"text": "Mahalanobis distance", "start": 23, "end": 43}}, {"head": {"text": "weighted rank loss", "start": 171, "end": 189}, "tail": {"text": "computer vision", "start": 234, "end": 249}}, {"head": {"text": "weighted rank loss", "start": 171, "end": 189}, "tail": {"text": "person re-identification", "start": 258, "end": 282}}, {"head": {"text": "stochastic gradient descent algorithm", "start": 428, "end": 465}, "tail": {"text": "learning problem", "start": 484, "end": 500}}, {"head": {"text": "kernel trick", "start": 573, "end": 585}, "tail": {"text": "non-linear extension of WARCA", "start": 530, "end": 559}}, {"head": {"text": "Kernel space embedding", "start": 587, "end": 609}, "tail": {"text": "inarbitrary distance measures", "start": 701, "end": 730}}, {"head": {"text": "regularizer", "start": 918, "end": 929}, "tail": {"text": "low-rank matrix optimization", "start": 868, "end": 896}}, {"head": {"text": "regularizer", "start": 918, "end": 929}, "tail": {"text": "or-thonormality", "start": 963, "end": 978}}], "feature_of": [{"head": {"text": "weighted sum", "start": 80, "end": 92}, "tail": {"text": "precision", "start": 100, "end": 109}}, {"head": {"text": "matrix rank degeneration", "start": 814, "end": 838}, "tail": {"text": "low-rank matrix optimization", "start": 868, "end": 896}}, {"head": {"text": "non-isolated minima", "start": 841, "end": 860}, "tail": {"text": "low-rank matrix optimization", "start": 868, "end": 896}}, {"head": {"text": "or-thonormality", "start": 963, "end": 978}, "tail": {"text": "learned matrix", "start": 986, "end": 1000}}], "hyponym_of": [{"head": {"text": "person re-identification", "start": 258, "end": 282}, "tail": {"text": "computer vision", "start": 234, "end": 249}}, {"head": {"text": "Weighted Approximate Rank Component Analysis -LRB- WARCA -RRB-", "start": 338, "end": 400}, "tail": {"text": "metric learning formulation", "start": 303, "end": 330}}, {"head": {"text": "scale Market-1501", "start": 1118, "end": 1135}, "tail": {"text": "person re-identification datasets", "start": 1064, "end": 1097}}, {"head": {"text": "CUHK03 datasets", "start": 1140, "end": 1155}, "tail": {"text": "person re-identification datasets", "start": 1064, "end": 1097}}], "conjunction": [{"head": {"text": "matrix rank degeneration", "start": 814, "end": 838}, "tail": {"text": "non-isolated minima", "start": 841, "end": 860}}, {"head": {"text": "CUHK03 datasets", "start": 1140, "end": 1155}, "tail": {"text": "scale Market-1501", "start": 1118, "end": 1135}}], "evaluate_for": [{"head": {"text": "person re-identification datasets", "start": 1064, "end": 1097}, "tail": {"text": "method", "start": 1040, "end": 1046}}]}}, "schema": []}
{"input": "In this paper, we discuss language model adaptation methods given a word list and a raw corpus. In this situation, the general method is to segment the raw corpus automatically using a word list, correct the output sentences by hand, and build a model from the segmented corpus. In this sentence-by-sentence error correction method, however, the annotator encounters grammatically complicated positions and this results in a decrease of productivity. In this paper, we propose to concentrate on correcting the positions in which the words in the list appear by taking a word as a correction unit. This method allows us to avoid these problems and go directly to capturing the statistical behavior of specific words in the application. In the experiments, we used a variety of methods for preparing a segmented corpus and compared the language models by their speech recognition accuracies. The results showed the advantages of our method.", "output": {"entities": {"method": [{"text": "language model adaptation methods", "start": 26, "end": 59}, {"text": "sentence-by-sentence error correction method", "start": 287, "end": 331}, {"text": "language models", "start": 834, "end": 849}], "other_scientific_term": [{"text": "word list", "start": 68, "end": 77}, {"text": "word list", "start": 185, "end": 194}], "material": [{"text": "raw corpus", "start": 84, "end": 94}, {"text": "raw corpus", "start": 152, "end": 162}, {"text": "segmented corpus", "start": 261, "end": 277}], "generic": [{"text": "method", "start": 52, "end": 58}, {"text": "model", "start": 35, "end": 40}, {"text": "method", "start": 127, "end": 133}, {"text": "methods", "start": 52, "end": 59}, {"text": "method", "start": 325, "end": 331}], "task": [{"text": "preparing a segmented corpus", "start": 788, "end": 816}], "metric": [{"text": "speech recognition accuracies", "start": 859, "end": 888}]}, "relations": {"used_for": [{"head": {"text": "word list", "start": 68, "end": 77}, "tail": {"text": "language model adaptation methods", "start": 26, "end": 59}}, {"head": {"text": "raw corpus", "start": 84, "end": 94}, "tail": {"text": "language model adaptation methods", "start": 26, "end": 59}}, {"head": {"text": "method", "start": 52, "end": 58}, "tail": {"text": "raw corpus", "start": 152, "end": 162}}, {"head": {"text": "word list", "start": 185, "end": 194}, "tail": {"text": "method", "start": 52, "end": 58}}, {"head": {"text": "segmented corpus", "start": 261, "end": 277}, "tail": {"text": "model", "start": 35, "end": 40}}, {"head": {"text": "methods", "start": 52, "end": 59}, "tail": {"text": "preparing a segmented corpus", "start": 788, "end": 816}}], "conjunction": [{"head": {"text": "word list", "start": 68, "end": 77}, "tail": {"text": "raw corpus", "start": 84, "end": 94}}], "evaluate_for": [{"head": {"text": "speech recognition accuracies", "start": 859, "end": 888}, "tail": {"text": "language models", "start": 834, "end": 849}}]}}, "schema": []}
{"input": "Many practical modeling problems involve discrete data that are best represented as draws from multinomial or categorical distributions. For example, nucleotides in a DNA sequence, children's names in a given state and year, and text documents are all commonly modeled with multinomial distributions. In all of these cases, we expect some form of dependency between the draws: the nucleotide at one position in the DNA strand may depend on the preceding nucleotides, children's names are highly correlated from year to year, and topics in text may be correlated and dynamic. These dependencies are not naturally captured by the typical Dirichlet-multinomial formulation. Here, we leverage a logistic stick-breaking representation and recent innovations in Pólya-gamma augmentation to reformu-late the multinomial distribution in terms of latent variables with jointly Gaussian likelihoods, enabling us to take advantage of a host of Bayesian inference techniques for Gaussian models with minimal overhead.", "output": {"entities": {"task": [{"text": "modeling problems", "start": 15, "end": 32}, {"text": "Pólya-gamma augmentation", "start": 756, "end": 780}], "material": [{"text": "discrete data", "start": 41, "end": 54}, {"text": "nucleotides in a DNA sequence", "start": 150, "end": 179}, {"text": "text documents", "start": 229, "end": 243}], "method": [{"text": "multinomial or categorical distributions", "start": 95, "end": 135}, {"text": "multinomial distributions", "start": 274, "end": 299}, {"text": "Dirichlet-multinomial formulation", "start": 636, "end": 669}, {"text": "logistic stick-breaking representation", "start": 691, "end": 729}, {"text": "multinomial distribution", "start": 274, "end": 298}, {"text": "Bayesian inference techniques", "start": 933, "end": 962}, {"text": "Gaussian models", "start": 967, "end": 982}], "other_scientific_term": [{"text": "nucleotide", "start": 150, "end": 160}, {"text": "DNA strand", "start": 415, "end": 425}, {"text": "preceding nucleotides", "start": 444, "end": 465}, {"text": "latent variables", "start": 838, "end": 854}, {"text": "jointly Gaussian likelihoods", "start": 860, "end": 888}, {"text": "minimal overhead", "start": 988, "end": 1004}]}, "relations": {"used_for": [{"head": {"text": "discrete data", "start": 41, "end": 54}, "tail": {"text": "modeling problems", "start": 15, "end": 32}}, {"head": {"text": "multinomial or categorical distributions", "start": 95, "end": 135}, "tail": {"text": "modeling problems", "start": 15, "end": 32}}, {"head": {"text": "multinomial distributions", "start": 274, "end": 299}, "tail": {"text": "nucleotides in a DNA sequence", "start": 150, "end": 179}}, {"head": {"text": "multinomial distributions", "start": 274, "end": 299}, "tail": {"text": "text documents", "start": 229, "end": 243}}, {"head": {"text": "logistic stick-breaking representation", "start": 691, "end": 729}, "tail": {"text": "multinomial distribution", "start": 274, "end": 298}}, {"head": {"text": "Pólya-gamma augmentation", "start": 756, "end": 780}, "tail": {"text": "multinomial distribution", "start": 274, "end": 298}}, {"head": {"text": "Bayesian inference techniques", "start": 933, "end": 962}, "tail": {"text": "Gaussian models", "start": 967, "end": 982}}], "part_of": [{"head": {"text": "latent variables", "start": 838, "end": 854}, "tail": {"text": "multinomial distribution", "start": 274, "end": 298}}], "feature_of": [{"head": {"text": "jointly Gaussian likelihoods", "start": 860, "end": 888}, "tail": {"text": "latent variables", "start": 838, "end": 854}}, {"head": {"text": "minimal overhead", "start": 988, "end": 1004}, "tail": {"text": "Gaussian models", "start": 967, "end": 982}}]}}, "schema": []}
{"input": "MINPRAN, a new robust operator, nds good ts in data sets where more than 50% of the points are outliers. Unlike other techniques that handle large outlier percentages, MINPRAN does not rely on a known error bound for the good data. Instead it assumes that the bad data are randomly -LRB- uniformly -RRB- distributed within the dynamic range of the sensor. Based on this, MINPRAN uses random sampling to search for the t and the number of inliers to the t that are least likely to have occurred randomly. It runs in time O -LRB- N 2 + SN log N -RRB-, where S is the number of random samples and N is the number of data points. We demonstrate analytically that MINPRAN distinguishes good ts from ts to random data, and that MINPRAN nds accurate ts and nearly the correct number of inliers, regardless of the percentage of true inliers. MINPRAN's properties are connrmed experimentally on synthetic data and compare favorably to least median of squares. Related work applies MINPRAN to complex range and intensity data 23 -RSB-.", "output": {"entities": {"method": [{"text": "MINPRAN", "start": 0, "end": 7}, {"text": "robust operator", "start": 15, "end": 30}, {"text": "MINPRAN", "start": 168, "end": 175}, {"text": "MINPRAN", "start": 371, "end": 378}, {"text": "random sampling", "start": 384, "end": 399}, {"text": "MINPRAN", "start": 659, "end": 666}, {"text": "MINPRAN", "start": 722, "end": 729}, {"text": "MINPRAN", "start": 834, "end": 841}, {"text": "least median of squares", "start": 926, "end": 949}, {"text": "MINPRAN", "start": 972, "end": 979}], "generic": [{"text": "techniques", "start": 118, "end": 128}, {"text": "it", "start": 240, "end": 242}, {"text": "It", "start": 504, "end": 506}], "metric": [{"text": "large outlier percentages", "start": 141, "end": 166}, {"text": "percentage of true inliers", "start": 806, "end": 832}], "other_scientific_term": [{"text": "error bound", "start": 201, "end": 212}, {"text": "dynamic range of the sensor", "start": 327, "end": 354}, {"text": "complex range", "start": 983, "end": 996}], "material": [{"text": "synthetic data", "start": 886, "end": 900}, {"text": "intensity data", "start": 1001, "end": 1015}]}, "relations": {"hyponym_of": [{"head": {"text": "MINPRAN", "start": 0, "end": 7}, "tail": {"text": "robust operator", "start": 15, "end": 30}}], "used_for": [{"head": {"text": "techniques", "start": 118, "end": 128}, "tail": {"text": "large outlier percentages", "start": 141, "end": 166}}, {"head": {"text": "random sampling", "start": 384, "end": 399}, "tail": {"text": "MINPRAN", "start": 371, "end": 378}}, {"head": {"text": "MINPRAN", "start": 972, "end": 979}, "tail": {"text": "complex range", "start": 983, "end": 996}}, {"head": {"text": "MINPRAN", "start": 972, "end": 979}, "tail": {"text": "intensity data", "start": 1001, "end": 1015}}], "compare": [{"head": {"text": "techniques", "start": 118, "end": 128}, "tail": {"text": "MINPRAN", "start": 168, "end": 175}}, {"head": {"text": "least median of squares", "start": 926, "end": 949}, "tail": {"text": "MINPRAN", "start": 834, "end": 841}}], "evaluate_for": [{"head": {"text": "synthetic data", "start": 886, "end": 900}, "tail": {"text": "MINPRAN", "start": 834, "end": 841}}]}}, "schema": []}
{"input": "Metagrammatical formalisms that combine context-free phrase structure rules and metarules -LRB- MPS grammars -RRB- allow concise statement of generalizations about the syntax of natural languages. Unconstrained MPS grammars, unfortunately, are not computationally safe. We evaluate several proposals for constraining them, basing our assessment on computational tractability and explanatory adequacy. We show that none of them satisfies both criteria, and suggest new directions for research on alternative metagrammatical formalisms.", "output": {"entities": {"method": [{"text": "Metagrammatical formalisms", "start": 0, "end": 26}, {"text": "Unconstrained MPS grammars", "start": 197, "end": 223}, {"text": "metagrammatical formalisms", "start": 507, "end": 533}], "other_scientific_term": [{"text": "context-free phrase structure rules", "start": 40, "end": 75}, {"text": "metarules -LRB- MPS grammars -RRB-", "start": 80, "end": 114}, {"text": "syntax of natural languages", "start": 168, "end": 195}], "generic": [{"text": "them", "start": 317, "end": 321}, {"text": "them", "start": 422, "end": 426}, {"text": "criteria", "start": 442, "end": 450}], "metric": [{"text": "computational tractability and explanatory adequacy", "start": 348, "end": 399}]}, "relations": {"part_of": [{"head": {"text": "context-free phrase structure rules", "start": 40, "end": 75}, "tail": {"text": "Metagrammatical formalisms", "start": 0, "end": 26}}, {"head": {"text": "metarules -LRB- MPS grammars -RRB-", "start": 80, "end": 114}, "tail": {"text": "Metagrammatical formalisms", "start": 0, "end": 26}}], "conjunction": [{"head": {"text": "context-free phrase structure rules", "start": 40, "end": 75}, "tail": {"text": "metarules -LRB- MPS grammars -RRB-", "start": 80, "end": 114}}], "evaluate_for": [{"head": {"text": "computational tractability and explanatory adequacy", "start": 348, "end": 399}, "tail": {"text": "them", "start": 317, "end": 321}}]}}, "schema": []}
{"input": "The unique properties of tree-adjoining grammars -LRB- TAG -RRB- present a challenge for the application of TAGs beyond the limited confines of syntax, for instance, to the task of semantic interpretation or automatic translation of natural language. We present a variant of TAGs, called synchronous TAGs, which characterize correspondences between languages. The formalism's intended usage is to relate expressions of natural languages to their associated semantics represented in a logical form language, or to their translates in another natural language; in summary, we intend it to allow TAGs to be used beyond their role in syntax proper. We discuss the application of synchronous TAGs to concrete examples, mentioning primarily in passing some computational issues that arise in its interpretation.", "output": {"entities": {"method": [{"text": "tree-adjoining grammars -LRB- TAG -RRB-", "start": 25, "end": 64}, {"text": "TAGs", "start": 108, "end": 112}, {"text": "variant of TAGs", "start": 264, "end": 279}, {"text": "TAGs", "start": 275, "end": 279}, {"text": "synchronous TAGs", "start": 288, "end": 304}, {"text": "TAGs", "start": 300, "end": 304}, {"text": "synchronous TAGs", "start": 675, "end": 691}], "other_scientific_term": [{"text": "syntax", "start": 144, "end": 150}, {"text": "expressions of natural languages", "start": 404, "end": 436}, {"text": "semantics", "start": 457, "end": 466}, {"text": "logical form language", "start": 484, "end": 505}, {"text": "natural language", "start": 233, "end": 249}, {"text": "syntax proper", "start": 630, "end": 643}], "task": [{"text": "semantic interpretation", "start": 181, "end": 204}, {"text": "automatic translation of natural language", "start": 208, "end": 249}]}, "relations": {"used_for": [{"head": {"text": "TAGs", "start": 108, "end": 112}, "tail": {"text": "semantic interpretation", "start": 181, "end": 204}}, {"head": {"text": "TAGs", "start": 108, "end": 112}, "tail": {"text": "automatic translation of natural language", "start": 208, "end": 249}}, {"head": {"text": "logical form language", "start": 484, "end": 505}, "tail": {"text": "semantics", "start": 457, "end": 466}}, {"head": {"text": "TAGs", "start": 300, "end": 304}, "tail": {"text": "syntax proper", "start": 630, "end": 643}}], "conjunction": [{"head": {"text": "semantic interpretation", "start": 181, "end": 204}, "tail": {"text": "automatic translation of natural language", "start": 208, "end": 249}}]}}, "schema": []}
{"input": "A model-based approach to on-line cursive handwriting analysis and recognition is presented and evaluated. In this model, on-line handwriting is considered as a modulation of a simple cycloidal pen motion, described by two coupled oscillations with a constant linear drift along the line of the writing. By slow modulations of the amplitudes and phase lags of the two oscillators, a general pen trajectory can be efficiently encoded. These parameters are then quantized into a small number of values without altering the writing intelligibility. A general procedure for the estimation and quantization of these cycloidal motion parameters for arbitrary handwriting is presented. The result is a discrete motor control representation of the continuous pen motion, via the quantized levels of the model parameters. This motor control representation enables successful word spotting and matching of cursive scripts. Our experiments clearly indicate the potential of this dynamic representation for complete cursive handwriting recognition.", "output": {"entities": {"method": [{"text": "model-based approach", "start": 2, "end": 22}, {"text": "discrete motor control representation", "start": 695, "end": 732}, {"text": "motor control representation", "start": 704, "end": 732}, {"text": "dynamic representation", "start": 968, "end": 990}], "task": [{"text": "on-line cursive handwriting analysis and recognition", "start": 26, "end": 78}, {"text": "on-line handwriting", "start": 122, "end": 141}, {"text": "cycloidal pen motion", "start": 184, "end": 204}, {"text": "word spotting", "start": 866, "end": 879}, {"text": "matching of cursive scripts", "start": 884, "end": 911}, {"text": "cursive handwriting recognition", "start": 1004, "end": 1035}], "generic": [{"text": "model", "start": 2, "end": 7}], "other_scientific_term": [{"text": "constant linear drift", "start": 251, "end": 272}, {"text": "pen trajectory", "start": 391, "end": 405}, {"text": "writing intelligibility", "start": 521, "end": 544}, {"text": "cycloidal motion parameters", "start": 611, "end": 638}, {"text": "continuous pen motion", "start": 740, "end": 761}], "material": [{"text": "arbitrary handwriting", "start": 643, "end": 664}]}, "relations": {"used_for": [{"head": {"text": "model-based approach", "start": 2, "end": 22}, "tail": {"text": "on-line cursive handwriting analysis and recognition", "start": 26, "end": 78}}, {"head": {"text": "model", "start": 2, "end": 7}, "tail": {"text": "on-line handwriting", "start": 122, "end": 141}}, {"head": {"text": "cycloidal motion parameters", "start": 611, "end": 638}, "tail": {"text": "arbitrary handwriting", "start": 643, "end": 664}}, {"head": {"text": "discrete motor control representation", "start": 695, "end": 732}, "tail": {"text": "continuous pen motion", "start": 740, "end": 761}}, {"head": {"text": "motor control representation", "start": 704, "end": 732}, "tail": {"text": "word spotting", "start": 866, "end": 879}}, {"head": {"text": "motor control representation", "start": 704, "end": 732}, "tail": {"text": "matching of cursive scripts", "start": 884, "end": 911}}, {"head": {"text": "dynamic representation", "start": 968, "end": 990}, "tail": {"text": "cursive handwriting recognition", "start": 1004, "end": 1035}}], "part_of": [{"head": {"text": "on-line handwriting", "start": 122, "end": 141}, "tail": {"text": "cycloidal pen motion", "start": 184, "end": 204}}], "conjunction": [{"head": {"text": "word spotting", "start": 866, "end": 879}, "tail": {"text": "matching of cursive scripts", "start": 884, "end": 911}}]}}, "schema": []}
{"input": "In the Object Recognition task, there exists a di-chotomy between the categorization of objects and estimating object pose, where the former necessitates a view-invariant representation, while the latter requires a representation capable of capturing pose information over different categories of objects. With the rise of deep archi-tectures, the prime focus has been on object category recognition. Deep learning methods have achieved wide success in this task. In contrast, object pose estimation using these approaches has received relatively less attention. In this work, we study how Convolutional Neural Networks -LRB- CNN -RRB- architectures can be adapted to the task of simultaneous object recognition and pose estimation. We investigate and analyze the layers of various CNN models and extensively compare between them with the goal of discovering how the layers of distributed representations within CNNs represent object pose information and how this contradicts with object category representations. We extensively experiment on two recent large and challenging multi-view datasets and we achieve better than the state-of-the-art.", "output": {"entities": {"task": [{"text": "Object Recognition task", "start": 7, "end": 30}, {"text": "categorization of objects", "start": 70, "end": 95}, {"text": "estimating object pose", "start": 100, "end": 122}, {"text": "object category recognition", "start": 372, "end": 399}, {"text": "object pose estimation", "start": 477, "end": 499}, {"text": "object recognition", "start": 693, "end": 711}, {"text": "pose estimation", "start": 484, "end": 499}], "generic": [{"text": "former", "start": 134, "end": 140}, {"text": "latter", "start": 197, "end": 203}, {"text": "representation", "start": 171, "end": 185}, {"text": "task", "start": 26, "end": 30}, {"text": "approaches", "start": 512, "end": 522}, {"text": "them", "start": 825, "end": 829}, {"text": "this", "start": 453, "end": 457}], "method": [{"text": "view-invariant representation", "start": 156, "end": 185}, {"text": "deep archi-tectures", "start": 323, "end": 342}, {"text": "Deep learning methods", "start": 401, "end": 422}, {"text": "Convolutional Neural Networks -LRB- CNN -RRB- architectures", "start": 590, "end": 649}, {"text": "CNN models", "start": 782, "end": 792}, {"text": "layers of distributed representations", "start": 867, "end": 904}, {"text": "CNNs", "start": 912, "end": 916}, {"text": "object category representations", "start": 981, "end": 1012}], "other_scientific_term": [{"text": "pose information", "start": 251, "end": 267}, {"text": "layers", "start": 764, "end": 770}, {"text": "object pose information", "start": 927, "end": 950}], "material": [{"text": "multi-view datasets", "start": 1076, "end": 1095}]}, "relations": {"part_of": [{"head": {"text": "categorization of objects", "start": 70, "end": 95}, "tail": {"text": "Object Recognition task", "start": 7, "end": 30}}, {"head": {"text": "estimating object pose", "start": 100, "end": 122}, "tail": {"text": "Object Recognition task", "start": 7, "end": 30}}, {"head": {"text": "layers", "start": 764, "end": 770}, "tail": {"text": "CNN models", "start": 782, "end": 792}}, {"head": {"text": "layers of distributed representations", "start": 867, "end": 904}, "tail": {"text": "CNNs", "start": 912, "end": 916}}], "conjunction": [{"head": {"text": "categorization of objects", "start": 70, "end": 95}, "tail": {"text": "estimating object pose", "start": 100, "end": 122}}, {"head": {"text": "object recognition", "start": 693, "end": 711}, "tail": {"text": "pose estimation", "start": 484, "end": 499}}], "used_for": [{"head": {"text": "view-invariant representation", "start": 156, "end": 185}, "tail": {"text": "former", "start": 134, "end": 140}}, {"head": {"text": "representation", "start": 171, "end": 185}, "tail": {"text": "latter", "start": 197, "end": 203}}, {"head": {"text": "representation", "start": 171, "end": 185}, "tail": {"text": "pose information", "start": 251, "end": 267}}, {"head": {"text": "deep archi-tectures", "start": 323, "end": 342}, "tail": {"text": "object category recognition", "start": 372, "end": 399}}, {"head": {"text": "approaches", "start": 512, "end": 522}, "tail": {"text": "object pose estimation", "start": 477, "end": 499}}, {"head": {"text": "Convolutional Neural Networks -LRB- CNN -RRB- architectures", "start": 590, "end": 649}, "tail": {"text": "object recognition", "start": 693, "end": 711}}, {"head": {"text": "Convolutional Neural Networks -LRB- CNN -RRB- architectures", "start": 590, "end": 649}, "tail": {"text": "pose estimation", "start": 484, "end": 499}}, {"head": {"text": "layers of distributed representations", "start": 867, "end": 904}, "tail": {"text": "object pose information", "start": 927, "end": 950}}], "compare": [{"head": {"text": "this", "start": 453, "end": 457}, "tail": {"text": "object category representations", "start": 981, "end": 1012}}]}}, "schema": []}
{"input": "In this paper we present our recent work on harvesting English-Chinese bitexts of the laws of Hong Kong from the Web and aligning them to the subparagraph level via utilizing the numbering system in the legal text hierarchy. Basic methodology and practical techniques are reported in detail. The resultant bilingual corpus, 10.4 M English words and 18.3 M Chinese characters, is an authoritative and comprehensive text collection covering the specific and special domain of HK laws. It is particularly valuable to empirical MT research. This piece of work has also laid a foundation for exploring and harvesting English-Chinese bitexts in a larger volume from the Web.", "output": {"entities": {"material": [{"text": "English-Chinese bitexts", "start": 55, "end": 78}, {"text": "bilingual corpus", "start": 306, "end": 322}, {"text": "English-Chinese bitexts", "start": 612, "end": 635}], "generic": [{"text": "them", "start": 130, "end": 134}, {"text": "It", "start": 483, "end": 485}], "method": [{"text": "numbering system", "start": 179, "end": 195}], "other_scientific_term": [{"text": "legal text hierarchy", "start": 203, "end": 223}], "task": [{"text": "empirical MT research", "start": 514, "end": 535}]}, "relations": {"used_for": [{"head": {"text": "It", "start": 483, "end": 485}, "tail": {"text": "empirical MT research", "start": 514, "end": 535}}]}}, "schema": []}
{"input": "Light fields are image-based representations that use densely sampled rays as a scene description. In this paper, we explore geometric structures of 3D lines in ray space for improving light field triangulation and stereo matching. The triangulation problem aims to fill in the ray space with continuous and non-overlapping simplices anchored at sampled points -LRB- rays -RRB-. Such a triangulation provides a piecewise-linear interpolant useful for light field super-resolution. We show that the light field space is largely bi-linear due to 3D line segments in the scene, and direct tri-angulation of these bilinear subspaces leads to large errors. We instead present a simple but effective algorithm to first map bilinear subspaces to line constraints and then apply Constrained Delaunay Triangulation -LRB- CDT -RRB-. Based on our analysis, we further develop a novel line-assisted graph-cut -LRB- LAGC -RRB- algorithm that effectively encodes 3D line constraints into light field stereo matching. Experiments on synthetic and real data show that both our triangulation and LAGC algorithms outperform state-of-the-art solutions in accuracy and visual quality.", "output": {"entities": {"other_scientific_term": [{"text": "Light fields", "start": 0, "end": 12}, {"text": "densely sampled rays", "start": 54, "end": 74}, {"text": "scene description", "start": 80, "end": 97}, {"text": "ray space", "start": 161, "end": 170}, {"text": "ray space", "start": 278, "end": 287}, {"text": "continuous and non-overlapping simplices", "start": 293, "end": 333}, {"text": "triangulation", "start": 197, "end": 210}, {"text": "piecewise-linear interpolant", "start": 411, "end": 439}, {"text": "light field space", "start": 498, "end": 515}, {"text": "3D line segments", "start": 544, "end": 560}, {"text": "bilinear subspaces", "start": 610, "end": 628}, {"text": "bilinear subspaces", "start": 717, "end": 735}, {"text": "line constraints", "start": 739, "end": 755}, {"text": "3D line constraints", "start": 949, "end": 968}], "method": [{"text": "image-based representations", "start": 17, "end": 44}, {"text": "Constrained Delaunay Triangulation -LRB- CDT -RRB-", "start": 771, "end": 821}, {"text": "line-assisted graph-cut -LRB- LAGC -RRB- algorithm", "start": 873, "end": 923}, {"text": "triangulation and LAGC algorithms", "start": 1061, "end": 1094}], "task": [{"text": "geometric structures of 3D lines", "start": 125, "end": 157}, {"text": "light field triangulation", "start": 185, "end": 210}, {"text": "stereo matching", "start": 215, "end": 230}, {"text": "triangulation problem", "start": 236, "end": 257}, {"text": "light field super-resolution", "start": 451, "end": 479}, {"text": "light field stereo matching", "start": 974, "end": 1001}], "material": [{"text": "synthetic and real data", "start": 1018, "end": 1041}], "generic": [{"text": "state-of-the-art solutions", "start": 1106, "end": 1132}], "metric": [{"text": "accuracy", "start": 1136, "end": 1144}, {"text": "visual quality", "start": 1149, "end": 1163}]}, "relations": {"used_for": [{"head": {"text": "geometric structures of 3D lines", "start": 125, "end": 157}, "tail": {"text": "light field triangulation", "start": 185, "end": 210}}, {"head": {"text": "geometric structures of 3D lines", "start": 125, "end": 157}, "tail": {"text": "stereo matching", "start": 215, "end": 230}}, {"head": {"text": "triangulation", "start": 197, "end": 210}, "tail": {"text": "piecewise-linear interpolant", "start": 411, "end": 439}}, {"head": {"text": "piecewise-linear interpolant", "start": 411, "end": 439}, "tail": {"text": "light field super-resolution", "start": 451, "end": 479}}], "feature_of": [{"head": {"text": "ray space", "start": 161, "end": 170}, "tail": {"text": "geometric structures of 3D lines", "start": 125, "end": 157}}], "conjunction": [{"head": {"text": "light field triangulation", "start": 185, "end": 210}, "tail": {"text": "stereo matching", "start": 215, "end": 230}}], "evaluate_for": [{"head": {"text": "synthetic and real data", "start": 1018, "end": 1041}, "tail": {"text": "triangulation and LAGC algorithms", "start": 1061, "end": 1094}}, {"head": {"text": "synthetic and real data", "start": 1018, "end": 1041}, "tail": {"text": "state-of-the-art solutions", "start": 1106, "end": 1132}}, {"head": {"text": "accuracy", "start": 1136, "end": 1144}, "tail": {"text": "triangulation and LAGC algorithms", "start": 1061, "end": 1094}}, {"head": {"text": "accuracy", "start": 1136, "end": 1144}, "tail": {"text": "state-of-the-art solutions", "start": 1106, "end": 1132}}, {"head": {"text": "visual quality", "start": 1149, "end": 1163}, "tail": {"text": "triangulation and LAGC algorithms", "start": 1061, "end": 1094}}, {"head": {"text": "visual quality", "start": 1149, "end": 1163}, "tail": {"text": "state-of-the-art solutions", "start": 1106, "end": 1132}}], "compare": [{"head": {"text": "triangulation and LAGC algorithms", "start": 1061, "end": 1094}, "tail": {"text": "state-of-the-art solutions", "start": 1106, "end": 1132}}]}}, "schema": []}
{"input": "This paper presents a phrase-based statistical machine translation method, based on non-contiguous phrases, i.e. phrases with gaps. A method for producing such phrases from a word-aligned corpora is proposed. A statistical translation model is also presented that deals such phrases, as well as a training method based on the maximization of translation accuracy, as measured with the NIST evaluation metric. Translations are produced by means of a beam-search decoder. Experimental results are presented, that demonstrate how the proposed method allows to better generalize from the training data.", "output": {"entities": {"method": [{"text": "phrase-based statistical machine translation method", "start": 22, "end": 73}, {"text": "statistical translation model", "start": 211, "end": 240}, {"text": "training method", "start": 297, "end": 312}, {"text": "beam-search decoder", "start": 449, "end": 468}], "material": [{"text": "non-contiguous phrases", "start": 84, "end": 106}, {"text": "word-aligned corpora", "start": 175, "end": 195}], "generic": [{"text": "method", "start": 67, "end": 73}, {"text": "phrases", "start": 99, "end": 106}, {"text": "phrases", "start": 113, "end": 120}, {"text": "method", "start": 134, "end": 140}], "metric": [{"text": "maximization of translation accuracy", "start": 326, "end": 362}, {"text": "NIST evaluation metric", "start": 385, "end": 407}], "other_scientific_term": [{"text": "Translations", "start": 409, "end": 421}]}, "relations": {"used_for": [{"head": {"text": "non-contiguous phrases", "start": 84, "end": 106}, "tail": {"text": "phrase-based statistical machine translation method", "start": 22, "end": 73}}, {"head": {"text": "method", "start": 67, "end": 73}, "tail": {"text": "phrases", "start": 99, "end": 106}}, {"head": {"text": "statistical translation model", "start": 211, "end": 240}, "tail": {"text": "phrases", "start": 113, "end": 120}}, {"head": {"text": "maximization of translation accuracy", "start": 326, "end": 362}, "tail": {"text": "training method", "start": 297, "end": 312}}, {"head": {"text": "beam-search decoder", "start": 449, "end": 468}, "tail": {"text": "Translations", "start": 409, "end": 421}}], "evaluate_for": [{"head": {"text": "word-aligned corpora", "start": 175, "end": 195}, "tail": {"text": "method", "start": 67, "end": 73}}, {"head": {"text": "NIST evaluation metric", "start": 385, "end": 407}, "tail": {"text": "statistical translation model", "start": 211, "end": 240}}]}}, "schema": []}
{"input": "GLOSSER is designed to support reading and learning to read in a foreign language. There are four language pairs currently supported by GLOSSER: English-Bulgarian, English-Estonian, English-Hungarian and French-Dutch. The program is operational on UNIX and Windows'95 platforms, and has undergone a pilot user-study. A demonstration -LRB- in UNIX -RRB- for Applied Natural Language Processing emphasizes components put to novel technical uses in intelligent computer-assisted morphological analysis -LRB- ICALL -RRB-, including disambiguated morphological analysis and lemmatized indexing for an aligned bilingual corpus of word examples.", "output": {"entities": {"method": [{"text": "GLOSSER", "start": 0, "end": 7}, {"text": "GLOSSER", "start": 136, "end": 143}, {"text": "disambiguated morphological analysis", "start": 528, "end": 564}, {"text": "lemmatized indexing", "start": 569, "end": 588}], "task": [{"text": "reading and learning", "start": 31, "end": 51}, {"text": "Applied Natural Language Processing", "start": 357, "end": 392}, {"text": "intelligent computer-assisted morphological analysis -LRB- ICALL -RRB-", "start": 446, "end": 516}], "generic": [{"text": "language pairs", "start": 98, "end": 112}, {"text": "program", "start": 222, "end": 229}, {"text": "components", "start": 404, "end": 414}], "material": [{"text": "English-Bulgarian", "start": 145, "end": 162}, {"text": "English-Estonian", "start": 164, "end": 180}, {"text": "English-Hungarian", "start": 182, "end": 199}, {"text": "French-Dutch", "start": 204, "end": 216}, {"text": "aligned bilingual corpus", "start": 596, "end": 620}], "other_scientific_term": [{"text": "UNIX and Windows'95 platforms", "start": 248, "end": 277}, {"text": "user-study", "start": 305, "end": 315}]}, "relations": {"used_for": [{"head": {"text": "GLOSSER", "start": 0, "end": 7}, "tail": {"text": "reading and learning", "start": 31, "end": 51}}, {"head": {"text": "language pairs", "start": 98, "end": 112}, "tail": {"text": "GLOSSER", "start": 136, "end": 143}}, {"head": {"text": "components", "start": 404, "end": 414}, "tail": {"text": "intelligent computer-assisted morphological analysis -LRB- ICALL -RRB-", "start": 446, "end": 516}}, {"head": {"text": "disambiguated morphological analysis", "start": 528, "end": 564}, "tail": {"text": "aligned bilingual corpus", "start": 596, "end": 620}}, {"head": {"text": "lemmatized indexing", "start": 569, "end": 588}, "tail": {"text": "aligned bilingual corpus", "start": 596, "end": 620}}], "hyponym_of": [{"head": {"text": "English-Bulgarian", "start": 145, "end": 162}, "tail": {"text": "language pairs", "start": 98, "end": 112}}, {"head": {"text": "English-Estonian", "start": 164, "end": 180}, "tail": {"text": "language pairs", "start": 98, "end": 112}}, {"head": {"text": "English-Hungarian", "start": 182, "end": 199}, "tail": {"text": "language pairs", "start": 98, "end": 112}}, {"head": {"text": "French-Dutch", "start": 204, "end": 216}, "tail": {"text": "language pairs", "start": 98, "end": 112}}, {"head": {"text": "disambiguated morphological analysis", "start": 528, "end": 564}, "tail": {"text": "components", "start": 404, "end": 414}}, {"head": {"text": "lemmatized indexing", "start": 569, "end": 588}, "tail": {"text": "components", "start": 404, "end": 414}}], "conjunction": [{"head": {"text": "English-Bulgarian", "start": 145, "end": 162}, "tail": {"text": "English-Estonian", "start": 164, "end": 180}}, {"head": {"text": "English-Estonian", "start": 164, "end": 180}, "tail": {"text": "English-Hungarian", "start": 182, "end": 199}}, {"head": {"text": "English-Hungarian", "start": 182, "end": 199}, "tail": {"text": "French-Dutch", "start": 204, "end": 216}}, {"head": {"text": "disambiguated morphological analysis", "start": 528, "end": 564}, "tail": {"text": "lemmatized indexing", "start": 569, "end": 588}}]}}, "schema": []}
{"input": "We present a new part-of-speech tagger that demonstrates the following ideas: -LRB- i -RRB- explicit use of both preceding and following tag contexts via a dependency network representation, -LRB- ii -RRB- broad use of lexical features, including jointly conditioning on multiple consecutive words, -LRB- iii -RRB- effective use of priors in conditional loglinear models, and -LRB- iv -RRB- fine-grained modeling of unknown word features. Using these ideas together, the resulting tagger gives a 97.24% accuracy on the Penn Treebank WSJ, an error reduction of 4.4% on the best previous single automatically learned tagging result.", "output": {"entities": {"method": [{"text": "part-of-speech tagger", "start": 17, "end": 38}, {"text": "dependency network representation", "start": 156, "end": 189}, {"text": "fine-grained modeling of unknown word features", "start": 391, "end": 437}, {"text": "tagger", "start": 32, "end": 38}], "other_scientific_term": [{"text": "tag contexts", "start": 137, "end": 149}, {"text": "lexical features", "start": 219, "end": 235}, {"text": "multiple consecutive words", "start": 271, "end": 297}, {"text": "priors in conditional loglinear models", "start": 332, "end": 370}], "metric": [{"text": "accuracy", "start": 503, "end": 511}, {"text": "error", "start": 541, "end": 546}], "material": [{"text": "Penn Treebank WSJ", "start": 519, "end": 536}], "task": [{"text": "tagging", "start": 615, "end": 622}]}, "relations": {"used_for": [{"head": {"text": "tag contexts", "start": 137, "end": 149}, "tail": {"text": "part-of-speech tagger", "start": 17, "end": 38}}, {"head": {"text": "dependency network representation", "start": 156, "end": 189}, "tail": {"text": "tag contexts", "start": 137, "end": 149}}, {"head": {"text": "lexical features", "start": 219, "end": 235}, "tail": {"text": "part-of-speech tagger", "start": 17, "end": 38}}, {"head": {"text": "priors in conditional loglinear models", "start": 332, "end": 370}, "tail": {"text": "part-of-speech tagger", "start": 17, "end": 38}}, {"head": {"text": "fine-grained modeling of unknown word features", "start": 391, "end": 437}, "tail": {"text": "part-of-speech tagger", "start": 17, "end": 38}}], "evaluate_for": [{"head": {"text": "accuracy", "start": 503, "end": 511}, "tail": {"text": "tagger", "start": 32, "end": 38}}, {"head": {"text": "Penn Treebank WSJ", "start": 519, "end": 536}, "tail": {"text": "tagger", "start": 32, "end": 38}}, {"head": {"text": "error", "start": 541, "end": 546}, "tail": {"text": "tagger", "start": 32, "end": 38}}]}}, "schema": []}
{"input": "Person re-identification is challenging due to the large variations of pose, illumination, occlusion and camera view. Owing to these variations, the pedestrian data is distributed as highly-curved manifolds in the feature space, despite the current convolutional neural networks -LRB- CNN -RRB-'s capability of feature extraction. However, the distribution is unknown, so it is difficult to use the geodesic distance when comparing two samples. In practice, the current deep embedding methods use the Euclidean distance for the training and test. On the other hand, the manifold learning methods suggest to use the Euclidean distance in the local range, combining with the graphical relationship between samples, for approximating the geodesic distance. From this point of view, selecting suitable positive -LRB- i.e. intra-class -RRB- training samples within a local range is critical for training the CNN embedding, especially when the data has large intra-class variations. In this paper, we propose a novel moderate positive sample mining method to train robust CNN for person re-identification, dealing with the problem of large variation. In addition, we improve the learning by a metric weight constraint, so that the learned metric has a better generalization ability. Experiments show that these two strategies are effective in learning robust deep metrics for person re-identification, and accordingly our deep model significantly outperforms the state-of-the-art methods on several benchmarks of person re-identification. Therefore, the study presented in this paper may be useful in inspiring new designs of deep models for person re-identification.", "output": {"entities": {"task": [{"text": "Person re-identification", "start": 0, "end": 24}, {"text": "person re-identification", "start": 1074, "end": 1098}, {"text": "person re-identification", "start": 1370, "end": 1394}, {"text": "person re-identification", "start": 1507, "end": 1531}, {"text": "person re-identification", "start": 1636, "end": 1660}], "other_scientific_term": [{"text": "pose", "start": 71, "end": 75}, {"text": "illumination", "start": 77, "end": 89}, {"text": "occlusion", "start": 91, "end": 100}, {"text": "camera view", "start": 105, "end": 116}, {"text": "feature space", "start": 214, "end": 227}, {"text": "feature extraction", "start": 311, "end": 329}, {"text": "geodesic distance", "start": 399, "end": 416}, {"text": "Euclidean distance", "start": 501, "end": 519}, {"text": "Euclidean distance", "start": 615, "end": 633}, {"text": "local range", "start": 641, "end": 652}, {"text": "graphical relationship", "start": 673, "end": 695}, {"text": "geodesic distance", "start": 735, "end": 752}, {"text": "local range", "start": 862, "end": 873}, {"text": "CNN embedding", "start": 903, "end": 916}, {"text": "intra-class variations", "start": 953, "end": 975}, {"text": "metric weight constraint", "start": 1187, "end": 1211}, {"text": "learned metric", "start": 1225, "end": 1239}, {"text": "generalization ability", "start": 1253, "end": 1275}, {"text": "robust deep metrics", "start": 1346, "end": 1365}], "material": [{"text": "pedestrian data", "start": 149, "end": 164}], "method": [{"text": "highly-curved manifolds", "start": 183, "end": 206}, {"text": "convolutional neural networks -LRB- CNN -RRB-", "start": 249, "end": 294}, {"text": "deep embedding methods", "start": 470, "end": 492}, {"text": "manifold learning methods", "start": 570, "end": 595}, {"text": "moderate positive sample mining method", "start": 1011, "end": 1049}, {"text": "robust CNN", "start": 1059, "end": 1069}, {"text": "deep model", "start": 1416, "end": 1426}, {"text": "deep models", "start": 1620, "end": 1631}], "generic": [{"text": "data", "start": 160, "end": 164}, {"text": "learning", "start": 579, "end": 587}, {"text": "state-of-the-art methods", "start": 1457, "end": 1481}]}, "relations": {"used_for": [{"head": {"text": "highly-curved manifolds", "start": 183, "end": 206}, "tail": {"text": "pedestrian data", "start": 149, "end": 164}}, {"head": {"text": "convolutional neural networks -LRB- CNN -RRB-", "start": 249, "end": 294}, "tail": {"text": "feature extraction", "start": 311, "end": 329}}, {"head": {"text": "Euclidean distance", "start": 501, "end": 519}, "tail": {"text": "deep embedding methods", "start": 470, "end": 492}}, {"head": {"text": "Euclidean distance", "start": 615, "end": 633}, "tail": {"text": "manifold learning methods", "start": 570, "end": 595}}, {"head": {"text": "Euclidean distance", "start": 615, "end": 633}, "tail": {"text": "geodesic distance", "start": 735, "end": 752}}, {"head": {"text": "graphical relationship", "start": 673, "end": 695}, "tail": {"text": "geodesic distance", "start": 735, "end": 752}}, {"head": {"text": "moderate positive sample mining method", "start": 1011, "end": 1049}, "tail": {"text": "robust CNN", "start": 1059, "end": 1069}}, {"head": {"text": "robust CNN", "start": 1059, "end": 1069}, "tail": {"text": "person re-identification", "start": 1074, "end": 1098}}, {"head": {"text": "metric weight constraint", "start": 1187, "end": 1211}, "tail": {"text": "learning", "start": 579, "end": 587}}, {"head": {"text": "robust deep metrics", "start": 1346, "end": 1365}, "tail": {"text": "person re-identification", "start": 1370, "end": 1394}}, {"head": {"text": "deep model", "start": 1416, "end": 1426}, "tail": {"text": "person re-identification", "start": 1507, "end": 1531}}, {"head": {"text": "state-of-the-art methods", "start": 1457, "end": 1481}, "tail": {"text": "person re-identification", "start": 1507, "end": 1531}}, {"head": {"text": "deep models", "start": 1620, "end": 1631}, "tail": {"text": "person re-identification", "start": 1636, "end": 1660}}], "feature_of": [{"head": {"text": "feature space", "start": 214, "end": 227}, "tail": {"text": "highly-curved manifolds", "start": 183, "end": 206}}, {"head": {"text": "local range", "start": 641, "end": 652}, "tail": {"text": "Euclidean distance", "start": 615, "end": 633}}, {"head": {"text": "intra-class variations", "start": 953, "end": 975}, "tail": {"text": "data", "start": 160, "end": 164}}, {"head": {"text": "generalization ability", "start": 1253, "end": 1275}, "tail": {"text": "learned metric", "start": 1225, "end": 1239}}], "conjunction": [{"head": {"text": "Euclidean distance", "start": 615, "end": 633}, "tail": {"text": "graphical relationship", "start": 673, "end": 695}}], "compare": [{"head": {"text": "deep model", "start": 1416, "end": 1426}, "tail": {"text": "state-of-the-art methods", "start": 1457, "end": 1481}}]}}, "schema": []}
{"input": "Utterance Verification -LRB- UV -RRB- is a critical function of an Automatic Speech Recognition -LRB- ASR -RRB- System working on real applications where spontaneous speech, out-of-vocabulary -LRB- OOV -RRB- words and acoustic noises are present. In this paper we present a new UV procedure with two major features: a -RRB- Confidence tests are applied to decoded string hypotheses obtained from using word and garbage models that represent OOV words and noises. Thus the ASR system is designed to deal with what we refer to as Word Spotting and Noise Spotting capabilities. b -RRB- The UV procedure is based on three different confidence tests, two based on acoustic measures and one founded on linguistic information, applied in a hierarchical structure. Experimental results from a real telephone application on a natural number recognition task show an 50% reduction in recognition errors with a moderate 12% rejection rate of correct utterances and a low 1.5% rate of false acceptance.", "output": {"entities": {"method": [{"text": "Utterance Verification -LRB- UV -RRB-", "start": 0, "end": 37}, {"text": "Automatic Speech Recognition -LRB- ASR -RRB- System", "start": 67, "end": 118}, {"text": "UV procedure", "start": 278, "end": 290}, {"text": "Confidence tests", "start": 324, "end": 340}, {"text": "ASR system", "start": 472, "end": 482}, {"text": "UV procedure", "start": 587, "end": 599}, {"text": "confidence tests", "start": 628, "end": 644}], "other_scientific_term": [{"text": "spontaneous speech", "start": 154, "end": 172}, {"text": "out-of-vocabulary -LRB- OOV -RRB- words", "start": 174, "end": 213}, {"text": "acoustic noises", "start": 218, "end": 233}, {"text": "decoded string hypotheses", "start": 356, "end": 381}, {"text": "OOV words", "start": 441, "end": 450}, {"text": "noises", "start": 227, "end": 233}, {"text": "linguistic information", "start": 696, "end": 718}, {"text": "hierarchical structure", "start": 733, "end": 755}], "task": [{"text": "Word Spotting", "start": 528, "end": 541}, {"text": "Noise Spotting capabilities", "start": 546, "end": 573}, {"text": "telephone application", "start": 790, "end": 811}, {"text": "natural number recognition task", "start": 817, "end": 848}], "generic": [{"text": "two", "start": 296, "end": 299}, {"text": "one", "start": 681, "end": 684}], "metric": [{"text": "acoustic measures", "start": 659, "end": 676}, {"text": "recognition errors", "start": 874, "end": 892}, {"text": "rejection rate", "start": 913, "end": 927}, {"text": "false acceptance", "start": 973, "end": 989}]}, "relations": {"hyponym_of": [{"head": {"text": "Utterance Verification -LRB- UV -RRB-", "start": 0, "end": 37}, "tail": {"text": "Automatic Speech Recognition -LRB- ASR -RRB- System", "start": 67, "end": 118}}, {"head": {"text": "two", "start": 296, "end": 299}, "tail": {"text": "confidence tests", "start": 628, "end": 644}}, {"head": {"text": "one", "start": 681, "end": 684}, "tail": {"text": "confidence tests", "start": 628, "end": 644}}], "used_for": [{"head": {"text": "Confidence tests", "start": 324, "end": 340}, "tail": {"text": "decoded string hypotheses", "start": 356, "end": 381}}, {"head": {"text": "ASR system", "start": 472, "end": 482}, "tail": {"text": "Word Spotting", "start": 528, "end": 541}}, {"head": {"text": "ASR system", "start": 472, "end": 482}, "tail": {"text": "Noise Spotting capabilities", "start": 546, "end": 573}}, {"head": {"text": "confidence tests", "start": 628, "end": 644}, "tail": {"text": "UV procedure", "start": 587, "end": 599}}, {"head": {"text": "confidence tests", "start": 628, "end": 644}, "tail": {"text": "hierarchical structure", "start": 733, "end": 755}}, {"head": {"text": "acoustic measures", "start": 659, "end": 676}, "tail": {"text": "two", "start": 296, "end": 299}}, {"head": {"text": "linguistic information", "start": 696, "end": 718}, "tail": {"text": "one", "start": 681, "end": 684}}], "conjunction": [{"head": {"text": "noises", "start": 227, "end": 233}, "tail": {"text": "OOV words", "start": 441, "end": 450}}], "feature_of": [{"head": {"text": "natural number recognition task", "start": 817, "end": 848}, "tail": {"text": "telephone application", "start": 790, "end": 811}}], "evaluate_for": [{"head": {"text": "recognition errors", "start": 874, "end": 892}, "tail": {"text": "natural number recognition task", "start": 817, "end": 848}}]}}, "schema": []}
{"input": "A critical step in encoding sound for neuronal processing occurs when the analog pressure wave is coded into discrete nerve-action potentials. Recent pool models of the inner hair cell synapse do not reproduce the dead time period after an intense stimulus, so we used visual inspection and automatic speech recognition -LRB- ASR -RRB- to investigate an offset adaptation -LRB- OA -RRB- model proposed by Zhang et al. -LSB- 1 -RSB-. OA improved phase locking in the auditory nerve -LRB- AN -RRB- and raised ASR accuracy for features derived from AN fibers -LRB- ANFs -RRB-. We also found that OA is crucial for auditory processing by onset neurons -LRB- ONs -RRB- in the next neuronal stage, the auditory brainstem. Multi-layer perceptrons -LRB- MLPs -RRB- performed much better than standard Gaussian mixture models -LRB- GMMs -RRB- for both our ANF-based and ON-based auditory features. Similar results were previously obtained with MSG -LRB- Modulation-filtered Spec-troGram -RRB- auditory features -LSB- 2 -RSB-. Thus we believe researchers working with novel features should consider trying MLPs.", "output": {"entities": {"task": [{"text": "encoding sound", "start": 19, "end": 33}, {"text": "neuronal processing", "start": 38, "end": 57}, {"text": "phase locking in the auditory nerve -LRB- AN -RRB-", "start": 445, "end": 495}, {"text": "auditory processing", "start": 611, "end": 630}], "other_scientific_term": [{"text": "analog pressure wave", "start": 74, "end": 94}, {"text": "discrete nerve-action potentials", "start": 109, "end": 141}, {"text": "inner hair cell synapse", "start": 169, "end": 192}, {"text": "features", "start": 524, "end": 532}, {"text": "AN fibers -LRB- ANFs -RRB-", "start": 546, "end": 572}, {"text": "onset neurons -LRB- ONs -RRB-", "start": 634, "end": 663}, {"text": "auditory brainstem", "start": 696, "end": 714}, {"text": "ANF-based and ON-based auditory features", "start": 847, "end": 887}, {"text": "MSG -LRB- Modulation-filtered Spec-troGram -RRB- auditory features", "start": 935, "end": 1001}, {"text": "features", "start": 879, "end": 887}], "method": [{"text": "pool models", "start": 150, "end": 161}, {"text": "visual inspection", "start": 269, "end": 286}, {"text": "automatic speech recognition -LRB- ASR -RRB-", "start": 291, "end": 335}, {"text": "offset adaptation -LRB- OA -RRB- model", "start": 354, "end": 392}, {"text": "OA", "start": 378, "end": 380}, {"text": "OA", "start": 433, "end": 435}, {"text": "Multi-layer perceptrons -LRB- MLPs -RRB-", "start": 716, "end": 756}, {"text": "Gaussian mixture models -LRB- GMMs -RRB-", "start": 793, "end": 833}, {"text": "MLPs", "start": 746, "end": 750}], "metric": [{"text": "ASR accuracy", "start": 507, "end": 519}]}, "relations": {"used_for": [{"head": {"text": "encoding sound", "start": 19, "end": 33}, "tail": {"text": "neuronal processing", "start": 38, "end": 57}}, {"head": {"text": "discrete nerve-action potentials", "start": 109, "end": 141}, "tail": {"text": "analog pressure wave", "start": 74, "end": 94}}, {"head": {"text": "pool models", "start": 150, "end": 161}, "tail": {"text": "inner hair cell synapse", "start": 169, "end": 192}}, {"head": {"text": "visual inspection", "start": 269, "end": 286}, "tail": {"text": "offset adaptation -LRB- OA -RRB- model", "start": 354, "end": 392}}, {"head": {"text": "automatic speech recognition -LRB- ASR -RRB-", "start": 291, "end": 335}, "tail": {"text": "offset adaptation -LRB- OA -RRB- model", "start": 354, "end": 392}}, {"head": {"text": "OA", "start": 378, "end": 380}, "tail": {"text": "phase locking in the auditory nerve -LRB- AN -RRB-", "start": 445, "end": 495}}, {"head": {"text": "OA", "start": 378, "end": 380}, "tail": {"text": "features", "start": 524, "end": 532}}, {"head": {"text": "AN fibers -LRB- ANFs -RRB-", "start": 546, "end": 572}, "tail": {"text": "features", "start": 524, "end": 532}}, {"head": {"text": "OA", "start": 433, "end": 435}, "tail": {"text": "auditory processing", "start": 611, "end": 630}}, {"head": {"text": "onset neurons -LRB- ONs -RRB-", "start": 634, "end": 663}, "tail": {"text": "OA", "start": 433, "end": 435}}, {"head": {"text": "Multi-layer perceptrons -LRB- MLPs -RRB-", "start": 716, "end": 756}, "tail": {"text": "ANF-based and ON-based auditory features", "start": 847, "end": 887}}, {"head": {"text": "Gaussian mixture models -LRB- GMMs -RRB-", "start": 793, "end": 833}, "tail": {"text": "ANF-based and ON-based auditory features", "start": 847, "end": 887}}], "conjunction": [{"head": {"text": "visual inspection", "start": 269, "end": 286}, "tail": {"text": "automatic speech recognition -LRB- ASR -RRB-", "start": 291, "end": 335}}], "evaluate_for": [{"head": {"text": "ASR accuracy", "start": 507, "end": 519}, "tail": {"text": "features", "start": 524, "end": 532}}], "compare": [{"head": {"text": "Multi-layer perceptrons -LRB- MLPs -RRB-", "start": 716, "end": 756}, "tail": {"text": "Gaussian mixture models -LRB- GMMs -RRB-", "start": 793, "end": 833}}]}}, "schema": []}
{"input": "Recent progress in computer vision has been driven by high-capacity models trained on large datasets. Unfortunately, creating large datasets with pixel-level labels has been extremely costly due to the amount of human effort required. In this paper, we present an approach to rapidly creating pixel-accurate semantic label maps for images extracted from modern computer games.", "output": {"entities": {"task": [{"text": "computer vision", "start": 19, "end": 34}], "method": [{"text": "high-capacity models", "start": 54, "end": 74}], "material": [{"text": "large datasets", "start": 86, "end": 100}, {"text": "large datasets", "start": 126, "end": 140}, {"text": "images", "start": 332, "end": 338}], "other_scientific_term": [{"text": "pixel-level labels", "start": 146, "end": 164}, {"text": "pixel-accurate semantic label maps", "start": 293, "end": 327}, {"text": "modern computer games", "start": 354, "end": 375}], "generic": [{"text": "approach", "start": 264, "end": 272}]}, "relations": {"used_for": [{"head": {"text": "high-capacity models", "start": 54, "end": 74}, "tail": {"text": "computer vision", "start": 19, "end": 34}}, {"head": {"text": "large datasets", "start": 86, "end": 100}, "tail": {"text": "high-capacity models", "start": 54, "end": 74}}, {"head": {"text": "approach", "start": 264, "end": 272}, "tail": {"text": "pixel-accurate semantic label maps", "start": 293, "end": 327}}, {"head": {"text": "pixel-accurate semantic label maps", "start": 293, "end": 327}, "tail": {"text": "images", "start": 332, "end": 338}}], "feature_of": [{"head": {"text": "pixel-level labels", "start": 146, "end": 164}, "tail": {"text": "large datasets", "start": 126, "end": 140}}], "part_of": [{"head": {"text": "images", "start": 332, "end": 338}, "tail": {"text": "modern computer games", "start": 354, "end": 375}}]}}, "schema": []}
{"input": "We propose a novel step toward the unsupervised seg-mentation of whole objects by combining'' hints'' of partial scene segmentation offered by multiple soft, binary mattes. These mattes are implied by a set of hypothesized object boundary fragments in the scene. Rather than trying to find or define a single'' best'' segmentation, we generate multiple segmentations of an image. This reflects contemporary methods for unsupervised object discovery from groups of images, and it allows us to define intuitive evaluation met-rics for our sets of segmentations based on the accurate and parsimonious delineation of scene objects. Our proposed approach builds on recent advances in spectral clustering, image matting, and boundary detection. It is demonstrated qualitatively and quantitatively on a dataset of scenes and is suitable for current work in unsupervised object discovery without top-down knowledge.", "output": {"entities": {"task": [{"text": "unsupervised seg-mentation of whole objects", "start": 35, "end": 78}, {"text": "partial scene segmentation", "start": 105, "end": 131}, {"text": "unsupervised object discovery", "start": 419, "end": 448}, {"text": "unsupervised object discovery", "start": 850, "end": 879}], "other_scientific_term": [{"text": "soft, binary mattes", "start": 152, "end": 171}, {"text": "hypothesized object boundary fragments", "start": 210, "end": 248}, {"text": "delineation of scene objects", "start": 598, "end": 626}], "generic": [{"text": "mattes", "start": 165, "end": 171}, {"text": "contemporary methods", "start": 394, "end": 414}, {"text": "approach", "start": 641, "end": 649}, {"text": "It", "start": 739, "end": 741}], "method": [{"text": "spectral clustering", "start": 679, "end": 698}, {"text": "image matting", "start": 700, "end": 713}, {"text": "boundary detection", "start": 719, "end": 737}], "material": [{"text": "dataset of scenes", "start": 796, "end": 813}]}, "relations": {"used_for": [{"head": {"text": "partial scene segmentation", "start": 105, "end": 131}, "tail": {"text": "unsupervised seg-mentation of whole objects", "start": 35, "end": 78}}, {"head": {"text": "soft, binary mattes", "start": 152, "end": 171}, "tail": {"text": "partial scene segmentation", "start": 105, "end": 131}}, {"head": {"text": "hypothesized object boundary fragments", "start": 210, "end": 248}, "tail": {"text": "mattes", "start": 165, "end": 171}}, {"head": {"text": "contemporary methods", "start": 394, "end": 414}, "tail": {"text": "unsupervised object discovery", "start": 419, "end": 448}}, {"head": {"text": "spectral clustering", "start": 679, "end": 698}, "tail": {"text": "approach", "start": 641, "end": 649}}, {"head": {"text": "image matting", "start": 700, "end": 713}, "tail": {"text": "approach", "start": 641, "end": 649}}, {"head": {"text": "boundary detection", "start": 719, "end": 737}, "tail": {"text": "approach", "start": 641, "end": 649}}, {"head": {"text": "It", "start": 739, "end": 741}, "tail": {"text": "unsupervised object discovery", "start": 850, "end": 879}}], "conjunction": [{"head": {"text": "spectral clustering", "start": 679, "end": 698}, "tail": {"text": "image matting", "start": 700, "end": 713}}, {"head": {"text": "image matting", "start": 700, "end": 713}, "tail": {"text": "boundary detection", "start": 719, "end": 737}}], "evaluate_for": [{"head": {"text": "dataset of scenes", "start": 796, "end": 813}, "tail": {"text": "It", "start": 739, "end": 741}}]}}, "schema": []}
{"input": "Language resource quality is crucial in NLP. Many of the resources used are derived from data created by human beings out of an NLP context, especially regarding MT and reference translations. Indeed, automatic evaluations need high-quality data that allow the comparison of both automatic and human translations. The validation of these resources is widely recommended before being used. This paper describes the impact of using different-quality references on evaluation. Surprisingly enough, similar scores are obtained in many cases regardless of the quality. Thus, the limitations of the automatic metrics used within MT are also discussed in this regard.", "output": {"entities": {"metric": [{"text": "Language resource quality", "start": 0, "end": 25}, {"text": "automatic metrics", "start": 593, "end": 610}], "task": [{"text": "NLP", "start": 40, "end": 43}, {"text": "NLP", "start": 128, "end": 131}, {"text": "MT", "start": 162, "end": 164}, {"text": "reference translations", "start": 169, "end": 191}, {"text": "automatic evaluations", "start": 201, "end": 222}, {"text": "MT", "start": 623, "end": 625}], "material": [{"text": "high-quality data", "start": 228, "end": 245}], "generic": [{"text": "resources", "start": 57, "end": 66}, {"text": "evaluation", "start": 211, "end": 221}], "other_scientific_term": [{"text": "different-quality references", "start": 430, "end": 458}]}, "relations": {"feature_of": [{"head": {"text": "Language resource quality", "start": 0, "end": 25}, "tail": {"text": "NLP", "start": 40, "end": 43}}], "hyponym_of": [{"head": {"text": "MT", "start": 162, "end": 164}, "tail": {"text": "NLP", "start": 128, "end": 131}}, {"head": {"text": "reference translations", "start": 169, "end": 191}, "tail": {"text": "NLP", "start": 128, "end": 131}}], "conjunction": [{"head": {"text": "MT", "start": 162, "end": 164}, "tail": {"text": "reference translations", "start": 169, "end": 191}}], "evaluate_for": [{"head": {"text": "high-quality data", "start": 228, "end": 245}, "tail": {"text": "automatic evaluations", "start": 201, "end": 222}}, {"head": {"text": "automatic metrics", "start": 593, "end": 610}, "tail": {"text": "MT", "start": 623, "end": 625}}], "used_for": [{"head": {"text": "different-quality references", "start": 430, "end": 458}, "tail": {"text": "evaluation", "start": 211, "end": 221}}]}}, "schema": []}
{"input": "This poster paper describes a full scale two-level morphological description -LRB- Karttunen, 1983; Koskenniemi, 1983 -RRB- of Turkish word structures. The description has been implemented using the PC-KIMMO environment -LRB- Antworth, 1990 -RRB- and is based on a root word lexicon of about 23,000 roots words. Almost all the special cases of and exceptions to phonological and morphological rules have been implemented. Turkish is an agglutinative language with word structures formed by productive affixations of derivational and inflectional suffixes to root words. Turkish has finite-state but nevertheless rather complex morphotactics. Morphemes added to a root word or a stem can convert the word from a nominal to a verbal structure or vice-versa, or can create adverbial constructs. The surface realizations of morphological constructions are constrained and modified by a number of phonetic rules such as vowel harmony.", "output": {"entities": {"task": [{"text": "full scale two-level morphological description", "start": 30, "end": 76}, {"text": "surface realizations of morphological constructions", "start": 796, "end": 847}], "material": [{"text": "Turkish word structures", "start": 127, "end": 150}, {"text": "root word lexicon", "start": 265, "end": 282}, {"text": "Turkish", "start": 127, "end": 134}, {"text": "agglutinative language", "start": 436, "end": 458}, {"text": "Turkish", "start": 422, "end": 429}], "generic": [{"text": "description", "start": 65, "end": 76}], "method": [{"text": "PC-KIMMO environment", "start": 199, "end": 219}], "other_scientific_term": [{"text": "phonological and morphological rules", "start": 362, "end": 398}, {"text": "word structures", "start": 135, "end": 150}, {"text": "productive affixations of derivational and inflectional suffixes", "start": 490, "end": 554}, {"text": "finite-state", "start": 582, "end": 594}, {"text": "Morphemes", "start": 642, "end": 651}, {"text": "verbal structure", "start": 724, "end": 740}, {"text": "adverbial constructs", "start": 770, "end": 790}, {"text": "phonetic rules", "start": 892, "end": 906}, {"text": "vowel harmony", "start": 915, "end": 928}]}, "relations": {"used_for": [{"head": {"text": "full scale two-level morphological description", "start": 30, "end": 76}, "tail": {"text": "Turkish word structures", "start": 127, "end": 150}}, {"head": {"text": "PC-KIMMO environment", "start": 199, "end": 219}, "tail": {"text": "description", "start": 65, "end": 76}}, {"head": {"text": "root word lexicon", "start": 265, "end": 282}, "tail": {"text": "description", "start": 65, "end": 76}}, {"head": {"text": "phonetic rules", "start": 892, "end": 906}, "tail": {"text": "surface realizations of morphological constructions", "start": 796, "end": 847}}], "hyponym_of": [{"head": {"text": "Turkish", "start": 127, "end": 134}, "tail": {"text": "agglutinative language", "start": 436, "end": 458}}, {"head": {"text": "vowel harmony", "start": 915, "end": 928}, "tail": {"text": "phonetic rules", "start": 892, "end": 906}}], "feature_of": [{"head": {"text": "word structures", "start": 135, "end": 150}, "tail": {"text": "agglutinative language", "start": 436, "end": 458}}], "part_of": [{"head": {"text": "productive affixations of derivational and inflectional suffixes", "start": 490, "end": 554}, "tail": {"text": "word structures", "start": 135, "end": 150}}]}}, "schema": []}
{"input": "This paper deals with the problem of generating the fundamental frequency -LRB- F0 -RRB- contour of speech from a text input for text-to-speech synthesis. We have previously introduced a statistical model describing the generating process of speech F0 contours, based on the discrete-time version of the Fujisaki model. One remarkable feature of this model is that it has allowed us to derive an efficient algorithm based on powerful statistical methods for estimating the Fujisaki-model parameters from raw F0 contours. To associate a sequence of the Fujisaki-model parameters with a text input based on statistical learning, this paper proposes extending this model to a context-dependent one. We further propose a parameter training algorithm for the present model based on a decision tree-based context clustering.", "output": {"entities": {"other_scientific_term": [{"text": "fundamental frequency -LRB- F0 -RRB- contour of speech", "start": 52, "end": 106}, {"text": "speech F0 contours", "start": 242, "end": 260}, {"text": "remarkable feature", "start": 324, "end": 342}, {"text": "Fujisaki-model parameters", "start": 473, "end": 498}, {"text": "raw F0 contours", "start": 504, "end": 519}, {"text": "Fujisaki-model parameters", "start": 552, "end": 577}], "material": [{"text": "text input", "start": 114, "end": 124}, {"text": "text input", "start": 585, "end": 595}], "task": [{"text": "text-to-speech synthesis", "start": 129, "end": 153}], "method": [{"text": "statistical model", "start": 187, "end": 204}, {"text": "Fujisaki model", "start": 304, "end": 318}, {"text": "statistical methods", "start": 434, "end": 453}, {"text": "statistical learning", "start": 605, "end": 625}, {"text": "parameter training algorithm", "start": 717, "end": 745}, {"text": "decision tree-based context clustering", "start": 779, "end": 817}], "generic": [{"text": "model", "start": 199, "end": 204}, {"text": "it", "start": 18, "end": 20}, {"text": "algorithm", "start": 406, "end": 415}, {"text": "model", "start": 313, "end": 318}, {"text": "model", "start": 351, "end": 356}]}, "relations": {"used_for": [{"head": {"text": "fundamental frequency -LRB- F0 -RRB- contour of speech", "start": 52, "end": 106}, "tail": {"text": "text-to-speech synthesis", "start": 129, "end": 153}}, {"head": {"text": "text input", "start": 114, "end": 124}, "tail": {"text": "fundamental frequency -LRB- F0 -RRB- contour of speech", "start": 52, "end": 106}}, {"head": {"text": "statistical model", "start": 187, "end": 204}, "tail": {"text": "speech F0 contours", "start": 242, "end": 260}}, {"head": {"text": "Fujisaki model", "start": 304, "end": 318}, "tail": {"text": "statistical model", "start": 187, "end": 204}}, {"head": {"text": "algorithm", "start": 406, "end": 415}, "tail": {"text": "Fujisaki-model parameters", "start": 473, "end": 498}}, {"head": {"text": "statistical methods", "start": 434, "end": 453}, "tail": {"text": "algorithm", "start": 406, "end": 415}}, {"head": {"text": "raw F0 contours", "start": 504, "end": 519}, "tail": {"text": "Fujisaki-model parameters", "start": 473, "end": 498}}, {"head": {"text": "text input", "start": 585, "end": 595}, "tail": {"text": "Fujisaki-model parameters", "start": 552, "end": 577}}, {"head": {"text": "statistical learning", "start": 605, "end": 625}, "tail": {"text": "Fujisaki-model parameters", "start": 552, "end": 577}}, {"head": {"text": "parameter training algorithm", "start": 717, "end": 745}, "tail": {"text": "model", "start": 351, "end": 356}}, {"head": {"text": "decision tree-based context clustering", "start": 779, "end": 817}, "tail": {"text": "parameter training algorithm", "start": 717, "end": 745}}], "feature_of": [{"head": {"text": "remarkable feature", "start": 324, "end": 342}, "tail": {"text": "model", "start": 199, "end": 204}}]}}, "schema": []}
{"input": "We introduce a method to accelerate the evaluation of object detection cascades with the help of a divide-and-conquer procedure in the space of candidate regions. Compared to the exhaustive procedure that thus far is the state-of-the-art for cascade evaluation, the proposed method requires fewer evaluations of the classifier functions, thereby speeding up the search. Furthermore, we show how the recently developed efficient subwindow search -LRB- ESS -RRB- procedure -LSB- 11 -RSB- can be integrated into the last stage of our method. This allows us to use our method to act not only as a faster procedure for cascade evaluation, but also as a tool to perform efficient branch-and-bound object detection with nonlinear quality functions, in particular kernel-ized support vector machines. Experiments on the PASCAL VOC 2006 dataset show an acceleration of more than 50% by our method compared to standard cascade evaluation.", "output": {"entities": {"generic": [{"text": "method", "start": 15, "end": 21}, {"text": "method", "start": 275, "end": 281}, {"text": "method", "start": 531, "end": 537}, {"text": "method", "start": 565, "end": 571}, {"text": "method", "start": 881, "end": 887}], "task": [{"text": "evaluation of object detection cascades", "start": 40, "end": 79}, {"text": "cascade evaluation", "start": 242, "end": 260}, {"text": "search", "start": 362, "end": 368}, {"text": "cascade evaluation", "start": 614, "end": 632}, {"text": "branch-and-bound object detection", "start": 674, "end": 707}], "method": [{"text": "divide-and-conquer procedure", "start": 99, "end": 127}, {"text": "space of candidate regions", "start": 135, "end": 161}, {"text": "exhaustive procedure", "start": 179, "end": 199}, {"text": "subwindow search -LRB- ESS -RRB- procedure", "start": 428, "end": 470}, {"text": "kernel-ized support vector machines", "start": 756, "end": 791}, {"text": "cascade evaluation", "start": 909, "end": 927}], "other_scientific_term": [{"text": "classifier functions", "start": 316, "end": 336}, {"text": "nonlinear quality functions", "start": 713, "end": 740}], "material": [{"text": "PASCAL VOC 2006 dataset", "start": 812, "end": 835}]}, "relations": {"used_for": [{"head": {"text": "method", "start": 15, "end": 21}, "tail": {"text": "evaluation of object detection cascades", "start": 40, "end": 79}}, {"head": {"text": "divide-and-conquer procedure", "start": 99, "end": 127}, "tail": {"text": "method", "start": 15, "end": 21}}, {"head": {"text": "exhaustive procedure", "start": 179, "end": 199}, "tail": {"text": "cascade evaluation", "start": 242, "end": 260}}, {"head": {"text": "method", "start": 275, "end": 281}, "tail": {"text": "search", "start": 362, "end": 368}}, {"head": {"text": "method", "start": 565, "end": 571}, "tail": {"text": "cascade evaluation", "start": 614, "end": 632}}, {"head": {"text": "method", "start": 565, "end": 571}, "tail": {"text": "branch-and-bound object detection", "start": 674, "end": 707}}, {"head": {"text": "nonlinear quality functions", "start": 713, "end": 740}, "tail": {"text": "branch-and-bound object detection", "start": 674, "end": 707}}], "feature_of": [{"head": {"text": "space of candidate regions", "start": 135, "end": 161}, "tail": {"text": "divide-and-conquer procedure", "start": 99, "end": 127}}], "compare": [{"head": {"text": "exhaustive procedure", "start": 179, "end": 199}, "tail": {"text": "method", "start": 275, "end": 281}}, {"head": {"text": "cascade evaluation", "start": 909, "end": 927}, "tail": {"text": "method", "start": 881, "end": 887}}], "part_of": [{"head": {"text": "subwindow search -LRB- ESS -RRB- procedure", "start": 428, "end": 470}, "tail": {"text": "method", "start": 531, "end": 537}}], "hyponym_of": [{"head": {"text": "kernel-ized support vector machines", "start": 756, "end": 791}, "tail": {"text": "nonlinear quality functions", "start": 713, "end": 740}}], "evaluate_for": [{"head": {"text": "PASCAL VOC 2006 dataset", "start": 812, "end": 835}, "tail": {"text": "method", "start": 881, "end": 887}}, {"head": {"text": "PASCAL VOC 2006 dataset", "start": 812, "end": 835}, "tail": {"text": "cascade evaluation", "start": 909, "end": 927}}]}}, "schema": []}
{"input": "Background modeling is an important component of many vision systems. Existing work in the area has mostly addressed scenes that consist of static or quasi-static structures. When the scene exhibits a persistent dynamic behavior in time, such an assumption is violated and detection performance deteriorates. In this paper, we propose a new method for the modeling and subtraction of such scenes. Towards the modeling of the dynamic characteristics, optical flow is computed and utilized as a feature in a higher dimensional space. Inherent ambiguities in the computation of features are addressed by using a data-dependent bandwidth for density estimation using kernels. Extensive experiments demonstrate the utility and performance of the proposed approach.", "output": {"entities": {"task": [{"text": "Background modeling", "start": 0, "end": 19}, {"text": "vision systems", "start": 54, "end": 68}, {"text": "detection", "start": 273, "end": 282}, {"text": "modeling and subtraction of such scenes", "start": 356, "end": 395}, {"text": "modeling of the dynamic characteristics", "start": 409, "end": 448}, {"text": "computation of features", "start": 560, "end": 583}, {"text": "density estimation", "start": 638, "end": 656}], "other_scientific_term": [{"text": "static or quasi-static structures", "start": 140, "end": 173}, {"text": "persistent dynamic behavior", "start": 201, "end": 228}, {"text": "optical flow", "start": 450, "end": 462}, {"text": "feature", "start": 493, "end": 500}, {"text": "higher dimensional space", "start": 506, "end": 530}, {"text": "ambiguities", "start": 541, "end": 552}, {"text": "data-dependent bandwidth", "start": 609, "end": 633}], "generic": [{"text": "scene", "start": 117, "end": 122}, {"text": "method", "start": 341, "end": 347}, {"text": "scenes", "start": 117, "end": 123}, {"text": "approach", "start": 750, "end": 758}], "method": [{"text": "kernels", "start": 663, "end": 670}]}, "relations": {"part_of": [{"head": {"text": "Background modeling", "start": 0, "end": 19}, "tail": {"text": "vision systems", "start": 54, "end": 68}}], "feature_of": [{"head": {"text": "persistent dynamic behavior", "start": 201, "end": 228}, "tail": {"text": "scene", "start": 117, "end": 122}}, {"head": {"text": "higher dimensional space", "start": 506, "end": 530}, "tail": {"text": "feature", "start": 493, "end": 500}}, {"head": {"text": "ambiguities", "start": 541, "end": 552}, "tail": {"text": "computation of features", "start": 560, "end": 583}}], "used_for": [{"head": {"text": "method", "start": 341, "end": 347}, "tail": {"text": "modeling and subtraction of such scenes", "start": 356, "end": 395}}, {"head": {"text": "optical flow", "start": 450, "end": 462}, "tail": {"text": "modeling of the dynamic characteristics", "start": 409, "end": 448}}, {"head": {"text": "optical flow", "start": 450, "end": 462}, "tail": {"text": "feature", "start": 493, "end": 500}}, {"head": {"text": "feature", "start": 493, "end": 500}, "tail": {"text": "modeling of the dynamic characteristics", "start": 409, "end": 448}}, {"head": {"text": "data-dependent bandwidth", "start": 609, "end": 633}, "tail": {"text": "ambiguities", "start": 541, "end": 552}}, {"head": {"text": "data-dependent bandwidth", "start": 609, "end": 633}, "tail": {"text": "density estimation", "start": 638, "end": 656}}, {"head": {"text": "kernels", "start": 663, "end": 670}, "tail": {"text": "density estimation", "start": 638, "end": 656}}]}}, "schema": []}
{"input": "Information distillation aims to extract relevant pieces of information related to a given query from massive, possibly multilingual, audio and textual document sources. In this paper, we present our approach for using information extraction annotations to augment document retrieval for distillation. We take advantage of the fact that some of the distillation queries can be associated with annotation elements introduced for the NIST Automatic Content Extraction -LRB- ACE -RRB- task. We experimentally show that using the ACE events to constrain the document set returned by an information retrieval engine significantly improves the precision at various recall rates for two different query templates.", "output": {"entities": {"task": [{"text": "Information distillation", "start": 0, "end": 24}, {"text": "document retrieval for distillation", "start": 265, "end": 300}, {"text": "NIST Automatic Content Extraction -LRB- ACE -RRB- task", "start": 432, "end": 486}], "material": [{"text": "massive, possibly multilingual, audio and textual document sources", "start": 102, "end": 168}], "other_scientific_term": [{"text": "information extraction annotations", "start": 219, "end": 253}, {"text": "distillation queries", "start": 349, "end": 369}, {"text": "annotation elements", "start": 393, "end": 412}, {"text": "ACE events", "start": 526, "end": 536}], "method": [{"text": "information retrieval engine", "start": 582, "end": 610}], "metric": [{"text": "precision", "start": 638, "end": 647}, {"text": "recall rates", "start": 659, "end": 671}]}, "relations": {"used_for": [{"head": {"text": "information extraction annotations", "start": 219, "end": 253}, "tail": {"text": "document retrieval for distillation", "start": 265, "end": 300}}]}}, "schema": []}
{"input": "This paper presents a novel representation for three-dimensional objects in terms of affine-invariant image patches and their spatial relationships. Multi-view constraints associated with groups of patches are combined with a normalized representation of their appearance to guide matching and reconstruction, allowing the acquisition of true three-dimensional affine and Euclidean models from multiple images and their recognition in a single photograph taken from an arbitrary viewpoint. The proposed approach does not require a separate segmentation stage and is applicable to cluttered scenes. Preliminary modeling and recognition results are presented.", "output": {"entities": {"generic": [{"text": "representation", "start": 28, "end": 42}, {"text": "approach", "start": 503, "end": 511}], "other_scientific_term": [{"text": "three-dimensional objects", "start": 47, "end": 72}, {"text": "affine-invariant image patches", "start": 85, "end": 115}, {"text": "spatial relationships", "start": 126, "end": 147}, {"text": "Multi-view constraints", "start": 149, "end": 171}, {"text": "cluttered scenes", "start": 580, "end": 596}], "method": [{"text": "normalized representation", "start": 226, "end": 251}, {"text": "segmentation stage", "start": 540, "end": 558}], "task": [{"text": "matching", "start": 281, "end": 289}, {"text": "reconstruction", "start": 294, "end": 308}, {"text": "acquisition of true three-dimensional affine and Euclidean models", "start": 323, "end": 388}, {"text": "recognition", "start": 420, "end": 431}], "material": [{"text": "images", "start": 403, "end": 409}]}, "relations": {"used_for": [{"head": {"text": "representation", "start": 28, "end": 42}, "tail": {"text": "three-dimensional objects", "start": 47, "end": 72}}, {"head": {"text": "Multi-view constraints", "start": 149, "end": 171}, "tail": {"text": "matching", "start": 281, "end": 289}}, {"head": {"text": "Multi-view constraints", "start": 149, "end": 171}, "tail": {"text": "reconstruction", "start": 294, "end": 308}}, {"head": {"text": "normalized representation", "start": 226, "end": 251}, "tail": {"text": "matching", "start": 281, "end": 289}}, {"head": {"text": "normalized representation", "start": 226, "end": 251}, "tail": {"text": "reconstruction", "start": 294, "end": 308}}, {"head": {"text": "images", "start": 403, "end": 409}, "tail": {"text": "acquisition of true three-dimensional affine and Euclidean models", "start": 323, "end": 388}}, {"head": {"text": "approach", "start": 503, "end": 511}, "tail": {"text": "cluttered scenes", "start": 580, "end": 596}}], "feature_of": [{"head": {"text": "affine-invariant image patches", "start": 85, "end": 115}, "tail": {"text": "three-dimensional objects", "start": 47, "end": 72}}, {"head": {"text": "spatial relationships", "start": 126, "end": 147}, "tail": {"text": "affine-invariant image patches", "start": 85, "end": 115}}], "conjunction": [{"head": {"text": "Multi-view constraints", "start": 149, "end": 171}, "tail": {"text": "normalized representation", "start": 226, "end": 251}}, {"head": {"text": "matching", "start": 281, "end": 289}, "tail": {"text": "reconstruction", "start": 294, "end": 308}}]}}, "schema": []}
{"input": "Fast algorithms for nearest neighbor -LRB- NN -RRB- search have in large part focused on 2 distance. Here we develop an approach for 1 distance that begins with an explicit and exactly distance-preserving embedding of the points into 2 2. We show how this can efficiently be combined with random-projection based methods for 2 NN search, such as locality-sensitive hashing -LRB- LSH -RRB- or random projection trees. We rigorously establish the correctness of the methodology and show by experimentation using LSH that it is competitive in practice with available alternatives.", "output": {"entities": {"generic": [{"text": "Fast algorithms", "start": 0, "end": 15}, {"text": "approach", "start": 120, "end": 128}, {"text": "this", "start": 251, "end": 255}, {"text": "it", "start": 10, "end": 12}, {"text": "alternatives", "start": 564, "end": 576}], "task": [{"text": "nearest neighbor -LRB- NN -RRB- search", "start": 20, "end": 58}, {"text": "NN search", "start": 327, "end": 336}], "other_scientific_term": [{"text": "distance", "start": 91, "end": 99}, {"text": "1 distance", "start": 133, "end": 143}, {"text": "distance-preserving embedding", "start": 185, "end": 214}], "method": [{"text": "random-projection based methods", "start": 289, "end": 320}, {"text": "locality-sensitive hashing -LRB- LSH -RRB-", "start": 346, "end": 388}, {"text": "random projection trees", "start": 392, "end": 415}, {"text": "LSH", "start": 379, "end": 382}]}, "relations": {"used_for": [{"head": {"text": "Fast algorithms", "start": 0, "end": 15}, "tail": {"text": "nearest neighbor -LRB- NN -RRB- search", "start": 20, "end": 58}}, {"head": {"text": "approach", "start": 120, "end": 128}, "tail": {"text": "1 distance", "start": 133, "end": 143}}, {"head": {"text": "random-projection based methods", "start": 289, "end": 320}, "tail": {"text": "NN search", "start": 327, "end": 336}}], "conjunction": [{"head": {"text": "this", "start": 251, "end": 255}, "tail": {"text": "random-projection based methods", "start": 289, "end": 320}}, {"head": {"text": "locality-sensitive hashing -LRB- LSH -RRB-", "start": 346, "end": 388}, "tail": {"text": "random projection trees", "start": 392, "end": 415}}], "hyponym_of": [{"head": {"text": "locality-sensitive hashing -LRB- LSH -RRB-", "start": 346, "end": 388}, "tail": {"text": "random-projection based methods", "start": 289, "end": 320}}, {"head": {"text": "random projection trees", "start": 392, "end": 415}, "tail": {"text": "random-projection based methods", "start": 289, "end": 320}}], "compare": [{"head": {"text": "it", "start": 10, "end": 12}, "tail": {"text": "alternatives", "start": 564, "end": 576}}]}}, "schema": []}