{"input": "This paper presents an algorithm for computing optical flow, shape, motion, lighting, and albedo from an image sequence of a rigidly-moving Lambertian object under distant illumination. The problem is formulated in a manner that subsumes structure from motion, multi-view stereo, and photo-metric stereo as special cases. The algorithm utilizes both spatial and temporal intensity variation as cues: the former constrains flow and the latter constrains surface orientation; combining both cues enables dense reconstruction of both textured and texture-less surfaces. The algorithm works by iteratively estimating affine camera parameters, illumination, shape, and albedo in an alternating fashion. Results are demonstrated on videos of hand-held objects moving in front of a fixed light and camera.", "output": {"entities": {"generic": [{"text": "algorithm", "start": 23, "end": 32}, {"text": "problem", "start": 190, "end": 197}, {"text": "algorithm", "start": 326, "end": 335}, {"text": "cues", "start": 394, "end": 398}, {"text": "former", "start": 404, "end": 410}, {"text": "latter", "start": 435, "end": 441}, {"text": "cues", "start": 489, "end": 493}, {"text": "algorithm", "start": 571, "end": 580}], "task": [{"text": "computing optical flow, shape, motion, lighting, and albedo", "start": 37, "end": 96}, {"text": "dense reconstruction of both textured and texture-less surfaces", "start": 502, "end": 565}], "material": [{"text": "image sequence", "start": 105, "end": 119}, {"text": "rigidly-moving Lambertian object", "start": 125, "end": 157}, {"text": "motion", "start": 68, "end": 74}, {"text": "multi-view stereo", "start": 261, "end": 278}, {"text": "photo-metric stereo", "start": 284, "end": 303}, {"text": "videos of hand-held objects", "start": 726, "end": 753}], "other_scientific_term": [{"text": "distant illumination", "start": 164, "end": 184}, {"text": "spatial and temporal intensity variation", "start": 350, "end": 390}, {"text": "flow", "start": 55, "end": 59}, {"text": "surface orientation", "start": 453, "end": 472}], "method": [{"text": "estimating affine camera parameters, illumination, shape, and albedo", "start": 602, "end": 670}]}, "relations": {"used_for": [{"head": {"text": "algorithm", "start": 23, "end": 32}, "tail": {"text": "computing optical flow, shape, motion, lighting, and albedo", "start": 37, "end": 96}}, {"head": {"text": "image sequence", "start": 105, "end": 119}, "tail": {"text": "algorithm", "start": 23, "end": 32}}, {"head": {"text": "spatial and temporal intensity variation", "start": 350, "end": 390}, "tail": {"text": "algorithm", "start": 326, "end": 335}}, {"head": {"text": "former", "start": 404, "end": 410}, "tail": {"text": "flow", "start": 55, "end": 59}}, {"head": {"text": "latter", "start": 435, "end": 441}, "tail": {"text": "surface orientation", "start": 453, "end": 472}}, {"head": {"text": "cues", "start": 489, "end": 493}, "tail": {"text": "dense reconstruction of both textured and texture-less surfaces", "start": 502, "end": 565}}, {"head": {"text": "estimating affine camera parameters, illumination, shape, and albedo", "start": 602, "end": 670}, "tail": {"text": "algorithm", "start": 571, "end": 580}}], "feature_of": [{"head": {"text": "rigidly-moving Lambertian object", "start": 125, "end": 157}, "tail": {"text": "image sequence", "start": 105, "end": 119}}, {"head": {"text": "distant illumination", "start": 164, "end": 184}, "tail": {"text": "rigidly-moving Lambertian object", "start": 125, "end": 157}}], "conjunction": [{"head": {"text": "motion", "start": 68, "end": 74}, "tail": {"text": "multi-view stereo", "start": 261, "end": 278}}, {"head": {"text": "multi-view stereo", "start": 261, "end": 278}, "tail": {"text": "photo-metric stereo", "start": 284, "end": 303}}, {"head": {"text": "former", "start": 404, "end": 410}, "tail": {"text": "latter", "start": 435, "end": 441}}], "hyponym_of": [{"head": {"text": "former", "start": 404, "end": 410}, "tail": {"text": "cues", "start": 394, "end": 398}}, {"head": {"text": "latter", "start": 435, "end": 441}, "tail": {"text": "cues", "start": 394, "end": 398}}]}}, "schema": []}
{"input": "Past work of generating referring expressions mainly utilized attributes of objects and binary relations between objects. However, such an approach does not work well when there is no distinctive attribute among objects. To overcome this limitation, this paper proposes a method utilizing the perceptual groups of objects and n-ary relations among them. The key is to identify groups of objects that are naturally recognized by humans. We conducted psychological experiments with 42 subjects to collect referring expressions in such situations, and built a generation algorithm based on the results. The evaluation using another 23 subjects showed that the proposed method could effectively generate proper referring expressions.", "output": {"entities": {"other_scientific_term": [{"text": "referring expressions", "start": 24, "end": 45}, {"text": "binary relations", "start": 88, "end": 104}, {"text": "n-ary relations", "start": 326, "end": 341}, {"text": "referring expressions", "start": 503, "end": 524}, {"text": "referring expressions", "start": 707, "end": 728}], "method": [{"text": "generation algorithm", "start": 557, "end": 577}], "generic": [{"text": "method", "start": 272, "end": 278}]}, "relations": {}}, "schema": []}
{"input": "An entity-oriented approach to restricted-domain parsing is proposed. In this approach, the definitions of the structure and surface representation of domain entities are grouped together. Like semantic grammar, this allows easy exploitation of limited domain semantics. In addition, it facilitates fragmentary recognition and the use of multiple parsing strategies, and so is particularly useful for robust recognition of extra-grammatical input. Several advantages from the point of view of language definition are also noted. Representative samples from an entity-oriented language definition are presented, along with a control structure for an entity-oriented parser, some parsing strategies that use the control structure, and worked examples of parses. A parser incorporating the control structure and the parsing strategies is currently under implementation.", "output": {"entities": {"method": [{"text": "entity-oriented approach", "start": 3, "end": 27}, {"text": "semantic grammar", "start": 194, "end": 210}, {"text": "multiple parsing strategies", "start": 338, "end": 365}, {"text": "entity-oriented parser", "start": 649, "end": 671}, {"text": "parsing strategies", "start": 347, "end": 365}, {"text": "parser", "start": 665, "end": 671}, {"text": "parsing strategies", "start": 678, "end": 696}], "task": [{"text": "restricted-domain parsing", "start": 31, "end": 56}, {"text": "fragmentary recognition", "start": 299, "end": 322}], "generic": [{"text": "approach", "start": 19, "end": 27}, {"text": "this", "start": 73, "end": 77}, {"text": "it", "start": 6, "end": 8}], "other_scientific_term": [{"text": "structure and surface representation of domain entities", "start": 111, "end": 166}, {"text": "limited domain semantics", "start": 245, "end": 269}, {"text": "recognition of extra-grammatical input", "start": 408, "end": 446}, {"text": "entity-oriented language definition", "start": 560, "end": 595}, {"text": "control structure", "start": 624, "end": 641}, {"text": "control structure", "start": 710, "end": 727}, {"text": "control structure", "start": 787, "end": 804}]}, "relations": {"used_for": [{"head": {"text": "entity-oriented approach", "start": 3, "end": 27}, "tail": {"text": "restricted-domain parsing", "start": 31, "end": 56}}, {"head": {"text": "this", "start": 73, "end": 77}, "tail": {"text": "limited domain semantics", "start": 245, "end": 269}}, {"head": {"text": "it", "start": 6, "end": 8}, "tail": {"text": "fragmentary recognition", "start": 299, "end": 322}}, {"head": {"text": "it", "start": 6, "end": 8}, "tail": {"text": "multiple parsing strategies", "start": 338, "end": 365}}, {"head": {"text": "multiple parsing strategies", "start": 338, "end": 365}, "tail": {"text": "recognition of extra-grammatical input", "start": 408, "end": 446}}, {"head": {"text": "control structure", "start": 624, "end": 641}, "tail": {"text": "entity-oriented parser", "start": 649, "end": 671}}, {"head": {"text": "control structure", "start": 710, "end": 727}, "tail": {"text": "parsing strategies", "start": 347, "end": 365}}], "part_of": [{"head": {"text": "control structure", "start": 787, "end": 804}, "tail": {"text": "parser", "start": 665, "end": 671}}]}}, "schema": []}
{"input": "This paper summarizes the formalism of Category Cooccurrence Restrictions -LRB- CCRs -RRB- and describes two parsing algorithms that interpret it. CCRs are Boolean conditions on the cooccurrence of categories in local trees which allow the statement of generalizations which can not be captured in other current syntax formalisms. The use of CCRs leads to syntactic descriptions formulated entirely with restrictive statements. The paper shows how conventional algorithms for the analysis of context free languages can be adapted to the CCR formalism. Special attention is given to the part of the parser that checks the fulfillment of logical well-formedness conditions on trees.", "output": {"entities": {"task": [{"text": "formalism of Category Cooccurrence Restrictions -LRB- CCRs -RRB-", "start": 26, "end": 90}, {"text": "CCR formalism", "start": 537, "end": 550}], "other_scientific_term": [{"text": "Category Cooccurrence Restrictions -LRB- CCRs -RRB-", "start": 39, "end": 90}, {"text": "CCRs", "start": 80, "end": 84}, {"text": "Boolean conditions", "start": 156, "end": 174}, {"text": "local trees", "start": 212, "end": 223}, {"text": "statement of generalizations", "start": 240, "end": 268}, {"text": "CCRs", "start": 147, "end": 151}, {"text": "syntactic descriptions", "start": 356, "end": 378}, {"text": "restrictive statements", "start": 404, "end": 426}, {"text": "logical well-formedness conditions", "start": 636, "end": 670}, {"text": "trees", "start": 218, "end": 223}], "method": [{"text": "parsing algorithms", "start": 109, "end": 127}, {"text": "syntax formalisms", "start": 312, "end": 329}, {"text": "parser", "start": 598, "end": 604}], "generic": [{"text": "it", "start": 122, "end": 124}, {"text": "algorithms", "start": 117, "end": 127}], "material": [{"text": "context free languages", "start": 492, "end": 514}]}, "relations": {"used_for": [{"head": {"text": "parsing algorithms", "start": 109, "end": 127}, "tail": {"text": "it", "start": 122, "end": 124}}, {"head": {"text": "algorithms", "start": 117, "end": 127}, "tail": {"text": "CCR formalism", "start": 537, "end": 550}}, {"head": {"text": "context free languages", "start": 492, "end": 514}, "tail": {"text": "algorithms", "start": 117, "end": 127}}], "feature_of": [{"head": {"text": "restrictive statements", "start": 404, "end": 426}, "tail": {"text": "syntactic descriptions", "start": 356, "end": 378}}, {"head": {"text": "logical well-formedness conditions", "start": 636, "end": 670}, "tail": {"text": "trees", "start": 218, "end": 223}}]}}, "schema": []}
{"input": "We present a text mining method for finding synonymous expressions based on the distributional hypothesis in a set of coherent corpora. This paper proposes a new methodology to improve the accuracy of a term aggregation system using each author's text as a coherent corpus. Our approach is based on the idea that one person tends to use one expression for one meaning. According to our assumption, most of the words with similar context features in each author's corpus tend not to be synonymous expressions. Our proposed method improves the accuracy of our term aggregation system, showing that our approach is successful.", "output": {"entities": {"method": [{"text": "text mining method", "start": 13, "end": 31}, {"text": "term aggregation system", "start": 203, "end": 226}, {"text": "term aggregation system", "start": 558, "end": 581}], "other_scientific_term": [{"text": "synonymous expressions", "start": 44, "end": 66}, {"text": "distributional hypothesis", "start": 80, "end": 105}, {"text": "similar context features", "start": 421, "end": 445}, {"text": "synonymous expressions", "start": 485, "end": 507}], "generic": [{"text": "methodology", "start": 162, "end": 173}, {"text": "approach", "start": 278, "end": 286}, {"text": "method", "start": 25, "end": 31}, {"text": "approach", "start": 600, "end": 608}], "metric": [{"text": "accuracy", "start": 189, "end": 197}, {"text": "accuracy", "start": 542, "end": 550}]}, "relations": {"used_for": [{"head": {"text": "text mining method", "start": 13, "end": 31}, "tail": {"text": "synonymous expressions", "start": 44, "end": 66}}, {"head": {"text": "distributional hypothesis", "start": 80, "end": 105}, "tail": {"text": "text mining method", "start": 13, "end": 31}}], "evaluate_for": [{"head": {"text": "accuracy", "start": 189, "end": 197}, "tail": {"text": "term aggregation system", "start": 203, "end": 226}}, {"head": {"text": "term aggregation system", "start": 203, "end": 226}, "tail": {"text": "methodology", "start": 162, "end": 173}}, {"head": {"text": "accuracy", "start": 542, "end": 550}, "tail": {"text": "term aggregation system", "start": 558, "end": 581}}, {"head": {"text": "term aggregation system", "start": 558, "end": 581}, "tail": {"text": "method", "start": 25, "end": 31}}]}}, "schema": []}
{"input": "In this work, we present a technique for robust estimation, which by explicitly incorporating the inherent uncertainty of the estimation procedure, results in a more efficient robust estimation algorithm. In addition, we build on recent work in randomized model verification, and use this to characterize the `non-randomness' of a solution. The combination of these two strategies results in a robust estimation procedure that provides a significant speed-up over existing RANSAC techniques, while requiring no prior information to guide the sampling process. In particular, our algorithm requires, on average, 3-10 times fewer samples than standard RANSAC, which is in close agreement with theoretical predictions. The efficiency of the algorithm is demonstrated on a selection of geometric estimation problems.", "output": {"entities": {"generic": [{"text": "technique", "start": 27, "end": 36}, {"text": "this", "start": 3, "end": 7}, {"text": "strategies", "start": 370, "end": 380}, {"text": "algorithm", "start": 194, "end": 203}, {"text": "algorithm", "start": 579, "end": 588}], "task": [{"text": "robust estimation", "start": 41, "end": 58}, {"text": "randomized model verification", "start": 245, "end": 274}, {"text": "geometric estimation problems", "start": 782, "end": 811}], "other_scientific_term": [{"text": "inherent uncertainty of the estimation procedure", "start": 98, "end": 146}, {"text": "prior information", "start": 511, "end": 528}, {"text": "sampling process", "start": 542, "end": 558}, {"text": "theoretical predictions", "start": 691, "end": 714}], "method": [{"text": "efficient robust estimation algorithm", "start": 166, "end": 203}, {"text": "robust estimation procedure", "start": 394, "end": 421}, {"text": "RANSAC techniques", "start": 473, "end": 490}, {"text": "RANSAC", "start": 473, "end": 479}]}, "relations": {"used_for": [{"head": {"text": "technique", "start": 27, "end": 36}, "tail": {"text": "robust estimation", "start": 41, "end": 58}}, {"head": {"text": "technique", "start": 27, "end": 36}, "tail": {"text": "efficient robust estimation algorithm", "start": 166, "end": 203}}, {"head": {"text": "inherent uncertainty of the estimation procedure", "start": 98, "end": 146}, "tail": {"text": "technique", "start": 27, "end": 36}}, {"head": {"text": "strategies", "start": 370, "end": 380}, "tail": {"text": "robust estimation procedure", "start": 394, "end": 421}}], "compare": [{"head": {"text": "RANSAC techniques", "start": 473, "end": 490}, "tail": {"text": "robust estimation procedure", "start": 394, "end": 421}}, {"head": {"text": "algorithm", "start": 194, "end": 203}, "tail": {"text": "RANSAC", "start": 473, "end": 479}}], "evaluate_for": [{"head": {"text": "geometric estimation problems", "start": 782, "end": 811}, "tail": {"text": "algorithm", "start": 579, "end": 588}}]}}, "schema": []}
{"input": "An attempt has been made to use an Augmented Transition Network as a procedural dialog model. The development of such a model appears to be important in several respects: as a device to represent and to use different dialog schemata proposed in empirical conversation analysis; as a device to represent and to use models of verbal interaction; as a device combining knowledge about dialog schemata and about verbal interaction with knowledge about task-oriented and goal-directed dialogs. A standard ATN should be further developed in order to account for the verbal interactions of task-oriented dialogs.", "output": {"entities": {"method": [{"text": "Augmented Transition Network", "start": 35, "end": 63}, {"text": "dialog model", "start": 80, "end": 92}, {"text": "conversation analysis", "start": 255, "end": 276}, {"text": "ATN", "start": 500, "end": 503}], "generic": [{"text": "model", "start": 87, "end": 92}, {"text": "device", "start": 176, "end": 182}, {"text": "device", "start": 283, "end": 289}, {"text": "models", "start": 314, "end": 320}, {"text": "device", "start": 349, "end": 355}], "other_scientific_term": [{"text": "dialog schemata", "start": 217, "end": 232}, {"text": "verbal interaction", "start": 324, "end": 342}, {"text": "dialog schemata", "start": 382, "end": 397}, {"text": "verbal interaction", "start": 408, "end": 426}, {"text": "verbal interactions", "start": 560, "end": 579}], "material": [{"text": "task-oriented and goal-directed dialogs", "start": 448, "end": 487}, {"text": "task-oriented dialogs", "start": 583, "end": 604}]}, "relations": {"hyponym_of": [{"head": {"text": "Augmented Transition Network", "start": 35, "end": 63}, "tail": {"text": "dialog model", "start": 80, "end": 92}}], "used_for": [{"head": {"text": "dialog schemata", "start": 217, "end": 232}, "tail": {"text": "device", "start": 176, "end": 182}}, {"head": {"text": "dialog schemata", "start": 217, "end": 232}, "tail": {"text": "conversation analysis", "start": 255, "end": 276}}, {"head": {"text": "models", "start": 314, "end": 320}, "tail": {"text": "device", "start": 283, "end": 289}}, {"head": {"text": "models", "start": 314, "end": 320}, "tail": {"text": "verbal interaction", "start": 324, "end": 342}}, {"head": {"text": "ATN", "start": 500, "end": 503}, "tail": {"text": "verbal interactions", "start": 560, "end": 579}}], "conjunction": [{"head": {"text": "dialog schemata", "start": 382, "end": 397}, "tail": {"text": "verbal interaction", "start": 408, "end": 426}}], "feature_of": [{"head": {"text": "verbal interactions", "start": 560, "end": 579}, "tail": {"text": "task-oriented dialogs", "start": 583, "end": 604}}]}}, "schema": []}
{"input": "We present a practically unsupervised learning method to produce single-snippet answers to definition questions in question answering systems that supplement Web search engines. The method exploits on-line encyclopedias and dictionaries to generate automatically an arbitrarily large number of positive and negative definition examples, which are then used to train an svm to separate the two classes. We show experimentally that the proposed method is viable, that it outperforms the alternative of training the system on questions and news articles from trec, and that it helps the search engine handle definition questions significantly better.", "output": {"entities": {"method": [{"text": "unsupervised learning method", "start": 25, "end": 53}, {"text": "question answering systems", "start": 115, "end": 141}, {"text": "Web search engines", "start": 158, "end": 176}, {"text": "svm", "start": 369, "end": 372}, {"text": "search engine", "start": 162, "end": 175}], "other_scientific_term": [{"text": "single-snippet answers", "start": 65, "end": 87}], "generic": [{"text": "method", "start": 47, "end": 53}, {"text": "method", "start": 182, "end": 188}, {"text": "it", "start": 96, "end": 98}, {"text": "alternative", "start": 485, "end": 496}, {"text": "system", "start": 134, "end": 140}, {"text": "it", "start": 194, "end": 196}], "material": [{"text": "on-line encyclopedias and dictionaries", "start": 198, "end": 236}, {"text": "positive and negative definition examples", "start": 294, "end": 335}, {"text": "news articles", "start": 537, "end": 550}, {"text": "trec", "start": 556, "end": 560}]}, "relations": {"used_for": [{"head": {"text": "unsupervised learning method", "start": 25, "end": 53}, "tail": {"text": "single-snippet answers", "start": 65, "end": 87}}, {"head": {"text": "question answering systems", "start": 115, "end": 141}, "tail": {"text": "Web search engines", "start": 158, "end": 176}}, {"head": {"text": "method", "start": 47, "end": 53}, "tail": {"text": "on-line encyclopedias and dictionaries", "start": 198, "end": 236}}, {"head": {"text": "on-line encyclopedias and dictionaries", "start": 198, "end": 236}, "tail": {"text": "positive and negative definition examples", "start": 294, "end": 335}}, {"head": {"text": "positive and negative definition examples", "start": 294, "end": 335}, "tail": {"text": "svm", "start": 369, "end": 372}}, {"head": {"text": "news articles", "start": 537, "end": 550}, "tail": {"text": "system", "start": 134, "end": 140}}, {"head": {"text": "it", "start": 194, "end": 196}, "tail": {"text": "search engine", "start": 162, "end": 175}}], "compare": [{"head": {"text": "it", "start": 96, "end": 98}, "tail": {"text": "alternative", "start": 485, "end": 496}}], "part_of": [{"head": {"text": "news articles", "start": 537, "end": 550}, "tail": {"text": "trec", "start": 556, "end": 560}}]}}, "schema": []}
{"input": "We revisit the classical decision-theoretic problem of weighted expert voting from a statistical learning perspective. In particular, we examine the consistency -LRB- both asymptotic and finitary -RRB- of the optimal Nitzan-Paroush weighted majority and related rules. In the case of known expert competence levels, we give sharp error estimates for the optimal rule. When the competence levels are unknown, they must be empirically estimated. We provide frequentist and Bayesian analyses for this situation. Some of our proof techniques are non-standard and may be of independent interest. The bounds we derive are nearly optimal, and several challenging open problems are posed. Experimental results are provided to illustrate the theory.", "output": {"entities": {"task": [{"text": "classical decision-theoretic problem of weighted expert voting", "start": 15, "end": 77}], "method": [{"text": "statistical learning perspective", "start": 85, "end": 117}, {"text": "sharp error estimates", "start": 324, "end": 345}, {"text": "Bayesian analyses", "start": 471, "end": 488}], "other_scientific_term": [{"text": "Nitzan-Paroush weighted majority", "start": 217, "end": 249}, {"text": "expert competence levels", "start": 290, "end": 314}, {"text": "optimal rule", "start": 354, "end": 366}, {"text": "competence levels", "start": 297, "end": 314}]}, "relations": {"used_for": [{"head": {"text": "statistical learning perspective", "start": 85, "end": 117}, "tail": {"text": "classical decision-theoretic problem of weighted expert voting", "start": 15, "end": 77}}, {"head": {"text": "sharp error estimates", "start": 324, "end": 345}, "tail": {"text": "optimal rule", "start": 354, "end": 366}}]}}, "schema": []}
{"input": "We analyze a reweighted version of the Kikuchi approximation for estimating the log partition function of a product distribution defined over a region graph. We establish sufficient conditions for the concavity of our reweighted objective function in terms of weight assignments in the Kikuchi expansion, and show that a reweighted version of the sum product algorithm applied to the Kikuchi region graph will produce global optima of the Kikuchi approximation whenever the algorithm converges. When the region graph has two layers, corresponding to a Bethe approximation, we show that our sufficient conditions for concavity are also necessary. Finally, we provide an explicit characterization of the polytope of concavity in terms of the cycle structure of the region graph. We conclude with simulations that demonstrate the advantages of the reweighted Kikuchi approach.", "output": {"entities": {"method": [{"text": "reweighted version of the Kikuchi approximation", "start": 13, "end": 60}, {"text": "Kikuchi approximation", "start": 39, "end": 60}, {"text": "reweighted version of the sum product algorithm", "start": 321, "end": 368}, {"text": "Kikuchi approximation", "start": 439, "end": 460}, {"text": "Bethe approximation", "start": 552, "end": 571}, {"text": "reweighted Kikuchi approach", "start": 845, "end": 872}], "task": [{"text": "log partition function of a product distribution", "start": 80, "end": 128}], "other_scientific_term": [{"text": "region graph", "start": 144, "end": 156}, {"text": "concavity", "start": 201, "end": 210}, {"text": "reweighted objective function", "start": 218, "end": 247}, {"text": "weight assignments", "start": 260, "end": 278}, {"text": "Kikuchi expansion", "start": 286, "end": 303}, {"text": "Kikuchi region graph", "start": 384, "end": 404}, {"text": "global optima", "start": 418, "end": 431}, {"text": "region graph", "start": 392, "end": 404}, {"text": "concavity", "start": 616, "end": 625}, {"text": "concavity", "start": 714, "end": 723}, {"text": "cycle structure", "start": 740, "end": 755}, {"text": "region graph", "start": 504, "end": 516}], "generic": [{"text": "algorithm", "start": 359, "end": 368}]}, "relations": {"used_for": [{"head": {"text": "reweighted version of the Kikuchi approximation", "start": 13, "end": 60}, "tail": {"text": "log partition function of a product distribution", "start": 80, "end": 128}}, {"head": {"text": "reweighted version of the sum product algorithm", "start": 321, "end": 368}, "tail": {"text": "Kikuchi region graph", "start": 384, "end": 404}}], "feature_of": [{"head": {"text": "log partition function of a product distribution", "start": 80, "end": 128}, "tail": {"text": "region graph", "start": 144, "end": 156}}, {"head": {"text": "concavity", "start": 201, "end": 210}, "tail": {"text": "reweighted objective function", "start": 218, "end": 247}}, {"head": {"text": "global optima", "start": 418, "end": 431}, "tail": {"text": "Kikuchi approximation", "start": 439, "end": 460}}, {"head": {"text": "cycle structure", "start": 740, "end": 755}, "tail": {"text": "region graph", "start": 504, "end": 516}}]}}, "schema": []}
{"input": "We apply a decision tree based approach to pronoun resolution in spoken dialogue. Our system deals with pronouns with NP-and non-NP-antecedents. We present a set of features designed for pronoun resolution in spoken dialogue and determine the most promising features. We evaluate the system on twenty Switchboard dialogues and show that it compares well to Byron's -LRB- 2002 -RRB- manually tuned system.", "output": {"entities": {"method": [{"text": "decision tree based approach", "start": 11, "end": 39}, {"text": "Byron's -LRB- 2002 -RRB- manually tuned system", "start": 357, "end": 403}], "task": [{"text": "pronoun resolution", "start": 43, "end": 61}, {"text": "spoken dialogue", "start": 65, "end": 80}, {"text": "pronoun resolution", "start": 187, "end": 205}, {"text": "spoken dialogue", "start": 209, "end": 224}], "generic": [{"text": "system", "start": 86, "end": 92}, {"text": "system", "start": 284, "end": 290}, {"text": "it", "start": 100, "end": 102}], "other_scientific_term": [{"text": "pronouns", "start": 104, "end": 112}, {"text": "NP-and non-NP-antecedents", "start": 118, "end": 143}, {"text": "features", "start": 165, "end": 173}, {"text": "features", "start": 258, "end": 266}], "material": [{"text": "Switchboard dialogues", "start": 301, "end": 322}]}, "relations": {"used_for": [{"head": {"text": "decision tree based approach", "start": 11, "end": 39}, "tail": {"text": "pronoun resolution", "start": 43, "end": 61}}, {"head": {"text": "pronoun resolution", "start": 43, "end": 61}, "tail": {"text": "spoken dialogue", "start": 65, "end": 80}}, {"head": {"text": "system", "start": 86, "end": 92}, "tail": {"text": "pronouns", "start": 104, "end": 112}}, {"head": {"text": "NP-and non-NP-antecedents", "start": 118, "end": 143}, "tail": {"text": "pronouns", "start": 104, "end": 112}}, {"head": {"text": "features", "start": 165, "end": 173}, "tail": {"text": "pronoun resolution", "start": 187, "end": 205}}, {"head": {"text": "pronoun resolution", "start": 187, "end": 205}, "tail": {"text": "spoken dialogue", "start": 209, "end": 224}}], "evaluate_for": [{"head": {"text": "Switchboard dialogues", "start": 301, "end": 322}, "tail": {"text": "system", "start": 284, "end": 290}}], "compare": [{"head": {"text": "it", "start": 100, "end": 102}, "tail": {"text": "Byron's -LRB- 2002 -RRB- manually tuned system", "start": 357, "end": 403}}]}}, "schema": []}
{"input": "We present a new approach for building an efficient and robust classifier for the two class problem, that localizes objects that may appear in the image under different orien-tations. In contrast to other works that address this problem using multiple classifiers, each one specialized for a specific orientation, we propose a simple two-step approach with an estimation stage and a classification stage. The estimator yields an initial set of potential object poses that are then validated by the classifier. This methodology allows reducing the time complexity of the algorithm while classification results remain high. The classifier we use in both stages is based on a boosted combination of Random Ferns over local histograms of oriented gradients -LRB- HOGs -RRB-, which we compute during a pre-processing step. Both the use of supervised learning and working on the gradient space makes our approach robust while being efficient at run-time. We show these properties by thorough testing on standard databases and on a new database made of motorbikes under planar rotations, and with challenging conditions such as cluttered backgrounds, changing illumination conditions and partial occlusions.", "output": {"entities": {"generic": [{"text": "approach", "start": 17, "end": 25}, {"text": "problem", "start": 92, "end": 99}, {"text": "approach", "start": 343, "end": 351}, {"text": "algorithm", "start": 570, "end": 579}, {"text": "approach", "start": 898, "end": 906}, {"text": "database", "start": 1006, "end": 1014}, {"text": "conditions", "start": 1102, "end": 1112}], "method": [{"text": "classifier", "start": 63, "end": 73}, {"text": "classifiers", "start": 252, "end": 263}, {"text": "estimation stage", "start": 360, "end": 376}, {"text": "classification stage", "start": 383, "end": 403}, {"text": "estimator", "start": 409, "end": 418}, {"text": "classifier", "start": 252, "end": 262}, {"text": "classifier", "start": 498, "end": 508}, {"text": "boosted combination of Random Ferns", "start": 673, "end": 708}, {"text": "pre-processing step", "start": 797, "end": 816}, {"text": "supervised learning", "start": 834, "end": 853}], "task": [{"text": "class problem", "start": 86, "end": 99}, {"text": "classification", "start": 383, "end": 397}], "material": [{"text": "image", "start": 147, "end": 152}, {"text": "motorbikes under planar rotations", "start": 1046, "end": 1079}], "other_scientific_term": [{"text": "orien-tations", "start": 169, "end": 182}, {"text": "object poses", "start": 454, "end": 466}, {"text": "local histograms of oriented gradients -LRB- HOGs -RRB-", "start": 714, "end": 769}, {"text": "gradient space", "start": 873, "end": 887}, {"text": "cluttered backgrounds", "start": 1121, "end": 1142}, {"text": "changing illumination conditions", "start": 1144, "end": 1176}, {"text": "partial occlusions", "start": 1181, "end": 1199}], "metric": [{"text": "time complexity", "start": 547, "end": 562}]}, "relations": {"used_for": [{"head": {"text": "approach", "start": 17, "end": 25}, "tail": {"text": "classifier", "start": 63, "end": 73}}, {"head": {"text": "classifier", "start": 63, "end": 73}, "tail": {"text": "class problem", "start": 86, "end": 99}}, {"head": {"text": "classifier", "start": 252, "end": 262}, "tail": {"text": "object poses", "start": 454, "end": 466}}, {"head": {"text": "boosted combination of Random Ferns", "start": 673, "end": 708}, "tail": {"text": "classifier", "start": 498, "end": 508}}, {"head": {"text": "pre-processing step", "start": 797, "end": 816}, "tail": {"text": "local histograms of oriented gradients -LRB- HOGs -RRB-", "start": 714, "end": 769}}, {"head": {"text": "supervised learning", "start": 834, "end": 853}, "tail": {"text": "approach", "start": 898, "end": 906}}, {"head": {"text": "gradient space", "start": 873, "end": 887}, "tail": {"text": "approach", "start": 898, "end": 906}}], "part_of": [{"head": {"text": "estimation stage", "start": 360, "end": 376}, "tail": {"text": "approach", "start": 343, "end": 351}}, {"head": {"text": "classification stage", "start": 383, "end": 403}, "tail": {"text": "approach", "start": 343, "end": 351}}], "conjunction": [{"head": {"text": "estimation stage", "start": 360, "end": 376}, "tail": {"text": "classification stage", "start": 383, "end": 403}}, {"head": {"text": "cluttered backgrounds", "start": 1121, "end": 1142}, "tail": {"text": "changing illumination conditions", "start": 1144, "end": 1176}}, {"head": {"text": "changing illumination conditions", "start": 1144, "end": 1176}, "tail": {"text": "partial occlusions", "start": 1181, "end": 1199}}], "evaluate_for": [{"head": {"text": "time complexity", "start": 547, "end": 562}, "tail": {"text": "algorithm", "start": 570, "end": 579}}], "feature_of": [{"head": {"text": "local histograms of oriented gradients -LRB- HOGs -RRB-", "start": 714, "end": 769}, "tail": {"text": "boosted combination of Random Ferns", "start": 673, "end": 708}}, {"head": {"text": "motorbikes under planar rotations", "start": 1046, "end": 1079}, "tail": {"text": "database", "start": 1006, "end": 1014}}, {"head": {"text": "conditions", "start": 1102, "end": 1112}, "tail": {"text": "database", "start": 1006, "end": 1014}}], "hyponym_of": [{"head": {"text": "cluttered backgrounds", "start": 1121, "end": 1142}, "tail": {"text": "conditions", "start": 1102, "end": 1112}}, {"head": {"text": "changing illumination conditions", "start": 1144, "end": 1176}, "tail": {"text": "conditions", "start": 1102, "end": 1112}}, {"head": {"text": "partial occlusions", "start": 1181, "end": 1199}, "tail": {"text": "conditions", "start": 1102, "end": 1112}}]}}, "schema": []}
{"input": "The following describes recent work on the Lincoln CSR system. Some new variations in semiphone modeling have been tested. A very simple improved duration model has reduced the error rate by about 10% in both triphone and semiphone systems. A new training strategy has been tested which, by itself, did not provide useful improvements but suggests that improvements can be obtained by a related rapid adaptation technique. Finally, the recognizer has been modified to use bigram back-off language models. The system was then transferred from the RM task to the ATIS CSR task and a limited number of development tests performed. Evaluation test results are presented for both the RM and ATIS CSR tasks.", "output": {"entities": {"method": [{"text": "Lincoln CSR system", "start": 43, "end": 61}, {"text": "semiphone modeling", "start": 86, "end": 104}, {"text": "duration model", "start": 146, "end": 160}, {"text": "triphone and semiphone systems", "start": 209, "end": 239}, {"text": "training strategy", "start": 247, "end": 264}, {"text": "rapid adaptation technique", "start": 395, "end": 421}, {"text": "recognizer", "start": 436, "end": 446}, {"text": "bigram back-off language models", "start": 472, "end": 503}], "metric": [{"text": "error rate", "start": 177, "end": 187}], "generic": [{"text": "system", "start": 55, "end": 61}], "task": [{"text": "RM task", "start": 546, "end": 553}, {"text": "ATIS CSR task", "start": 561, "end": 574}, {"text": "RM and ATIS CSR tasks", "start": 679, "end": 700}]}, "relations": {"used_for": [{"head": {"text": "duration model", "start": 146, "end": 160}, "tail": {"text": "triphone and semiphone systems", "start": 209, "end": 239}}, {"head": {"text": "rapid adaptation technique", "start": 395, "end": 421}, "tail": {"text": "training strategy", "start": 247, "end": 264}}, {"head": {"text": "bigram back-off language models", "start": 472, "end": 503}, "tail": {"text": "recognizer", "start": 436, "end": 446}}, {"head": {"text": "system", "start": 55, "end": 61}, "tail": {"text": "RM task", "start": 546, "end": 553}}, {"head": {"text": "system", "start": 55, "end": 61}, "tail": {"text": "ATIS CSR task", "start": 561, "end": 574}}], "evaluate_for": [{"head": {"text": "error rate", "start": 177, "end": 187}, "tail": {"text": "triphone and semiphone systems", "start": 209, "end": 239}}], "conjunction": [{"head": {"text": "RM task", "start": 546, "end": 553}, "tail": {"text": "ATIS CSR task", "start": 561, "end": 574}}]}}, "schema": []}
{"input": "A new approach for Interactive Machine Translation where the author interacts during the creation or the modification of the document is proposed. The explanation of an ambiguity or an error for the purposes of correction does not use any concepts of the underlying linguistic theory: it is a reformulation of the erroneous or ambiguous sentence. The interaction is limited to the analysis step of the translation process. This paper presents a new interactive disambiguation scheme based on the paraphrasing of a parser's multiple output. Some examples of paraphrasing ambiguous sentences are presented.", "output": {"entities": {"generic": [{"text": "approach", "start": 6, "end": 14}], "task": [{"text": "Interactive Machine Translation", "start": 19, "end": 50}], "method": [{"text": "linguistic theory", "start": 266, "end": 283}, {"text": "translation process", "start": 402, "end": 421}, {"text": "interactive disambiguation scheme", "start": 449, "end": 482}, {"text": "paraphrasing", "start": 496, "end": 508}]}, "relations": {"used_for": [{"head": {"text": "approach", "start": 6, "end": 14}, "tail": {"text": "Interactive Machine Translation", "start": 19, "end": 50}}, {"head": {"text": "paraphrasing", "start": 496, "end": 508}, "tail": {"text": "interactive disambiguation scheme", "start": 449, "end": 482}}]}}, "schema": []}
{"input": "We describe a novel approach to statistical machine translation that combines syntactic information in the source language with recent advances in phrasal translation. This method requires a source-language dependency parser, target language word segmentation and an unsupervised word alignment component. We align a parallel corpus, project the source dependency parse onto the target sentence, extract dependency treelet translation pairs, and train a tree-based ordering model. We describe an efficient decoder and show that using these tree-based models in combination with conventional SMT models provides a promising approach that incorporates the power of phrasal SMT with the linguistic generality available in a parser.", "output": {"entities": {"generic": [{"text": "approach", "start": 20, "end": 28}, {"text": "method", "start": 173, "end": 179}, {"text": "approach", "start": 623, "end": 631}], "task": [{"text": "statistical machine translation", "start": 32, "end": 63}, {"text": "phrasal translation", "start": 147, "end": 166}], "other_scientific_term": [{"text": "syntactic information", "start": 78, "end": 99}, {"text": "source dependency parse", "start": 346, "end": 369}, {"text": "dependency treelet translation pairs", "start": 404, "end": 440}, {"text": "linguistic generality", "start": 684, "end": 705}], "method": [{"text": "source-language dependency parser", "start": 191, "end": 224}, {"text": "target language word segmentation", "start": 226, "end": 259}, {"text": "unsupervised word alignment component", "start": 267, "end": 304}, {"text": "tree-based ordering model", "start": 454, "end": 479}, {"text": "decoder", "start": 506, "end": 513}, {"text": "tree-based models", "start": 540, "end": 557}, {"text": "SMT models", "start": 591, "end": 601}, {"text": "phrasal SMT", "start": 663, "end": 674}, {"text": "parser", "start": 218, "end": 224}], "material": [{"text": "parallel corpus", "start": 317, "end": 332}]}, "relations": {"used_for": [{"head": {"text": "approach", "start": 20, "end": 28}, "tail": {"text": "statistical machine translation", "start": 32, "end": 63}}, {"head": {"text": "source-language dependency parser", "start": 191, "end": 224}, "tail": {"text": "method", "start": 173, "end": 179}}, {"head": {"text": "target language word segmentation", "start": 226, "end": 259}, "tail": {"text": "method", "start": 173, "end": 179}}, {"head": {"text": "unsupervised word alignment component", "start": 267, "end": 304}, "tail": {"text": "method", "start": 173, "end": 179}}, {"head": {"text": "tree-based models", "start": 540, "end": 557}, "tail": {"text": "approach", "start": 623, "end": 631}}, {"head": {"text": "SMT models", "start": 591, "end": 601}, "tail": {"text": "approach", "start": 623, "end": 631}}, {"head": {"text": "phrasal SMT", "start": 663, "end": 674}, "tail": {"text": "parser", "start": 218, "end": 224}}], "part_of": [{"head": {"text": "syntactic information", "start": 78, "end": 99}, "tail": {"text": "approach", "start": 20, "end": 28}}, {"head": {"text": "phrasal translation", "start": 147, "end": 166}, "tail": {"text": "approach", "start": 20, "end": 28}}], "conjunction": [{"head": {"text": "syntactic information", "start": 78, "end": 99}, "tail": {"text": "phrasal translation", "start": 147, "end": 166}}, {"head": {"text": "source-language dependency parser", "start": 191, "end": 224}, "tail": {"text": "target language word segmentation", "start": 226, "end": 259}}, {"head": {"text": "target language word segmentation", "start": 226, "end": 259}, "tail": {"text": "unsupervised word alignment component", "start": 267, "end": 304}}, {"head": {"text": "tree-based models", "start": 540, "end": 557}, "tail": {"text": "SMT models", "start": 591, "end": 601}}, {"head": {"text": "phrasal SMT", "start": 663, "end": 674}, "tail": {"text": "linguistic generality", "start": 684, "end": 705}}], "feature_of": [{"head": {"text": "linguistic generality", "start": 684, "end": 705}, "tail": {"text": "parser", "start": 218, "end": 224}}]}}, "schema": []}
{"input": "Video provides not only rich visual cues such as motion and appearance, but also much less explored long-range temporal interactions among objects. We aim to capture such interactions and to construct a powerful intermediate-level video representation for subsequent recognition. Motivated by this goal, we seek to obtain spatio-temporal over-segmentation of a video into regions that respect object boundaries and, at the same time, associate object pix-els over many video frames. The contributions of this paper are twofold. First, we develop an efficient spatio-temporal video segmentation algorithm, which naturally incorporates long-range motion cues from the past and future frames in the form of clusters of point tracks with coherent motion. Second, we devise a new track clustering cost function that includes occlusion reasoning, in the form of depth ordering constraints, as well as motion similarity along the tracks. We evaluate the proposed approach on a challenging set of video sequences of office scenes from feature length movies.", "output": {"entities": {"material": [{"text": "Video", "start": 0, "end": 5}, {"text": "video sequences of office scenes", "start": 989, "end": 1021}, {"text": "movies", "start": 1042, "end": 1048}], "other_scientific_term": [{"text": "visual cues", "start": 29, "end": 40}, {"text": "motion", "start": 49, "end": 55}, {"text": "appearance", "start": 60, "end": 70}, {"text": "long-range temporal interactions", "start": 100, "end": 132}, {"text": "spatio-temporal over-segmentation", "start": 322, "end": 355}, {"text": "object boundaries", "start": 393, "end": 410}, {"text": "long-range motion cues", "start": 634, "end": 656}, {"text": "clusters of point tracks", "start": 704, "end": 728}, {"text": "occlusion reasoning", "start": 820, "end": 839}, {"text": "depth ordering constraints", "start": 856, "end": 882}, {"text": "motion similarity", "start": 895, "end": 912}], "generic": [{"text": "interactions", "start": 120, "end": 132}, {"text": "approach", "start": 956, "end": 964}], "method": [{"text": "intermediate-level video representation", "start": 212, "end": 251}, {"text": "spatio-temporal video segmentation algorithm", "start": 559, "end": 603}, {"text": "track clustering cost function", "start": 775, "end": 805}], "task": [{"text": "recognition", "start": 267, "end": 278}]}, "relations": {"feature_of": [{"head": {"text": "visual cues", "start": 29, "end": 40}, "tail": {"text": "Video", "start": 0, "end": 5}}, {"head": {"text": "depth ordering constraints", "start": 856, "end": 882}, "tail": {"text": "occlusion reasoning", "start": 820, "end": 839}}], "hyponym_of": [{"head": {"text": "motion", "start": 49, "end": 55}, "tail": {"text": "visual cues", "start": 29, "end": 40}}, {"head": {"text": "appearance", "start": 60, "end": 70}, "tail": {"text": "visual cues", "start": 29, "end": 40}}], "conjunction": [{"head": {"text": "motion", "start": 49, "end": 55}, "tail": {"text": "appearance", "start": 60, "end": 70}}], "used_for": [{"head": {"text": "intermediate-level video representation", "start": 212, "end": 251}, "tail": {"text": "recognition", "start": 267, "end": 278}}, {"head": {"text": "long-range motion cues", "start": 634, "end": 656}, "tail": {"text": "spatio-temporal video segmentation algorithm", "start": 559, "end": 603}}, {"head": {"text": "clusters of point tracks", "start": 704, "end": 728}, "tail": {"text": "long-range motion cues", "start": 634, "end": 656}}], "part_of": [{"head": {"text": "occlusion reasoning", "start": 820, "end": 839}, "tail": {"text": "track clustering cost function", "start": 775, "end": 805}}, {"head": {"text": "motion similarity", "start": 895, "end": 912}, "tail": {"text": "track clustering cost function", "start": 775, "end": 805}}], "evaluate_for": [{"head": {"text": "video sequences of office scenes", "start": 989, "end": 1021}, "tail": {"text": "approach", "start": 956, "end": 964}}]}}, "schema": []}
{"input": "In this paper, we introduce KAZE features, a novel multiscale 2D feature detection and description algorithm in nonlinear scale spaces. Previous approaches detect and describe features at different scale levels by building or approximating the Gaussian scale space of an image. However, Gaussian blurring does not respect the natural boundaries of objects and smoothes to the same degree both details and noise, reducing localization accuracy and distinctiveness. In contrast, we detect and describe 2D features in a nonlinear scale space by means of nonlinear diffusion filtering. In this way, we can make blurring locally adaptive to the image data, reducing noise but retaining object boundaries, obtaining superior localization accuracy and distinctiviness. The nonlinear scale space is built using efficient Additive Operator Splitting -LRB- AOS -RRB- techniques and variable con-ductance diffusion. We present an extensive evaluation on benchmark datasets and a practical matching application on deformable surfaces. Even though our features are somewhat more expensive to compute than SURF due to the construction of the nonlinear scale space, but comparable to SIFT, our results reveal a step forward in performance both in detection and description against previous state-of-the-art methods.", "output": {"entities": {"method": [{"text": "KAZE features", "start": 28, "end": 41}, {"text": "multiscale 2D feature detection and description algorithm", "start": 51, "end": 108}, {"text": "2D features", "start": 500, "end": 511}, {"text": "nonlinear diffusion filtering", "start": 551, "end": 580}, {"text": "Additive Operator Splitting -LRB- AOS -RRB- techniques", "start": 813, "end": 867}, {"text": "variable con-ductance diffusion", "start": 872, "end": 903}, {"text": "SURF", "start": 1092, "end": 1096}, {"text": "SIFT", "start": 1169, "end": 1173}], "other_scientific_term": [{"text": "nonlinear scale spaces", "start": 112, "end": 134}, {"text": "Gaussian scale space", "start": 244, "end": 264}, {"text": "Gaussian blurring", "start": 287, "end": 304}, {"text": "boundaries of objects", "start": 334, "end": 355}, {"text": "nonlinear scale space", "start": 112, "end": 133}, {"text": "object boundaries", "start": 681, "end": 698}, {"text": "nonlinear scale space", "start": 517, "end": 538}, {"text": "nonlinear scale space", "start": 766, "end": 787}], "metric": [{"text": "localization accuracy and distinctiveness", "start": 421, "end": 462}, {"text": "localization accuracy and distinctiviness", "start": 719, "end": 760}], "material": [{"text": "image data", "start": 640, "end": 650}, {"text": "benchmark datasets", "start": 943, "end": 961}], "task": [{"text": "matching application on deformable surfaces", "start": 978, "end": 1021}, {"text": "detection", "start": 73, "end": 82}, {"text": "description", "start": 87, "end": 98}], "generic": [{"text": "features", "start": 33, "end": 41}, {"text": "results", "start": 1179, "end": 1186}, {"text": "state-of-the-art methods", "start": 1275, "end": 1299}]}, "relations": {"hyponym_of": [{"head": {"text": "KAZE features", "start": 28, "end": 41}, "tail": {"text": "multiscale 2D feature detection and description algorithm", "start": 51, "end": 108}}], "feature_of": [{"head": {"text": "nonlinear scale spaces", "start": 112, "end": 134}, "tail": {"text": "multiscale 2D feature detection and description algorithm", "start": 51, "end": 108}}, {"head": {"text": "nonlinear scale space", "start": 112, "end": 133}, "tail": {"text": "2D features", "start": 500, "end": 511}}], "used_for": [{"head": {"text": "nonlinear diffusion filtering", "start": 551, "end": 580}, "tail": {"text": "2D features", "start": 500, "end": 511}}, {"head": {"text": "Additive Operator Splitting -LRB- AOS -RRB- techniques", "start": 813, "end": 867}, "tail": {"text": "nonlinear scale space", "start": 517, "end": 538}}, {"head": {"text": "variable con-ductance diffusion", "start": 872, "end": 903}, "tail": {"text": "nonlinear scale space", "start": 517, "end": 538}}], "conjunction": [{"head": {"text": "Additive Operator Splitting -LRB- AOS -RRB- techniques", "start": 813, "end": 867}, "tail": {"text": "variable con-ductance diffusion", "start": 872, "end": 903}}, {"head": {"text": "detection", "start": 73, "end": 82}, "tail": {"text": "description", "start": 87, "end": 98}}], "compare": [{"head": {"text": "features", "start": 33, "end": 41}, "tail": {"text": "SURF", "start": 1092, "end": 1096}}, {"head": {"text": "features", "start": 33, "end": 41}, "tail": {"text": "SIFT", "start": 1169, "end": 1173}}, {"head": {"text": "results", "start": 1179, "end": 1186}, "tail": {"text": "state-of-the-art methods", "start": 1275, "end": 1299}}], "evaluate_for": [{"head": {"text": "detection", "start": 73, "end": 82}, "tail": {"text": "results", "start": 1179, "end": 1186}}, {"head": {"text": "detection", "start": 73, "end": 82}, "tail": {"text": "state-of-the-art methods", "start": 1275, "end": 1299}}, {"head": {"text": "description", "start": 87, "end": 98}, "tail": {"text": "results", "start": 1179, "end": 1186}}, {"head": {"text": "description", "start": 87, "end": 98}, "tail": {"text": "state-of-the-art methods", "start": 1275, "end": 1299}}]}}, "schema": []}
{"input": "Semantic Web documents that encode facts about entities on the Web have been growing rapidly in size and evolving over time. Creating summaries on lengthy Semantic Web documents for quick identification of the corresponding entity has been of great contemporary interest. In this paper, we explore automatic summa-rization techniques that characterize and enable identification of an entity and create summaries that are human friendly. Specifically, we highlight the importance of diversified -LRB- faceted -RRB- summaries by combining three dimensions: diversity, uniqueness, and popularity. Our novel diversity-aware entity summarization approach mimics human conceptual clustering techniques to group facts, and picks representative facts from each group to form concise -LRB- i.e., short -RRB- and comprehensive -LRB- i.e., improved coverage through diversity -RRB- summaries. We evaluate our approach against the state-of-the-art techniques and show that our work improves both the quality and the efficiency of entity summarization.", "output": {"entities": {"material": [{"text": "Semantic Web documents", "start": 0, "end": 22}, {"text": "lengthy Semantic Web documents", "start": 147, "end": 177}], "task": [{"text": "Creating summaries", "start": 125, "end": 143}, {"text": "identification of the corresponding entity", "start": 188, "end": 230}, {"text": "entity summarization", "start": 620, "end": 640}], "method": [{"text": "automatic summa-rization techniques", "start": 298, "end": 333}, {"text": "diversity-aware entity summarization approach", "start": 604, "end": 649}, {"text": "human conceptual clustering techniques", "start": 657, "end": 695}], "other_scientific_term": [{"text": "diversified -LRB- faceted -RRB- summaries", "start": 482, "end": 523}, {"text": "diversity", "start": 555, "end": 564}, {"text": "uniqueness", "start": 566, "end": 576}, {"text": "popularity", "start": 582, "end": 592}], "generic": [{"text": "approach", "start": 641, "end": 649}, {"text": "state-of-the-art techniques", "start": 919, "end": 946}], "metric": [{"text": "quality", "start": 988, "end": 995}, {"text": "efficiency", "start": 1004, "end": 1014}]}, "relations": {"used_for": [{"head": {"text": "Creating summaries", "start": 125, "end": 143}, "tail": {"text": "identification of the corresponding entity", "start": 188, "end": 230}}, {"head": {"text": "lengthy Semantic Web documents", "start": 147, "end": 177}, "tail": {"text": "Creating summaries", "start": 125, "end": 143}}, {"head": {"text": "human conceptual clustering techniques", "start": 657, "end": 695}, "tail": {"text": "diversity-aware entity summarization approach", "start": 604, "end": 649}}, {"head": {"text": "approach", "start": 641, "end": 649}, "tail": {"text": "entity summarization", "start": 620, "end": 640}}, {"head": {"text": "state-of-the-art techniques", "start": 919, "end": 946}, "tail": {"text": "entity summarization", "start": 620, "end": 640}}], "feature_of": [{"head": {"text": "diversity", "start": 555, "end": 564}, "tail": {"text": "diversified -LRB- faceted -RRB- summaries", "start": 482, "end": 523}}, {"head": {"text": "uniqueness", "start": 566, "end": 576}, "tail": {"text": "diversified -LRB- faceted -RRB- summaries", "start": 482, "end": 523}}, {"head": {"text": "popularity", "start": 582, "end": 592}, "tail": {"text": "diversified -LRB- faceted -RRB- summaries", "start": 482, "end": 523}}], "conjunction": [{"head": {"text": "diversity", "start": 555, "end": 564}, "tail": {"text": "uniqueness", "start": 566, "end": 576}}, {"head": {"text": "uniqueness", "start": 566, "end": 576}, "tail": {"text": "popularity", "start": 582, "end": 592}}], "compare": [{"head": {"text": "state-of-the-art techniques", "start": 919, "end": 946}, "tail": {"text": "approach", "start": 641, "end": 649}}], "evaluate_for": [{"head": {"text": "quality", "start": 988, "end": 995}, "tail": {"text": "entity summarization", "start": 620, "end": 640}}, {"head": {"text": "efficiency", "start": 1004, "end": 1014}, "tail": {"text": "entity summarization", "start": 620, "end": 640}}]}}, "schema": []}
{"input": "We present a framework for the fast computation of lexical affinity models. The framework is composed of a novel algorithm to efficiently compute the co-occurrence distribution between pairs of terms, an independence model, and a parametric affinity model. In comparison with previous models, which either use arbitrary windows to compute similarity between words or use lexical affinity to create sequential models, in this paper we focus on models intended to capture the co-occurrence patterns of any pair of words or phrases at any distance in the corpus. The framework is flexible, allowing fast adaptation to applications and it is scalable. We apply it in combination with a terabyte corpus to answer natural language tests, achieving encouraging results.", "output": {"entities": {"generic": [{"text": "framework", "start": 13, "end": 22}, {"text": "framework", "start": 80, "end": 89}, {"text": "algorithm", "start": 113, "end": 122}, {"text": "models", "start": 68, "end": 74}, {"text": "models", "start": 285, "end": 291}, {"text": "framework", "start": 564, "end": 573}, {"text": "it", "start": 64, "end": 66}], "task": [{"text": "fast computation of lexical affinity models", "start": 31, "end": 74}, {"text": "natural language tests", "start": 708, "end": 730}], "other_scientific_term": [{"text": "co-occurrence distribution", "start": 150, "end": 176}, {"text": "similarity", "start": 339, "end": 349}, {"text": "lexical affinity", "start": 51, "end": 67}, {"text": "co-occurrence patterns", "start": 474, "end": 496}], "method": [{"text": "independence model", "start": 204, "end": 222}, {"text": "parametric affinity model", "start": 230, "end": 255}, {"text": "sequential models", "start": 398, "end": 415}], "material": [{"text": "terabyte corpus", "start": 682, "end": 697}]}, "relations": {"used_for": [{"head": {"text": "framework", "start": 13, "end": 22}, "tail": {"text": "fast computation of lexical affinity models", "start": 31, "end": 74}}, {"head": {"text": "algorithm", "start": 113, "end": 122}, "tail": {"text": "co-occurrence distribution", "start": 150, "end": 176}}, {"head": {"text": "lexical affinity", "start": 51, "end": 67}, "tail": {"text": "sequential models", "start": 398, "end": 415}}, {"head": {"text": "models", "start": 285, "end": 291}, "tail": {"text": "co-occurrence patterns", "start": 474, "end": 496}}, {"head": {"text": "it", "start": 64, "end": 66}, "tail": {"text": "natural language tests", "start": 708, "end": 730}}], "part_of": [{"head": {"text": "algorithm", "start": 113, "end": 122}, "tail": {"text": "framework", "start": 80, "end": 89}}, {"head": {"text": "independence model", "start": 204, "end": 222}, "tail": {"text": "framework", "start": 80, "end": 89}}, {"head": {"text": "parametric affinity model", "start": 230, "end": 255}, "tail": {"text": "framework", "start": 80, "end": 89}}], "conjunction": [{"head": {"text": "algorithm", "start": 113, "end": 122}, "tail": {"text": "independence model", "start": 204, "end": 222}}, {"head": {"text": "independence model", "start": 204, "end": 222}, "tail": {"text": "parametric affinity model", "start": 230, "end": 255}}], "compare": [{"head": {"text": "models", "start": 285, "end": 291}, "tail": {"text": "models", "start": 68, "end": 74}}], "evaluate_for": [{"head": {"text": "terabyte corpus", "start": 682, "end": 697}, "tail": {"text": "it", "start": 64, "end": 66}}]}}, "schema": []}
{"input": "This paper introduces a system for categorizing unknown words. The system is based on a multi-component architecture where each component is responsible for identifying one class of unknown words. The focus of this paper is the components that identify names and spelling errors. Each component uses a decision tree architecture to combine multiple types of evidence about the unknown word. The system is evaluated using data from live closed captions-a genre replete with a wide variety of unknown words.", "output": {"entities": {"generic": [{"text": "system", "start": 24, "end": 30}, {"text": "system", "start": 67, "end": 73}, {"text": "component", "start": 94, "end": 103}, {"text": "components", "start": 228, "end": 238}, {"text": "component", "start": 128, "end": 137}, {"text": "system", "start": 395, "end": 401}], "task": [{"text": "categorizing unknown words", "start": 35, "end": 61}], "method": [{"text": "multi-component architecture", "start": 88, "end": 116}, {"text": "decision tree architecture", "start": 302, "end": 328}], "other_scientific_term": [{"text": "unknown words", "start": 48, "end": 61}, {"text": "names", "start": 253, "end": 258}, {"text": "spelling errors", "start": 263, "end": 278}, {"text": "unknown word", "start": 48, "end": 60}, {"text": "unknown words", "start": 182, "end": 195}], "material": [{"text": "live closed captions", "start": 431, "end": 451}]}, "relations": {"used_for": [{"head": {"text": "system", "start": 24, "end": 30}, "tail": {"text": "categorizing unknown words", "start": 35, "end": 61}}, {"head": {"text": "multi-component architecture", "start": 88, "end": 116}, "tail": {"text": "system", "start": 67, "end": 73}}, {"head": {"text": "component", "start": 94, "end": 103}, "tail": {"text": "unknown words", "start": 48, "end": 61}}, {"head": {"text": "components", "start": 228, "end": 238}, "tail": {"text": "names", "start": 253, "end": 258}}, {"head": {"text": "components", "start": 228, "end": 238}, "tail": {"text": "spelling errors", "start": 263, "end": 278}}, {"head": {"text": "decision tree architecture", "start": 302, "end": 328}, "tail": {"text": "component", "start": 128, "end": 137}}], "part_of": [{"head": {"text": "component", "start": 94, "end": 103}, "tail": {"text": "multi-component architecture", "start": 88, "end": 116}}], "conjunction": [{"head": {"text": "names", "start": 253, "end": 258}, "tail": {"text": "spelling errors", "start": 263, "end": 278}}], "evaluate_for": [{"head": {"text": "live closed captions", "start": 431, "end": 451}, "tail": {"text": "system", "start": 395, "end": 401}}]}}, "schema": []}
{"input": "At MIT Lincoln Laboratory, we have been developing a Korean-to-English machine translation system CCLINC -LRB- Common Coalition Language System at Lincoln Laboratory -RRB-. The CCLINC Korean-to-English translation system consists of two core modules, language understanding and generation modules mediated by a language neutral meaning representation called a semantic frame. The key features of the system include: -LRB- i -RRB- Robust efficient parsing of Korean -LRB- a verb final language with overt case markers, relatively free word order, and frequent omissions of arguments -RRB-. -LRB- ii -RRB- High quality translation via word sense disambiguation and accurate word order generation of the target language. -LRB- iii -RRB- Rapid system development and porting to new domains via knowledge-based automated acquisition of grammars. Having been trained on Korean newspaper articles on missiles and chemical biological warfare, the system produces the translation output sufficient for content understanding of the original document.", "output": {"entities": {"method": [{"text": "Korean-to-English machine translation system", "start": 53, "end": 97}, {"text": "CCLINC -LRB- Common Coalition Language System at Lincoln Laboratory -RRB-", "start": 98, "end": 171}, {"text": "CCLINC Korean-to-English translation system", "start": 177, "end": 220}, {"text": "language understanding and generation modules", "start": 251, "end": 296}, {"text": "language neutral meaning representation", "start": 311, "end": 350}, {"text": "word sense disambiguation", "start": 633, "end": 658}, {"text": "knowledge-based automated acquisition of grammars", "start": 790, "end": 839}], "generic": [{"text": "core modules", "start": 237, "end": 249}, {"text": "system", "start": 91, "end": 97}, {"text": "system", "start": 214, "end": 220}], "other_scientific_term": [{"text": "semantic frame", "start": 360, "end": 374}, {"text": "overt case markers", "start": 498, "end": 516}], "task": [{"text": "parsing of Korean", "start": 447, "end": 464}, {"text": "translation", "start": 79, "end": 90}, {"text": "word order generation", "start": 672, "end": 693}], "material": [{"text": "Korean", "start": 53, "end": 59}, {"text": "verb final language", "start": 473, "end": 492}, {"text": "Korean newspaper articles", "start": 864, "end": 889}, {"text": "missiles and chemical biological warfare", "start": 893, "end": 933}]}, "relations": {"hyponym_of": [{"head": {"text": "CCLINC -LRB- Common Coalition Language System at Lincoln Laboratory -RRB-", "start": 98, "end": 171}, "tail": {"text": "Korean-to-English machine translation system", "start": 53, "end": 97}}, {"head": {"text": "semantic frame", "start": 360, "end": 374}, "tail": {"text": "language neutral meaning representation", "start": 311, "end": 350}}, {"head": {"text": "Korean", "start": 53, "end": 59}, "tail": {"text": "verb final language", "start": 473, "end": 492}}], "part_of": [{"head": {"text": "core modules", "start": 237, "end": 249}, "tail": {"text": "CCLINC Korean-to-English translation system", "start": 177, "end": 220}}], "used_for": [{"head": {"text": "language neutral meaning representation", "start": 311, "end": 350}, "tail": {"text": "language understanding and generation modules", "start": 251, "end": 296}}, {"head": {"text": "word sense disambiguation", "start": 633, "end": 658}, "tail": {"text": "translation", "start": 79, "end": 90}}, {"head": {"text": "word order generation", "start": 672, "end": 693}, "tail": {"text": "translation", "start": 79, "end": 90}}, {"head": {"text": "Korean newspaper articles", "start": 864, "end": 889}, "tail": {"text": "system", "start": 214, "end": 220}}], "feature_of": [{"head": {"text": "overt case markers", "start": 498, "end": 516}, "tail": {"text": "verb final language", "start": 473, "end": 492}}, {"head": {"text": "missiles and chemical biological warfare", "start": 893, "end": 933}, "tail": {"text": "Korean newspaper articles", "start": 864, "end": 889}}], "conjunction": [{"head": {"text": "word sense disambiguation", "start": 633, "end": 658}, "tail": {"text": "word order generation", "start": 672, "end": 693}}]}}, "schema": []}
{"input": "The JAVELIN system integrates a flexible, planning-based architecture with a variety of language processing modules to provide an open-domain question answering capability on free text. The demonstration will focus on how JAVELIN processes questions and retrieves the most likely answer candidates from the given text corpus. The operation of the system will be explained in depth through browsing the repository of data objects created by the system during each question answering session.", "output": {"entities": {"method": [{"text": "JAVELIN system", "start": 4, "end": 18}, {"text": "planning-based architecture", "start": 42, "end": 69}, {"text": "language processing modules", "start": 88, "end": 115}, {"text": "JAVELIN", "start": 4, "end": 11}], "task": [{"text": "open-domain question answering capability", "start": 130, "end": 171}, {"text": "question answering", "start": 142, "end": 160}], "generic": [{"text": "system", "start": 12, "end": 18}, {"text": "system", "start": 347, "end": 353}]}, "relations": {"used_for": [{"head": {"text": "JAVELIN system", "start": 4, "end": 18}, "tail": {"text": "open-domain question answering capability", "start": 130, "end": 171}}], "part_of": [{"head": {"text": "planning-based architecture", "start": 42, "end": 69}, "tail": {"text": "JAVELIN system", "start": 4, "end": 18}}, {"head": {"text": "language processing modules", "start": 88, "end": 115}, "tail": {"text": "JAVELIN system", "start": 4, "end": 18}}], "conjunction": [{"head": {"text": "language processing modules", "start": 88, "end": 115}, "tail": {"text": "planning-based architecture", "start": 42, "end": 69}}]}}, "schema": []}
{"input": "We present the first application of the head-driven statistical parsing model of Collins -LRB- 1999 -RRB- as a simultaneous language model and parser for large-vocabulary speech recognition. The model is adapted to an online left to right chart-parser for word lattices, integrating acoustic, n-gram, and parser probabilities. The parser uses structural and lexical dependencies not considered by n-gram models, conditioning recognition on more linguistically-grounded relationships. Experiments on the Wall Street Journal treebank and lattice corpora show word error rates competitive with the standard n-gram language model while extracting additional structural information useful for speech understanding.", "output": {"entities": {"method": [{"text": "head-driven statistical parsing model", "start": 40, "end": 77}, {"text": "simultaneous language model", "start": 111, "end": 138}, {"text": "parser", "start": 143, "end": 149}, {"text": "online left to right chart-parser", "start": 218, "end": 251}, {"text": "parser", "start": 245, "end": 251}, {"text": "n-gram models", "start": 397, "end": 410}, {"text": "n-gram language model", "start": 604, "end": 625}], "task": [{"text": "large-vocabulary speech recognition", "start": 154, "end": 189}, {"text": "speech understanding", "start": 688, "end": 708}], "generic": [{"text": "model", "start": 72, "end": 77}], "other_scientific_term": [{"text": "word lattices", "start": 256, "end": 269}, {"text": "acoustic, n-gram, and parser probabilities", "start": 283, "end": 325}, {"text": "structural and lexical dependencies", "start": 343, "end": 378}, {"text": "structural information", "start": 654, "end": 676}], "material": [{"text": "Wall Street Journal treebank", "start": 503, "end": 531}, {"text": "lattice corpora", "start": 536, "end": 551}], "metric": [{"text": "word error rates", "start": 557, "end": 573}]}, "relations": {"used_for": [{"head": {"text": "head-driven statistical parsing model", "start": 40, "end": 77}, "tail": {"text": "simultaneous language model", "start": 111, "end": 138}}, {"head": {"text": "head-driven statistical parsing model", "start": 40, "end": 77}, "tail": {"text": "parser", "start": 143, "end": 149}}, {"head": {"text": "simultaneous language model", "start": 111, "end": 138}, "tail": {"text": "large-vocabulary speech recognition", "start": 154, "end": 189}}, {"head": {"text": "parser", "start": 143, "end": 149}, "tail": {"text": "large-vocabulary speech recognition", "start": 154, "end": 189}}, {"head": {"text": "model", "start": 72, "end": 77}, "tail": {"text": "online left to right chart-parser", "start": 218, "end": 251}}, {"head": {"text": "online left to right chart-parser", "start": 218, "end": 251}, "tail": {"text": "word lattices", "start": 256, "end": 269}}, {"head": {"text": "structural and lexical dependencies", "start": 343, "end": 378}, "tail": {"text": "parser", "start": 245, "end": 251}}, {"head": {"text": "structural information", "start": 654, "end": 676}, "tail": {"text": "speech understanding", "start": 688, "end": 708}}], "conjunction": [{"head": {"text": "simultaneous language model", "start": 111, "end": 138}, "tail": {"text": "parser", "start": 143, "end": 149}}, {"head": {"text": "Wall Street Journal treebank", "start": 503, "end": 531}, "tail": {"text": "lattice corpora", "start": 536, "end": 551}}], "part_of": [{"head": {"text": "acoustic, n-gram, and parser probabilities", "start": 283, "end": 325}, "tail": {"text": "online left to right chart-parser", "start": 218, "end": 251}}], "evaluate_for": [{"head": {"text": "Wall Street Journal treebank", "start": 503, "end": 531}, "tail": {"text": "n-gram language model", "start": 604, "end": 625}}, {"head": {"text": "lattice corpora", "start": 536, "end": 551}, "tail": {"text": "n-gram language model", "start": 604, "end": 625}}, {"head": {"text": "word error rates", "start": 557, "end": 573}, "tail": {"text": "n-gram language model", "start": 604, "end": 625}}]}}, "schema": []}
{"input": "Image composition -LRB- or mosaicing -RRB- has attracted a growing attention in recent years as one of the main elements in video analysis and representation. In this paper we deal with the problem of global alignment and super-resolution. We also propose to evaluate the quality of the resulting mosaic by measuring the amount of blurring. Global registration is achieved by combining a graph-based technique -- that exploits the topological structure of the sequence induced by the spatial overlap -- with a bundle adjustment which uses only the homographies computed in the previous steps. Experimental comparison with other techniques shows the effectiveness of our approach.", "output": {"entities": {"task": [{"text": "Image composition -LRB- or mosaicing -RRB-", "start": 0, "end": 42}, {"text": "video analysis and representation", "start": 124, "end": 157}, {"text": "global alignment", "start": 201, "end": 217}, {"text": "super-resolution", "start": 222, "end": 238}, {"text": "mosaic", "start": 27, "end": 33}, {"text": "Global registration", "start": 341, "end": 360}], "metric": [{"text": "amount of blurring", "start": 321, "end": 339}], "method": [{"text": "graph-based technique", "start": 388, "end": 409}, {"text": "bundle adjustment", "start": 510, "end": 527}], "other_scientific_term": [{"text": "topological structure", "start": 431, "end": 452}, {"text": "spatial overlap", "start": 484, "end": 499}, {"text": "homographies", "start": 548, "end": 560}], "generic": [{"text": "techniques", "start": 628, "end": 638}, {"text": "approach", "start": 670, "end": 678}]}, "relations": {"part_of": [{"head": {"text": "Image composition -LRB- or mosaicing -RRB-", "start": 0, "end": 42}, "tail": {"text": "video analysis and representation", "start": 124, "end": 157}}], "conjunction": [{"head": {"text": "global alignment", "start": 201, "end": 217}, "tail": {"text": "super-resolution", "start": 222, "end": 238}}, {"head": {"text": "graph-based technique", "start": 388, "end": 409}, "tail": {"text": "bundle adjustment", "start": 510, "end": 527}}], "evaluate_for": [{"head": {"text": "amount of blurring", "start": 321, "end": 339}, "tail": {"text": "mosaic", "start": 27, "end": 33}}], "used_for": [{"head": {"text": "graph-based technique", "start": 388, "end": 409}, "tail": {"text": "Global registration", "start": 341, "end": 360}}, {"head": {"text": "graph-based technique", "start": 388, "end": 409}, "tail": {"text": "topological structure", "start": 431, "end": 452}}, {"head": {"text": "bundle adjustment", "start": 510, "end": 527}, "tail": {"text": "Global registration", "start": 341, "end": 360}}, {"head": {"text": "homographies", "start": 548, "end": 560}, "tail": {"text": "bundle adjustment", "start": 510, "end": 527}}], "compare": [{"head": {"text": "approach", "start": 670, "end": 678}, "tail": {"text": "techniques", "start": 628, "end": 638}}]}}, "schema": []}
{"input": "The project presented here is a part of a long term research program aiming at a full lexicon grammar for Polish -LRB- SyntLex -RRB-. The main of this project is computer-assisted acquisition and morpho-syntactic description of verb-noun collocations in Polish. We present methodology and resources obtained in three main project phases which are: dictionary-based acquisition of collocation lexicon, feasibility study for corpus-based lexicon enlargement phase, corpus-based lexicon enlargement and collocation description. In this paper we focus on the results of the third phase. The presented here corpus-based approach permitted us to triple the size the verb-noun collocation dictionary for Polish. In the paper we describe the SyntLex Dictionary of Collocations and announce some future research intended to be a separate project continuation.", "output": {"entities": {"method": [{"text": "lexicon grammar for Polish -LRB- SyntLex -RRB-", "start": 86, "end": 132}, {"text": "corpus-based approach", "start": 602, "end": 623}], "task": [{"text": "computer-assisted acquisition and morpho-syntactic description of verb-noun collocations", "start": 162, "end": 250}, {"text": "dictionary-based acquisition of collocation lexicon", "start": 348, "end": 399}, {"text": "feasibility study", "start": 401, "end": 418}, {"text": "corpus-based lexicon enlargement phase", "start": 423, "end": 461}, {"text": "corpus-based lexicon enlargement and collocation description", "start": 463, "end": 523}], "material": [{"text": "Polish", "start": 106, "end": 112}, {"text": "verb-noun collocation dictionary", "start": 660, "end": 692}, {"text": "Polish", "start": 254, "end": 260}, {"text": "SyntLex Dictionary of Collocations", "start": 734, "end": 768}], "generic": [{"text": "phases", "start": 330, "end": 336}]}, "relations": {"used_for": [{"head": {"text": "Polish", "start": 106, "end": 112}, "tail": {"text": "computer-assisted acquisition and morpho-syntactic description of verb-noun collocations", "start": 162, "end": 250}}, {"head": {"text": "feasibility study", "start": 401, "end": 418}, "tail": {"text": "corpus-based lexicon enlargement phase", "start": 423, "end": 461}}, {"head": {"text": "corpus-based approach", "start": 602, "end": 623}, "tail": {"text": "verb-noun collocation dictionary", "start": 660, "end": 692}}], "hyponym_of": [{"head": {"text": "dictionary-based acquisition of collocation lexicon", "start": 348, "end": 399}, "tail": {"text": "phases", "start": 330, "end": 336}}, {"head": {"text": "feasibility study", "start": 401, "end": 418}, "tail": {"text": "phases", "start": 330, "end": 336}}, {"head": {"text": "corpus-based lexicon enlargement and collocation description", "start": 463, "end": 523}, "tail": {"text": "phases", "start": 330, "end": 336}}], "conjunction": [{"head": {"text": "dictionary-based acquisition of collocation lexicon", "start": 348, "end": 399}, "tail": {"text": "feasibility study", "start": 401, "end": 418}}, {"head": {"text": "corpus-based lexicon enlargement and collocation description", "start": 463, "end": 523}, "tail": {"text": "feasibility study", "start": 401, "end": 418}}], "feature_of": [{"head": {"text": "Polish", "start": 254, "end": 260}, "tail": {"text": "verb-noun collocation dictionary", "start": 660, "end": 692}}]}}, "schema": []}
{"input": "Along with the increasing requirements, the hash-tag recommendation task for microblogs has been receiving considerable attention in recent years. Various researchers have studied the problem from different aspects. However, most of these methods usually need handcrafted features. Motivated by the successful use of convolutional neural networks -LRB- CNNs -RRB- for many natural language processing tasks, in this paper, we adopt CNNs to perform the hashtag recommendation problem. To incorporate the trigger words whose effectiveness have been experimentally evaluated in several previous works, we propose a novel architecture with an attention mechanism. The results of experiments on the data collected from a real world microblogging service demonstrated that the proposed model outperforms state-of-the-art methods. By incorporating trigger words into the consideration, the relative improvement of the proposed method over the state-of-the-art method is around 9.4% in the F1-score.", "output": {"entities": {"task": [{"text": "hash-tag recommendation task", "start": 44, "end": 72}, {"text": "natural language processing tasks", "start": 373, "end": 406}, {"text": "hashtag recommendation problem", "start": 452, "end": 482}], "material": [{"text": "microblogs", "start": 77, "end": 87}], "other_scientific_term": [{"text": "handcrafted features", "start": 260, "end": 280}, {"text": "trigger words", "start": 503, "end": 516}, {"text": "attention mechanism", "start": 639, "end": 658}, {"text": "trigger words", "start": 841, "end": 854}], "method": [{"text": "convolutional neural networks -LRB- CNNs -RRB-", "start": 317, "end": 363}, {"text": "CNNs", "start": 353, "end": 357}], "generic": [{"text": "architecture", "start": 618, "end": 630}, {"text": "data", "start": 694, "end": 698}, {"text": "model", "start": 780, "end": 785}, {"text": "state-of-the-art methods", "start": 798, "end": 822}, {"text": "method", "start": 239, "end": 245}, {"text": "state-of-the-art method", "start": 798, "end": 821}], "metric": [{"text": "F1-score", "start": 982, "end": 990}]}, "relations": {"used_for": [{"head": {"text": "hash-tag recommendation task", "start": 44, "end": 72}, "tail": {"text": "microblogs", "start": 77, "end": 87}}, {"head": {"text": "convolutional neural networks -LRB- CNNs -RRB-", "start": 317, "end": 363}, "tail": {"text": "natural language processing tasks", "start": 373, "end": 406}}, {"head": {"text": "architecture", "start": 618, "end": 630}, "tail": {"text": "trigger words", "start": 503, "end": 516}}], "feature_of": [{"head": {"text": "attention mechanism", "start": 639, "end": 658}, "tail": {"text": "architecture", "start": 618, "end": 630}}], "evaluate_for": [{"head": {"text": "data", "start": 694, "end": 698}, "tail": {"text": "model", "start": 780, "end": 785}}, {"head": {"text": "F1-score", "start": 982, "end": 990}, "tail": {"text": "state-of-the-art method", "start": 798, "end": 821}}], "compare": [{"head": {"text": "model", "start": 780, "end": 785}, "tail": {"text": "state-of-the-art methods", "start": 798, "end": 822}}, {"head": {"text": "method", "start": 239, "end": 245}, "tail": {"text": "state-of-the-art method", "start": 798, "end": 821}}]}}, "schema": []}
{"input": "In this paper, we improve an unsupervised learning method using the Expectation-Maximization -LRB- EM -RRB- algorithm proposed by Nigam et al. for text classification problems in order to apply it to word sense disambiguation -LRB- WSD -RRB- problems. The improved method stops the EM algorithm at the optimum iteration number. To estimate that number, we propose two methods. In experiments, we solved 50 noun WSD problems in the Japanese Dictionary Task in SENSEVAL2. The score of our method is a match for the best public score of this task. Furthermore, our methods were confirmed to be effective also for verb WSD problems.", "output": {"entities": {"method": [{"text": "unsupervised learning method", "start": 29, "end": 57}, {"text": "Expectation-Maximization -LRB- EM -RRB- algorithm", "start": 68, "end": 117}, {"text": "EM algorithm", "start": 282, "end": 294}], "task": [{"text": "text classification problems", "start": 147, "end": 175}, {"text": "word sense disambiguation -LRB- WSD -RRB- problems", "start": 200, "end": 250}, {"text": "noun WSD problems", "start": 406, "end": 423}, {"text": "Japanese Dictionary Task", "start": 431, "end": 455}, {"text": "verb WSD problems", "start": 610, "end": 627}], "generic": [{"text": "it", "start": 113, "end": 115}, {"text": "method", "start": 51, "end": 57}, {"text": "number", "start": 320, "end": 326}, {"text": "method", "start": 265, "end": 271}, {"text": "task", "start": 539, "end": 543}, {"text": "methods", "start": 368, "end": 375}], "other_scientific_term": [{"text": "optimum iteration number", "start": 302, "end": 326}], "material": [{"text": "SENSEVAL2", "start": 459, "end": 468}]}, "relations": {"used_for": [{"head": {"text": "Expectation-Maximization -LRB- EM -RRB- algorithm", "start": 68, "end": 117}, "tail": {"text": "unsupervised learning method", "start": 29, "end": 57}}, {"head": {"text": "Expectation-Maximization -LRB- EM -RRB- algorithm", "start": 68, "end": 117}, "tail": {"text": "text classification problems", "start": 147, "end": 175}}, {"head": {"text": "it", "start": 113, "end": 115}, "tail": {"text": "word sense disambiguation -LRB- WSD -RRB- problems", "start": 200, "end": 250}}, {"head": {"text": "methods", "start": 368, "end": 375}, "tail": {"text": "verb WSD problems", "start": 610, "end": 627}}], "feature_of": [{"head": {"text": "Japanese Dictionary Task", "start": 431, "end": 455}, "tail": {"text": "SENSEVAL2", "start": 459, "end": 468}}]}}, "schema": []}
{"input": "Dividing sentences in chunks of words is a useful preprocessing step for parsing, information extraction and information retrieval. -LRB- Ramshaw and Marcus, 1995 -RRB- have introduced a ``convenient'' data representation for chunking by converting it to a tagging task. In this paper we will examine seven different data representations for the problem of recognizing noun phrase chunks. We will show that the data representation choice has a minor influence on chunking performance. However, equipped with the most suitable data representation, our memory-based learning chunker was able to improve the best published chunking results for a standard data set.", "output": {"entities": {"task": [{"text": "Dividing sentences in chunks of words", "start": 0, "end": 37}, {"text": "parsing", "start": 73, "end": 80}, {"text": "information extraction", "start": 82, "end": 104}, {"text": "information retrieval", "start": 109, "end": 130}, {"text": "chunking", "start": 226, "end": 234}, {"text": "tagging task", "start": 257, "end": 269}, {"text": "recognizing noun phrase chunks", "start": 357, "end": 387}, {"text": "chunking", "start": 463, "end": 471}, {"text": "chunking", "start": 620, "end": 628}], "method": [{"text": "data representation", "start": 202, "end": 221}, {"text": "data representations", "start": 317, "end": 337}, {"text": "data representation", "start": 317, "end": 336}, {"text": "data representation", "start": 411, "end": 430}, {"text": "memory-based learning chunker", "start": 551, "end": 580}], "generic": [{"text": "it", "start": 249, "end": 251}], "material": [{"text": "data set", "start": 652, "end": 660}]}, "relations": {"used_for": [{"head": {"text": "Dividing sentences in chunks of words", "start": 0, "end": 37}, "tail": {"text": "parsing", "start": 73, "end": 80}}, {"head": {"text": "Dividing sentences in chunks of words", "start": 0, "end": 37}, "tail": {"text": "information extraction", "start": 82, "end": 104}}, {"head": {"text": "Dividing sentences in chunks of words", "start": 0, "end": 37}, "tail": {"text": "information retrieval", "start": 109, "end": 130}}, {"head": {"text": "data representation", "start": 202, "end": 221}, "tail": {"text": "chunking", "start": 226, "end": 234}}, {"head": {"text": "data representations", "start": 317, "end": 337}, "tail": {"text": "recognizing noun phrase chunks", "start": 357, "end": 387}}, {"head": {"text": "data representation", "start": 411, "end": 430}, "tail": {"text": "memory-based learning chunker", "start": 551, "end": 580}}], "conjunction": [{"head": {"text": "parsing", "start": 73, "end": 80}, "tail": {"text": "information extraction", "start": 82, "end": 104}}, {"head": {"text": "information extraction", "start": 82, "end": 104}, "tail": {"text": "information retrieval", "start": 109, "end": 130}}], "evaluate_for": [{"head": {"text": "data set", "start": 652, "end": 660}, "tail": {"text": "memory-based learning chunker", "start": 551, "end": 580}}]}}, "schema": []}
{"input": "In this paper we describe and evaluate a Question Answering system that goes beyond answering factoid questions. We focus on FAQ-like questions and answers, and build our system around a noisy-channel architecture which exploits both a language model for answers and a transformation model for answer/question terms, trained on a corpus of 1 million question/answer pairs collected from the Web.", "output": {"entities": {"method": [{"text": "Question Answering system", "start": 41, "end": 66}, {"text": "noisy-channel architecture", "start": 187, "end": 213}, {"text": "language model", "start": 236, "end": 250}, {"text": "transformation model", "start": 269, "end": 289}], "material": [{"text": "FAQ-like questions and answers", "start": 125, "end": 155}, {"text": "Web", "start": 391, "end": 394}], "generic": [{"text": "system", "start": 60, "end": 66}]}, "relations": {"used_for": [{"head": {"text": "system", "start": 60, "end": 66}, "tail": {"text": "FAQ-like questions and answers", "start": 125, "end": 155}}, {"head": {"text": "noisy-channel architecture", "start": 187, "end": 213}, "tail": {"text": "system", "start": 60, "end": 66}}, {"head": {"text": "noisy-channel architecture", "start": 187, "end": 213}, "tail": {"text": "language model", "start": 236, "end": 250}}, {"head": {"text": "noisy-channel architecture", "start": 187, "end": 213}, "tail": {"text": "transformation model", "start": 269, "end": 289}}]}}, "schema": []}
{"input": "In this paper we evaluate four objective measures of speech with regards to intelligibility prediction of synthesized speech in diverse noisy situations. We evaluated three intel-ligibility measures, the Dau measure, the glimpse proportion and the Speech Intelligibility Index -LRB- SII -RRB- and a quality measure, the Perceptual Evaluation of Speech Quality -LRB- PESQ -RRB-. For the generation of synthesized speech we used a state of the art HMM-based speech synthesis system. The noisy conditions comprised four additive noises. The measures were compared with subjective intelligibility scores obtained in listening tests. The results show the Dau and the glimpse measures to be the best predictors of intelligibility, with correlations of around 0.83 to subjective scores. All measures gave less accurate predictions of intelligibility for synthetic speech than have previously been found for natural speech; in particular the SII measure. In additional experiments, we processed the synthesized speech by an ideal binary mask before adding noise. The Glimpse measure gave the most accurate intelligibility predictions in this situation.", "output": {"entities": {"metric": [{"text": "measures of speech", "start": 41, "end": 59}, {"text": "intel-ligibility measures", "start": 173, "end": 198}, {"text": "Dau measure", "start": 204, "end": 215}, {"text": "glimpse proportion", "start": 221, "end": 239}, {"text": "Speech Intelligibility Index -LRB- SII -RRB-", "start": 248, "end": 292}, {"text": "quality measure", "start": 299, "end": 314}, {"text": "Perceptual Evaluation of Speech Quality -LRB- PESQ -RRB-", "start": 320, "end": 376}, {"text": "subjective intelligibility scores", "start": 566, "end": 599}, {"text": "Dau", "start": 204, "end": 207}, {"text": "glimpse measures", "start": 662, "end": 678}, {"text": "correlations", "start": 730, "end": 742}, {"text": "SII measure", "start": 934, "end": 945}, {"text": "Glimpse measure", "start": 1059, "end": 1074}], "task": [{"text": "intelligibility prediction", "start": 76, "end": 102}, {"text": "generation of synthesized speech", "start": 386, "end": 418}, {"text": "predictions of intelligibility", "start": 812, "end": 842}, {"text": "intelligibility predictions", "start": 1098, "end": 1125}], "material": [{"text": "synthesized speech", "start": 106, "end": 124}, {"text": "synthetic speech", "start": 847, "end": 863}, {"text": "natural speech", "start": 900, "end": 914}, {"text": "synthesized speech", "start": 400, "end": 418}], "other_scientific_term": [{"text": "diverse noisy situations", "start": 128, "end": 152}, {"text": "noisy conditions", "start": 485, "end": 501}, {"text": "additive noises", "start": 517, "end": 532}, {"text": "subjective scores", "start": 761, "end": 778}], "method": [{"text": "HMM-based speech synthesis system", "start": 446, "end": 479}, {"text": "predictors of intelligibility", "start": 694, "end": 723}, {"text": "ideal binary mask", "start": 1016, "end": 1033}], "generic": [{"text": "measures", "start": 41, "end": 49}, {"text": "measures", "start": 190, "end": 198}]}, "relations": {"evaluate_for": [{"head": {"text": "measures of speech", "start": 41, "end": 59}, "tail": {"text": "intelligibility prediction", "start": 76, "end": 102}}, {"head": {"text": "correlations", "start": 730, "end": 742}, "tail": {"text": "Dau", "start": 204, "end": 207}}, {"head": {"text": "correlations", "start": 730, "end": 742}, "tail": {"text": "glimpse measures", "start": 662, "end": 678}}, {"head": {"text": "measures", "start": 190, "end": 198}, "tail": {"text": "predictions of intelligibility", "start": 812, "end": 842}}], "used_for": [{"head": {"text": "synthesized speech", "start": 106, "end": 124}, "tail": {"text": "intelligibility prediction", "start": 76, "end": 102}}, {"head": {"text": "HMM-based speech synthesis system", "start": 446, "end": 479}, "tail": {"text": "generation of synthesized speech", "start": 386, "end": 418}}, {"head": {"text": "synthetic speech", "start": 847, "end": 863}, "tail": {"text": "predictions of intelligibility", "start": 812, "end": 842}}, {"head": {"text": "ideal binary mask", "start": 1016, "end": 1033}, "tail": {"text": "synthesized speech", "start": 400, "end": 418}}, {"head": {"text": "Glimpse measure", "start": 1059, "end": 1074}, "tail": {"text": "intelligibility predictions", "start": 1098, "end": 1125}}], "feature_of": [{"head": {"text": "diverse noisy situations", "start": 128, "end": 152}, "tail": {"text": "synthesized speech", "start": 106, "end": 124}}], "conjunction": [{"head": {"text": "intel-ligibility measures", "start": 173, "end": 198}, "tail": {"text": "quality measure", "start": 299, "end": 314}}, {"head": {"text": "Dau measure", "start": 204, "end": 215}, "tail": {"text": "glimpse proportion", "start": 221, "end": 239}}, {"head": {"text": "glimpse proportion", "start": 221, "end": 239}, "tail": {"text": "Speech Intelligibility Index -LRB- SII -RRB-", "start": 248, "end": 292}}, {"head": {"text": "Dau", "start": 204, "end": 207}, "tail": {"text": "glimpse measures", "start": 662, "end": 678}}], "hyponym_of": [{"head": {"text": "Dau measure", "start": 204, "end": 215}, "tail": {"text": "intel-ligibility measures", "start": 173, "end": 198}}, {"head": {"text": "glimpse proportion", "start": 221, "end": 239}, "tail": {"text": "intel-ligibility measures", "start": 173, "end": 198}}, {"head": {"text": "Speech Intelligibility Index -LRB- SII -RRB-", "start": 248, "end": 292}, "tail": {"text": "intel-ligibility measures", "start": 173, "end": 198}}, {"head": {"text": "Perceptual Evaluation of Speech Quality -LRB- PESQ -RRB-", "start": 320, "end": 376}, "tail": {"text": "quality measure", "start": 299, "end": 314}}, {"head": {"text": "Dau", "start": 204, "end": 207}, "tail": {"text": "predictors of intelligibility", "start": 694, "end": 723}}, {"head": {"text": "glimpse measures", "start": 662, "end": 678}, "tail": {"text": "predictors of intelligibility", "start": 694, "end": 723}}, {"head": {"text": "SII measure", "start": 934, "end": 945}, "tail": {"text": "measures", "start": 190, "end": 198}}], "part_of": [{"head": {"text": "additive noises", "start": 517, "end": 532}, "tail": {"text": "noisy conditions", "start": 485, "end": 501}}], "compare": [{"head": {"text": "measures", "start": 41, "end": 49}, "tail": {"text": "subjective intelligibility scores", "start": 566, "end": 599}}, {"head": {"text": "Dau", "start": 204, "end": 207}, "tail": {"text": "subjective scores", "start": 761, "end": 778}}, {"head": {"text": "glimpse measures", "start": 662, "end": 678}, "tail": {"text": "subjective scores", "start": 761, "end": 778}}, {"head": {"text": "synthetic speech", "start": 847, "end": 863}, "tail": {"text": "natural speech", "start": 900, "end": 914}}]}}, "schema": []}
{"input": "A'' graphics for vision'' approach is proposed to address the problem of reconstruction from a large and imperfect data set: reconstruction on demand by tensor voting, or ROD-TV. ROD-TV simultaneously delivers good efficiency and robust-ness, by adapting to a continuum of primitive connectivity, view dependence, and levels of detail -LRB- LOD -RRB-. Locally inferred surface elements are robust to noise and better capture local shapes. By inferring per-vertex normals at sub-voxel precision on the fly, we can achieve interpolative shading. Since these missing details can be recovered at the current level of detail, our result is not upper bounded by the scanning resolution. By relaxing the mesh connectivity requirement, we extend ROD-TV and propose a simple but effective multiscale feature extraction algorithm. ROD-TV consists of a hierarchical data structure that encodes different levels of detail. The local reconstruction algorithm is tensor voting. It is applied on demand to the visible subset of data at a desired level of detail, by traversing the data hierarchy and collecting tensorial support in a neighborhood. We compare our approach and present encouraging results.", "output": {"entities": {"method": [{"text": "'' graphics for vision'' approach", "start": 1, "end": 34}, {"text": "tensor voting", "start": 153, "end": 166}, {"text": "ROD-TV", "start": 171, "end": 177}, {"text": "ROD-TV", "start": 179, "end": 185}, {"text": "ROD-TV", "start": 738, "end": 744}, {"text": "multiscale feature extraction algorithm", "start": 780, "end": 819}, {"text": "ROD-TV", "start": 821, "end": 827}, {"text": "hierarchical data structure", "start": 842, "end": 869}, {"text": "local reconstruction algorithm", "start": 915, "end": 945}, {"text": "tensor voting", "start": 949, "end": 962}, {"text": "traversing the data hierarchy", "start": 1051, "end": 1080}, {"text": "collecting tensorial support", "start": 1085, "end": 1113}], "task": [{"text": "reconstruction", "start": 73, "end": 87}, {"text": "reconstruction", "start": 125, "end": 139}, {"text": "interpolative shading", "start": 521, "end": 542}], "material": [{"text": "large and imperfect data set", "start": 95, "end": 123}], "metric": [{"text": "efficiency", "start": 215, "end": 225}, {"text": "robust-ness", "start": 230, "end": 241}, {"text": "sub-voxel precision", "start": 474, "end": 493}], "other_scientific_term": [{"text": "primitive connectivity", "start": 273, "end": 295}, {"text": "view dependence", "start": 297, "end": 312}, {"text": "levels of detail -LRB- LOD -RRB-", "start": 318, "end": 350}, {"text": "Locally inferred surface elements", "start": 352, "end": 385}, {"text": "noise", "start": 400, "end": 405}, {"text": "local shapes", "start": 425, "end": 437}, {"text": "per-vertex normals", "start": 452, "end": 470}, {"text": "scanning resolution", "start": 660, "end": 679}, {"text": "mesh connectivity requirement", "start": 697, "end": 726}], "generic": [{"text": "It", "start": 964, "end": 966}, {"text": "approach", "start": 26, "end": 34}]}, "relations": {"used_for": [{"head": {"text": "'' graphics for vision'' approach", "start": 1, "end": 34}, "tail": {"text": "reconstruction", "start": 73, "end": 87}}, {"head": {"text": "large and imperfect data set", "start": 95, "end": 123}, "tail": {"text": "reconstruction", "start": 73, "end": 87}}, {"head": {"text": "tensor voting", "start": 153, "end": 166}, "tail": {"text": "reconstruction", "start": 125, "end": 139}}, {"head": {"text": "ROD-TV", "start": 171, "end": 177}, "tail": {"text": "reconstruction", "start": 125, "end": 139}}, {"head": {"text": "Locally inferred surface elements", "start": 352, "end": 385}, "tail": {"text": "local shapes", "start": 425, "end": 437}}, {"head": {"text": "per-vertex normals", "start": 452, "end": 470}, "tail": {"text": "interpolative shading", "start": 521, "end": 542}}, {"head": {"text": "mesh connectivity requirement", "start": 697, "end": 726}, "tail": {"text": "multiscale feature extraction algorithm", "start": 780, "end": 819}}, {"head": {"text": "ROD-TV", "start": 738, "end": 744}, "tail": {"text": "multiscale feature extraction algorithm", "start": 780, "end": 819}}, {"head": {"text": "traversing the data hierarchy", "start": 1051, "end": 1080}, "tail": {"text": "It", "start": 964, "end": 966}}, {"head": {"text": "collecting tensorial support", "start": 1085, "end": 1113}, "tail": {"text": "It", "start": 964, "end": 966}}], "conjunction": [{"head": {"text": "tensor voting", "start": 153, "end": 166}, "tail": {"text": "ROD-TV", "start": 171, "end": 177}}, {"head": {"text": "robust-ness", "start": 230, "end": 241}, "tail": {"text": "efficiency", "start": 215, "end": 225}}, {"head": {"text": "view dependence", "start": 297, "end": 312}, "tail": {"text": "primitive connectivity", "start": 273, "end": 295}}, {"head": {"text": "levels of detail -LRB- LOD -RRB-", "start": 318, "end": 350}, "tail": {"text": "view dependence", "start": 297, "end": 312}}, {"head": {"text": "traversing the data hierarchy", "start": 1051, "end": 1080}, "tail": {"text": "collecting tensorial support", "start": 1085, "end": 1113}}], "evaluate_for": [{"head": {"text": "efficiency", "start": 215, "end": 225}, "tail": {"text": "ROD-TV", "start": 179, "end": 185}}, {"head": {"text": "robust-ness", "start": 230, "end": 241}, "tail": {"text": "ROD-TV", "start": 179, "end": 185}}], "feature_of": [{"head": {"text": "sub-voxel precision", "start": 474, "end": 493}, "tail": {"text": "per-vertex normals", "start": 452, "end": 470}}], "part_of": [{"head": {"text": "hierarchical data structure", "start": 842, "end": 869}, "tail": {"text": "ROD-TV", "start": 821, "end": 827}}], "hyponym_of": [{"head": {"text": "tensor voting", "start": 949, "end": 962}, "tail": {"text": "local reconstruction algorithm", "start": 915, "end": 945}}]}}, "schema": []}
{"input": "Both rhetorical structure and punctuation have been helpful in discourse processing. Based on a corpus annotation project, this paper reports the discursive usage of 6 Chinese punctuation marks in news commentary texts: Colon, Dash, Ellipsis, Exclamation Mark, Question Mark, and Semicolon. The rhetorical patterns of these marks are compared against patterns around cue phrases in general. Results show that these Chinese punctuation marks, though fewer in number than cue phrases, are easy to identify, have strong correlation with certain relations, and can be used as distinctive indicators of nuclearity in Chinese texts.", "output": {"entities": {"other_scientific_term": [{"text": "rhetorical structure", "start": 5, "end": 25}, {"text": "punctuation", "start": 30, "end": 41}, {"text": "Chinese punctuation marks", "start": 168, "end": 193}, {"text": "Colon", "start": 220, "end": 225}, {"text": "Dash", "start": 227, "end": 231}, {"text": "Ellipsis", "start": 233, "end": 241}, {"text": "Exclamation Mark", "start": 243, "end": 259}, {"text": "Question Mark", "start": 261, "end": 274}, {"text": "Semicolon", "start": 280, "end": 289}, {"text": "rhetorical patterns", "start": 295, "end": 314}, {"text": "patterns around cue phrases", "start": 351, "end": 378}, {"text": "Chinese punctuation marks", "start": 415, "end": 440}, {"text": "cue phrases", "start": 367, "end": 378}, {"text": "indicators of nuclearity", "start": 584, "end": 608}], "task": [{"text": "discourse processing", "start": 63, "end": 83}, {"text": "corpus annotation project", "start": 96, "end": 121}, {"text": "discursive usage of 6 Chinese punctuation marks", "start": 146, "end": 193}], "material": [{"text": "news commentary texts", "start": 197, "end": 218}, {"text": "Chinese texts", "start": 612, "end": 625}], "generic": [{"text": "marks", "start": 188, "end": 193}]}, "relations": {"conjunction": [{"head": {"text": "rhetorical structure", "start": 5, "end": 25}, "tail": {"text": "punctuation", "start": 30, "end": 41}}, {"head": {"text": "Colon", "start": 220, "end": 225}, "tail": {"text": "Dash", "start": 227, "end": 231}}, {"head": {"text": "Dash", "start": 227, "end": 231}, "tail": {"text": "Ellipsis", "start": 233, "end": 241}}, {"head": {"text": "Ellipsis", "start": 233, "end": 241}, "tail": {"text": "Exclamation Mark", "start": 243, "end": 259}}, {"head": {"text": "Exclamation Mark", "start": 243, "end": 259}, "tail": {"text": "Question Mark", "start": 261, "end": 274}}, {"head": {"text": "Question Mark", "start": 261, "end": 274}, "tail": {"text": "Semicolon", "start": 280, "end": 289}}], "used_for": [{"head": {"text": "rhetorical structure", "start": 5, "end": 25}, "tail": {"text": "discourse processing", "start": 63, "end": 83}}, {"head": {"text": "punctuation", "start": 30, "end": 41}, "tail": {"text": "discourse processing", "start": 63, "end": 83}}, {"head": {"text": "Chinese punctuation marks", "start": 415, "end": 440}, "tail": {"text": "indicators of nuclearity", "start": 584, "end": 608}}], "part_of": [{"head": {"text": "Chinese punctuation marks", "start": 168, "end": 193}, "tail": {"text": "news commentary texts", "start": 197, "end": 218}}], "hyponym_of": [{"head": {"text": "Colon", "start": 220, "end": 225}, "tail": {"text": "Chinese punctuation marks", "start": 168, "end": 193}}, {"head": {"text": "Dash", "start": 227, "end": 231}, "tail": {"text": "Chinese punctuation marks", "start": 168, "end": 193}}, {"head": {"text": "Ellipsis", "start": 233, "end": 241}, "tail": {"text": "Chinese punctuation marks", "start": 168, "end": 193}}, {"head": {"text": "Exclamation Mark", "start": 243, "end": 259}, "tail": {"text": "Chinese punctuation marks", "start": 168, "end": 193}}, {"head": {"text": "Question Mark", "start": 261, "end": 274}, "tail": {"text": "Chinese punctuation marks", "start": 168, "end": 193}}, {"head": {"text": "Semicolon", "start": 280, "end": 289}, "tail": {"text": "Chinese punctuation marks", "start": 168, "end": 193}}], "feature_of": [{"head": {"text": "rhetorical patterns", "start": 295, "end": 314}, "tail": {"text": "marks", "start": 188, "end": 193}}, {"head": {"text": "Chinese texts", "start": 612, "end": 625}, "tail": {"text": "indicators of nuclearity", "start": 584, "end": 608}}], "compare": [{"head": {"text": "rhetorical patterns", "start": 295, "end": 314}, "tail": {"text": "patterns around cue phrases", "start": 351, "end": 378}}, {"head": {"text": "Chinese punctuation marks", "start": 415, "end": 440}, "tail": {"text": "cue phrases", "start": 367, "end": 378}}]}}, "schema": []}
{"input": "The features based on Markov random field -LRB- MRF -RRB- models are usually sensitive to the rotation of image textures. This paper develops an anisotropic circular Gaussian MRF -LRB- ACGMRF -RRB- model for modelling rotated image textures and retrieving rotation-invariant texture features. To overcome the singularity problem of the least squares estimate -LRB- LSE -RRB- method, an approximate least squares estimate -LRB- ALSE -RRB- method is proposed to estimate the parameters of the ACGMRF model. The rotation-invariant features can be obtained from the parameters of the ACGMRF model by the one-dimensional -LRB- 1-D -RRB- discrete Fourier transform -LRB- DFT -RRB-. Significantly improved accuracy can be achieved by applying the rotation-invariant features to classify SAR -LRB- synthetic aperture radar -RRB- sea ice and Brodatz imagery.", "output": {"entities": {"other_scientific_term": [{"text": "features", "start": 4, "end": 12}, {"text": "rotation of image textures", "start": 94, "end": 120}, {"text": "singularity problem", "start": 309, "end": 328}, {"text": "parameters of the ACGMRF model", "start": 473, "end": 503}, {"text": "rotation-invariant features", "start": 509, "end": 536}, {"text": "parameters of the ACGMRF model", "start": 562, "end": 592}, {"text": "rotation-invariant features", "start": 740, "end": 767}, {"text": "SAR -LRB- synthetic aperture radar", "start": 780, "end": 814}], "method": [{"text": "Markov random field -LRB- MRF -RRB- models", "start": 22, "end": 64}, {"text": "anisotropic circular Gaussian MRF -LRB- ACGMRF -RRB- model", "start": 145, "end": 203}, {"text": "least squares estimate -LRB- LSE -RRB- method", "start": 336, "end": 381}, {"text": "approximate least squares estimate -LRB- ALSE -RRB- method", "start": 386, "end": 444}, {"text": "ACGMRF model", "start": 491, "end": 503}, {"text": "ACGMRF model", "start": 580, "end": 592}, {"text": "one-dimensional -LRB- 1-D -RRB- discrete Fourier transform -LRB- DFT -RRB-", "start": 600, "end": 674}], "task": [{"text": "modelling rotated image textures", "start": 208, "end": 240}, {"text": "retrieving rotation-invariant texture features", "start": 245, "end": 291}], "metric": [{"text": "accuracy", "start": 699, "end": 707}]}, "relations": {"used_for": [{"head": {"text": "Markov random field -LRB- MRF -RRB- models", "start": 22, "end": 64}, "tail": {"text": "features", "start": 4, "end": 12}}, {"head": {"text": "anisotropic circular Gaussian MRF -LRB- ACGMRF -RRB- model", "start": 145, "end": 203}, "tail": {"text": "modelling rotated image textures", "start": 208, "end": 240}}, {"head": {"text": "anisotropic circular Gaussian MRF -LRB- ACGMRF -RRB- model", "start": 145, "end": 203}, "tail": {"text": "retrieving rotation-invariant texture features", "start": 245, "end": 291}}, {"head": {"text": "approximate least squares estimate -LRB- ALSE -RRB- method", "start": 386, "end": 444}, "tail": {"text": "parameters of the ACGMRF model", "start": 473, "end": 503}}, {"head": {"text": "parameters of the ACGMRF model", "start": 562, "end": 592}, "tail": {"text": "rotation-invariant features", "start": 509, "end": 536}}, {"head": {"text": "one-dimensional -LRB- 1-D -RRB- discrete Fourier transform -LRB- DFT -RRB-", "start": 600, "end": 674}, "tail": {"text": "rotation-invariant features", "start": 509, "end": 536}}, {"head": {"text": "rotation-invariant features", "start": 740, "end": 767}, "tail": {"text": "SAR -LRB- synthetic aperture radar", "start": 780, "end": 814}}], "conjunction": [{"head": {"text": "modelling rotated image textures", "start": 208, "end": 240}, "tail": {"text": "retrieving rotation-invariant texture features", "start": 245, "end": 291}}], "feature_of": [{"head": {"text": "singularity problem", "start": 309, "end": 328}, "tail": {"text": "least squares estimate -LRB- LSE -RRB- method", "start": 336, "end": 381}}]}}, "schema": []}
{"input": "Despite much recent progress on accurate semantic role labeling, previous work has largely used independent classifiers, possibly combined with separate label sequence models via Viterbi decoding. This stands in stark contrast to the linguistic observation that a core argument frame is a joint structure, with strong dependencies between arguments. We show how to build a joint model of argument frames, incorporating novel features that model these interactions into discriminative log-linear models. This system achieves an error reduction of 22% on all arguments and 32% on core arguments over a state-of-the art independent classifier for gold-standard parse trees on PropBank.", "output": {"entities": {"task": [{"text": "semantic role labeling", "start": 41, "end": 63}], "method": [{"text": "independent classifiers", "start": 96, "end": 119}, {"text": "label sequence models", "start": 153, "end": 174}, {"text": "Viterbi decoding", "start": 179, "end": 195}, {"text": "joint model of argument frames", "start": 373, "end": 403}, {"text": "discriminative log-linear models", "start": 469, "end": 501}, {"text": "independent classifier", "start": 96, "end": 118}], "other_scientific_term": [{"text": "core argument frame", "start": 264, "end": 283}, {"text": "features", "start": 425, "end": 433}, {"text": "gold-standard parse trees", "start": 644, "end": 669}], "generic": [{"text": "system", "start": 508, "end": 514}], "metric": [{"text": "error reduction", "start": 527, "end": 542}], "material": [{"text": "PropBank", "start": 673, "end": 681}]}, "relations": {"used_for": [{"head": {"text": "independent classifiers", "start": 96, "end": 119}, "tail": {"text": "semantic role labeling", "start": 41, "end": 63}}, {"head": {"text": "Viterbi decoding", "start": 179, "end": 195}, "tail": {"text": "label sequence models", "start": 153, "end": 174}}], "conjunction": [{"head": {"text": "independent classifiers", "start": 96, "end": 119}, "tail": {"text": "label sequence models", "start": 153, "end": 174}}], "part_of": [{"head": {"text": "features", "start": 425, "end": 433}, "tail": {"text": "discriminative log-linear models", "start": 469, "end": 501}}, {"head": {"text": "gold-standard parse trees", "start": 644, "end": 669}, "tail": {"text": "PropBank", "start": 673, "end": 681}}], "evaluate_for": [{"head": {"text": "error reduction", "start": 527, "end": 542}, "tail": {"text": "system", "start": 508, "end": 514}}, {"head": {"text": "error reduction", "start": 527, "end": 542}, "tail": {"text": "independent classifier", "start": 96, "end": 118}}, {"head": {"text": "gold-standard parse trees", "start": 644, "end": 669}, "tail": {"text": "system", "start": 508, "end": 514}}, {"head": {"text": "gold-standard parse trees", "start": 644, "end": 669}, "tail": {"text": "independent classifier", "start": 96, "end": 118}}], "compare": [{"head": {"text": "independent classifier", "start": 96, "end": 118}, "tail": {"text": "system", "start": 508, "end": 514}}]}}, "schema": []}
{"input": "One of the major problems one is faced with when decomposing words into their constituent parts is ambiguity: the generation of multiple analyses for one input word, many of which are implausible. In order to deal with ambiguity, the MORphological PArser MORPA is provided with a probabilistic context-free grammar -LRB- PCFG -RRB-, i.e. it combines a ``conventional'' context-free morphological grammar to filter out ungrammatical segmentations with a probability-based scoring function which determines the likelihood of each successful parse. Consequently, remaining analyses can be ordered along a scale of plausibility. Test performance data will show that a PCFG yields good results in morphological parsing. MORPA is a fully implemented parser developed for use in a text-to-speech conversion system.", "output": {"entities": {"other_scientific_term": [{"text": "ambiguity", "start": 99, "end": 108}, {"text": "ambiguity", "start": 219, "end": 228}, {"text": "ungrammatical segmentations", "start": 418, "end": 445}], "task": [{"text": "generation", "start": 114, "end": 124}, {"text": "parse", "start": 539, "end": 544}, {"text": "morphological parsing", "start": 692, "end": 713}, {"text": "text-to-speech conversion system", "start": 774, "end": 806}], "method": [{"text": "MORphological PArser MORPA", "start": 234, "end": 260}, {"text": "probabilistic context-free grammar -LRB- PCFG -RRB-", "start": 280, "end": 331}, {"text": "``conventional'' context-free morphological grammar", "start": 352, "end": 403}, {"text": "probability-based scoring function", "start": 453, "end": 487}, {"text": "PCFG", "start": 321, "end": 325}, {"text": "MORPA", "start": 255, "end": 260}, {"text": "parser", "start": 744, "end": 750}], "generic": [{"text": "it", "start": 40, "end": 42}]}, "relations": {"used_for": [{"head": {"text": "MORphological PArser MORPA", "start": 234, "end": 260}, "tail": {"text": "ambiguity", "start": 219, "end": 228}}, {"head": {"text": "probabilistic context-free grammar -LRB- PCFG -RRB-", "start": 280, "end": 331}, "tail": {"text": "MORphological PArser MORPA", "start": 234, "end": 260}}, {"head": {"text": "``conventional'' context-free morphological grammar", "start": 352, "end": 403}, "tail": {"text": "it", "start": 40, "end": 42}}, {"head": {"text": "``conventional'' context-free morphological grammar", "start": 352, "end": 403}, "tail": {"text": "ungrammatical segmentations", "start": 418, "end": 445}}, {"head": {"text": "probability-based scoring function", "start": 453, "end": 487}, "tail": {"text": "it", "start": 40, "end": 42}}, {"head": {"text": "probability-based scoring function", "start": 453, "end": 487}, "tail": {"text": "parse", "start": 539, "end": 544}}, {"head": {"text": "PCFG", "start": 321, "end": 325}, "tail": {"text": "morphological parsing", "start": 692, "end": 713}}, {"head": {"text": "MORPA", "start": 255, "end": 260}, "tail": {"text": "text-to-speech conversion system", "start": 774, "end": 806}}, {"head": {"text": "parser", "start": 744, "end": 750}, "tail": {"text": "text-to-speech conversion system", "start": 774, "end": 806}}], "conjunction": [{"head": {"text": "probability-based scoring function", "start": 453, "end": 487}, "tail": {"text": "``conventional'' context-free morphological grammar", "start": 352, "end": 403}}], "hyponym_of": [{"head": {"text": "MORPA", "start": 255, "end": 260}, "tail": {"text": "parser", "start": 744, "end": 750}}]}}, "schema": []}
{"input": "This paper describes the framework of a Korean phonological knowledge base system using the unification-based grammar formalism: Korean Phonology Structure Grammar -LRB- KPSG -RRB-. The approach of KPSG provides an explicit development model for constructing a computational phonological system: speech recognition and synthesis system. We show that the proposed approach is more describable than other approaches such as those employing a traditional generative phonological approach.", "output": {"entities": {"task": [{"text": "Korean phonological knowledge base system", "start": 40, "end": 81}, {"text": "phonological system", "start": 275, "end": 294}, {"text": "speech recognition and synthesis system", "start": 296, "end": 335}], "method": [{"text": "unification-based grammar formalism", "start": 92, "end": 127}, {"text": "Korean Phonology Structure Grammar -LRB- KPSG -RRB-", "start": 129, "end": 180}, {"text": "KPSG", "start": 170, "end": 174}, {"text": "generative phonological approach", "start": 452, "end": 484}], "generic": [{"text": "approach", "start": 186, "end": 194}, {"text": "approach", "start": 363, "end": 371}, {"text": "approaches", "start": 403, "end": 413}, {"text": "those", "start": 422, "end": 427}]}, "relations": {"used_for": [{"head": {"text": "unification-based grammar formalism", "start": 92, "end": 127}, "tail": {"text": "Korean phonological knowledge base system", "start": 40, "end": 81}}, {"head": {"text": "approach", "start": 186, "end": 194}, "tail": {"text": "KPSG", "start": 170, "end": 174}}, {"head": {"text": "KPSG", "start": 170, "end": 174}, "tail": {"text": "phonological system", "start": 275, "end": 294}}, {"head": {"text": "generative phonological approach", "start": 452, "end": 484}, "tail": {"text": "those", "start": 422, "end": 427}}], "hyponym_of": [{"head": {"text": "Korean Phonology Structure Grammar -LRB- KPSG -RRB-", "start": 129, "end": 180}, "tail": {"text": "unification-based grammar formalism", "start": 92, "end": 127}}], "compare": [{"head": {"text": "approach", "start": 363, "end": 371}, "tail": {"text": "approaches", "start": 403, "end": 413}}]}}, "schema": []}
{"input": "In some auction domains, there is uncertainty regarding the final availability of the goods being auctioned off. For example, a government may auction off spectrum from its public safety network, but it may need this spectrum back in times of emergency. In such a domain, standard combinatorial auctions perform poorly because they lead to violations of individual rationality -LRB- IR -RRB-, even in expectation, and to very low efficiency. In this paper, we study the design of core-selecting payment rules for such domains. Surprisingly, we show that in this new domain, there does not exist a payment rule with is guaranteed to be ex-post core-selecting. However, we show that by designing rules that are'' execution-contingent,'' i.e., by charging payments that are conditioned on the realization of the availability of the goods, we can reduce IR violations. We design two core-selecting rules that always satisfy IR in expectation. To study the performance of our rules we perform a computational Bayes-Nash equilibrium analysis. We show that, in equilibrium, our new rules have better incentives, higher efficiency, and a lower rate of ex-post IR violations than standard core-selecting rules.", "output": {"entities": {"task": [{"text": "auction domains", "start": 8, "end": 23}, {"text": "design of core-selecting payment rules", "start": 470, "end": 508}], "generic": [{"text": "domain", "start": 16, "end": 22}, {"text": "they", "start": 327, "end": 331}, {"text": "domains", "start": 16, "end": 23}, {"text": "domain", "start": 264, "end": 270}, {"text": "rules", "start": 694, "end": 699}, {"text": "rules", "start": 894, "end": 899}], "method": [{"text": "combinatorial auctions", "start": 281, "end": 303}, {"text": "computational Bayes-Nash equilibrium analysis", "start": 990, "end": 1035}], "other_scientific_term": [{"text": "violations of individual rationality -LRB- IR -RRB-", "start": 340, "end": 391}, {"text": "individual rationality -LRB- IR -RRB-", "start": 354, "end": 391}, {"text": "payment rule", "start": 495, "end": 507}, {"text": "rules", "start": 503, "end": 508}, {"text": "IR violations", "start": 850, "end": 863}, {"text": "core-selecting rules", "start": 879, "end": 899}, {"text": "IR", "start": 383, "end": 385}, {"text": "core-selecting rules", "start": 1180, "end": 1200}], "metric": [{"text": "rate of ex-post IR violations", "start": 1136, "end": 1165}]}, "relations": {"used_for": [{"head": {"text": "design of core-selecting payment rules", "start": 470, "end": 508}, "tail": {"text": "domains", "start": 16, "end": 23}}, {"head": {"text": "core-selecting rules", "start": 879, "end": 899}, "tail": {"text": "IR", "start": 383, "end": 385}}, {"head": {"text": "computational Bayes-Nash equilibrium analysis", "start": 990, "end": 1035}, "tail": {"text": "rules", "start": 694, "end": 699}}], "compare": [{"head": {"text": "rules", "start": 894, "end": 899}, "tail": {"text": "core-selecting rules", "start": 1180, "end": 1200}}], "evaluate_for": [{"head": {"text": "rate of ex-post IR violations", "start": 1136, "end": 1165}, "tail": {"text": "rules", "start": 894, "end": 899}}, {"head": {"text": "rate of ex-post IR violations", "start": 1136, "end": 1165}, "tail": {"text": "core-selecting rules", "start": 1180, "end": 1200}}]}}, "schema": []}
{"input": "In this paper, we will describe a search tool for a huge set of ngrams. The tool supports queries with an arbitrary number of wildcards. It takes a fraction of a second for a search, and can provide the fillers of the wildcards. The system runs on a single Linux PC with reasonable size memory -LRB- less than 4GB -RRB- and disk space -LRB- less than 400GB -RRB-. This system can be a very useful tool for linguistic knowledge discovery and other NLP tasks.", "output": {"entities": {"method": [{"text": "search tool", "start": 34, "end": 45}], "other_scientific_term": [{"text": "ngrams", "start": 64, "end": 70}, {"text": "memory", "start": 287, "end": 293}, {"text": "disk space", "start": 324, "end": 334}], "generic": [{"text": "tool", "start": 41, "end": 45}, {"text": "It", "start": 137, "end": 139}, {"text": "system", "start": 233, "end": 239}, {"text": "system", "start": 369, "end": 375}, {"text": "tool", "start": 76, "end": 80}], "task": [{"text": "linguistic knowledge discovery", "start": 406, "end": 436}, {"text": "NLP tasks", "start": 447, "end": 456}]}, "relations": {"used_for": [{"head": {"text": "search tool", "start": 34, "end": 45}, "tail": {"text": "ngrams", "start": 64, "end": 70}}, {"head": {"text": "tool", "start": 76, "end": 80}, "tail": {"text": "linguistic knowledge discovery", "start": 406, "end": 436}}, {"head": {"text": "tool", "start": 76, "end": 80}, "tail": {"text": "NLP tasks", "start": 447, "end": 456}}], "conjunction": [{"head": {"text": "linguistic knowledge discovery", "start": 406, "end": 436}, "tail": {"text": "NLP tasks", "start": 447, "end": 456}}]}}, "schema": []}
{"input": "For intelligent interactive systems to communicate with humans in a natural manner, they must have knowledge about the system users. This paper explores the role of user modeling in such systems. It begins with a characterization of what a user model is and how it can be used. The types of information that a user model may be required to keep about a user are then identified and discussed. User models themselves can vary greatly depending on the requirements of the situation and the implementation, so several dimensions along which they can be classified are presented. Since acquiring the knowledge for a user model is a fundamental problem in user modeling, a section is devoted to this topic. Next, the benefits and costs of implementing a user modeling component for a system are weighed in light of several aspects of the interaction requirements that may be imposed by the system. Finally, the current state of research in user modeling is summarized, and future research topics that must be addressed in order to achieve powerful, general user modeling systems are assessed.", "output": {"entities": {"method": [{"text": "intelligent interactive systems", "start": 4, "end": 35}, {"text": "systems", "start": 28, "end": 35}, {"text": "user model", "start": 165, "end": 175}, {"text": "user model", "start": 240, "end": 250}, {"text": "User models", "start": 393, "end": 404}, {"text": "user model", "start": 310, "end": 320}, {"text": "user modeling component", "start": 749, "end": 772}, {"text": "user modeling systems", "start": 1052, "end": 1073}], "generic": [{"text": "they", "start": 84, "end": 88}, {"text": "it", "start": 52, "end": 54}, {"text": "they", "start": 538, "end": 542}, {"text": "system", "start": 28, "end": 34}, {"text": "system", "start": 119, "end": 125}], "task": [{"text": "user modeling", "start": 165, "end": 178}, {"text": "user modeling", "start": 651, "end": 664}, {"text": "user modeling", "start": 749, "end": 762}]}, "relations": {"part_of": [{"head": {"text": "user modeling", "start": 165, "end": 178}, "tail": {"text": "systems", "start": 28, "end": 35}}, {"head": {"text": "user modeling component", "start": 749, "end": 772}, "tail": {"text": "system", "start": 28, "end": 34}}], "used_for": [{"head": {"text": "user model", "start": 310, "end": 320}, "tail": {"text": "user modeling", "start": 651, "end": 664}}]}}, "schema": []}
{"input": "Information extraction techniques automatically create structured databases from unstructured data sources, such as the Web or newswire documents. Despite the successes of these systems, accuracy will always be imperfect. For many reasons, it is highly desirable to accurately estimate the confidence the system has in the correctness of each extracted field. The information extraction system we evaluate is based on a linear-chain conditional random field -LRB- CRF -RRB-, a probabilistic model which has performed well on information extraction tasks because of its ability to capture arbitrary, overlapping features of the input in a Markov model. We implement several techniques to estimate the confidence of both extracted fields and entire multi-field records, obtaining an average precision of 98% for retrieving correct fields and 87% for multi-field records.", "output": {"entities": {"method": [{"text": "Information extraction techniques", "start": 0, "end": 33}, {"text": "information extraction system", "start": 364, "end": 393}, {"text": "linear-chain conditional random field -LRB- CRF -RRB-", "start": 420, "end": 473}, {"text": "probabilistic model", "start": 477, "end": 496}, {"text": "Markov model", "start": 638, "end": 650}], "material": [{"text": "structured databases", "start": 55, "end": 75}, {"text": "unstructured data sources", "start": 81, "end": 106}, {"text": "Web", "start": 120, "end": 123}, {"text": "newswire documents", "start": 127, "end": 145}, {"text": "multi-field records", "start": 747, "end": 766}, {"text": "multi-field records", "start": 848, "end": 867}], "generic": [{"text": "systems", "start": 178, "end": 185}, {"text": "system", "start": 178, "end": 184}, {"text": "input", "start": 627, "end": 632}, {"text": "techniques", "start": 23, "end": 33}, {"text": "extracted fields", "start": 719, "end": 735}], "metric": [{"text": "accuracy", "start": 187, "end": 195}, {"text": "average precision", "start": 781, "end": 798}], "task": [{"text": "information extraction tasks", "start": 525, "end": 553}], "other_scientific_term": [{"text": "arbitrary, overlapping features", "start": 588, "end": 619}]}, "relations": {"used_for": [{"head": {"text": "Information extraction techniques", "start": 0, "end": 33}, "tail": {"text": "structured databases", "start": 55, "end": 75}}, {"head": {"text": "unstructured data sources", "start": 81, "end": 106}, "tail": {"text": "Information extraction techniques", "start": 0, "end": 33}}, {"head": {"text": "linear-chain conditional random field -LRB- CRF -RRB-", "start": 420, "end": 473}, "tail": {"text": "information extraction system", "start": 364, "end": 393}}, {"head": {"text": "probabilistic model", "start": 477, "end": 496}, "tail": {"text": "information extraction tasks", "start": 525, "end": 553}}, {"head": {"text": "probabilistic model", "start": 477, "end": 496}, "tail": {"text": "arbitrary, overlapping features", "start": 588, "end": 619}}], "hyponym_of": [{"head": {"text": "Web", "start": 120, "end": 123}, "tail": {"text": "unstructured data sources", "start": 81, "end": 106}}, {"head": {"text": "newswire documents", "start": 127, "end": 145}, "tail": {"text": "unstructured data sources", "start": 81, "end": 106}}, {"head": {"text": "linear-chain conditional random field -LRB- CRF -RRB-", "start": 420, "end": 473}, "tail": {"text": "probabilistic model", "start": 477, "end": 496}}], "conjunction": [{"head": {"text": "Web", "start": 120, "end": 123}, "tail": {"text": "newswire documents", "start": 127, "end": 145}}, {"head": {"text": "extracted fields", "start": 719, "end": 735}, "tail": {"text": "multi-field records", "start": 747, "end": 766}}], "evaluate_for": [{"head": {"text": "accuracy", "start": 187, "end": 195}, "tail": {"text": "systems", "start": 178, "end": 185}}, {"head": {"text": "average precision", "start": 781, "end": 798}, "tail": {"text": "techniques", "start": 23, "end": 33}}], "feature_of": [{"head": {"text": "arbitrary, overlapping features", "start": 588, "end": 619}, "tail": {"text": "input", "start": 627, "end": 632}}], "part_of": [{"head": {"text": "arbitrary, overlapping features", "start": 588, "end": 619}, "tail": {"text": "Markov model", "start": 638, "end": 650}}]}}, "schema": []}
{"input": "In this paper, we use the information redundancy in multilingual input to correct errors in machine translation and thus improve the quality of multilingual summaries. We consider the case of multi-document summarization, where the input documents are in Arabic, and the output summary is in English. Typically, information that makes it to a summary appears in many different lexical-syntactic forms in the input documents. Further, the use of multiple machine translation systems provides yet more redundancy, yielding different ways to realize that information in English. We demonstrate how errors in the machine translations of the input Arabic documents can be corrected by identifying and generating from such redundancy, focusing on noun phrases.", "output": {"entities": {"other_scientific_term": [{"text": "information redundancy in multilingual input", "start": 26, "end": 70}, {"text": "lexical-syntactic forms", "start": 377, "end": 400}], "task": [{"text": "machine translation", "start": 92, "end": 111}, {"text": "multilingual summaries", "start": 144, "end": 166}, {"text": "multi-document summarization", "start": 192, "end": 220}, {"text": "machine translations", "start": 609, "end": 629}], "material": [{"text": "Arabic", "start": 255, "end": 261}, {"text": "English", "start": 292, "end": 299}, {"text": "English", "start": 567, "end": 574}, {"text": "Arabic documents", "start": 643, "end": 659}], "method": [{"text": "machine translation systems", "start": 454, "end": 481}]}, "relations": {"used_for": [{"head": {"text": "information redundancy in multilingual input", "start": 26, "end": 70}, "tail": {"text": "machine translation", "start": 92, "end": 111}}, {"head": {"text": "information redundancy in multilingual input", "start": 26, "end": 70}, "tail": {"text": "multilingual summaries", "start": 144, "end": 166}}, {"head": {"text": "Arabic documents", "start": 643, "end": 659}, "tail": {"text": "machine translations", "start": 609, "end": 629}}]}}, "schema": []}
{"input": "In this paper, we propose a new approach to generate oriented object proposals -LRB- OOPs -RRB- to reduce the detection error caused by various orientations of the object. To this end, we propose to efficiently locate object regions according to pixelwise object probability, rather than measuring the objectness from a set of sampled windows. We formulate the proposal generation problem as a generative proba-bilistic model such that object proposals of different shapes -LRB- i.e., sizes and orientations -RRB- can be produced by locating the local maximum likelihoods. The new approach has three main advantages. First, it helps the object detector handle objects of different orientations. Second, as the shapes of the proposals may vary to fit the objects, the resulting proposals are tighter than the sampling windows with fixed sizes. Third, it avoids massive window sampling, and thereby reducing the number of proposals while maintaining a high recall. Experiments on the PASCAL VOC 2007 dataset show that the proposed OOP outperforms the state-of-the-art fast methods. Further experiments show that the rotation invariant property helps a class-specific object detector achieve better performance than the state-of-the-art proposal generation methods in either object rotation scenarios or general scenarios. Generating OOPs is very fast and takes only 0.5 s per image.", "output": {"entities": {"generic": [{"text": "approach", "start": 32, "end": 40}, {"text": "approach", "start": 581, "end": 589}, {"text": "it", "start": 271, "end": 273}, {"text": "it", "start": 624, "end": 626}, {"text": "state-of-the-art fast methods", "start": 1049, "end": 1078}], "task": [{"text": "oriented object proposals -LRB- OOPs -RRB-", "start": 53, "end": 95}, {"text": "proposal generation problem", "start": 361, "end": 388}, {"text": "OOPs", "start": 85, "end": 89}], "metric": [{"text": "detection error", "start": 110, "end": 125}, {"text": "recall", "start": 955, "end": 961}], "other_scientific_term": [{"text": "orientations of the object", "start": 144, "end": 170}, {"text": "object regions", "start": 218, "end": 232}, {"text": "pixelwise object probability", "start": 246, "end": 274}, {"text": "objectness", "start": 302, "end": 312}, {"text": "object proposals", "start": 62, "end": 78}, {"text": "shapes", "start": 466, "end": 472}, {"text": "sizes", "start": 485, "end": 490}, {"text": "orientations", "start": 144, "end": 156}, {"text": "local maximum likelihoods", "start": 546, "end": 571}, {"text": "orientations", "start": 495, "end": 507}, {"text": "shapes of the proposals", "start": 710, "end": 733}, {"text": "sampling windows", "start": 808, "end": 824}, {"text": "number of proposals", "start": 910, "end": 929}, {"text": "rotation invariant property", "start": 1114, "end": 1141}], "method": [{"text": "generative proba-bilistic model", "start": 394, "end": 425}, {"text": "object detector", "start": 637, "end": 652}, {"text": "massive window sampling", "start": 860, "end": 883}, {"text": "OOP", "start": 85, "end": 88}, {"text": "class-specific object detector", "start": 1150, "end": 1180}, {"text": "proposal generation methods", "start": 1234, "end": 1261}], "material": [{"text": "PASCAL VOC 2007 dataset", "start": 982, "end": 1005}, {"text": "object rotation scenarios", "start": 1272, "end": 1297}, {"text": "general scenarios", "start": 1301, "end": 1318}]}, "relations": {"used_for": [{"head": {"text": "approach", "start": 32, "end": 40}, "tail": {"text": "oriented object proposals -LRB- OOPs -RRB-", "start": 53, "end": 95}}, {"head": {"text": "pixelwise object probability", "start": 246, "end": 274}, "tail": {"text": "object regions", "start": 218, "end": 232}}, {"head": {"text": "generative proba-bilistic model", "start": 394, "end": 425}, "tail": {"text": "proposal generation problem", "start": 361, "end": 388}}, {"head": {"text": "local maximum likelihoods", "start": 546, "end": 571}, "tail": {"text": "object proposals", "start": 62, "end": 78}}, {"head": {"text": "object detector", "start": 637, "end": 652}, "tail": {"text": "orientations", "start": 495, "end": 507}}, {"head": {"text": "it", "start": 624, "end": 626}, "tail": {"text": "number of proposals", "start": 910, "end": 929}}, {"head": {"text": "rotation invariant property", "start": 1114, "end": 1141}, "tail": {"text": "class-specific object detector", "start": 1150, "end": 1180}}], "evaluate_for": [{"head": {"text": "detection error", "start": 110, "end": 125}, "tail": {"text": "oriented object proposals -LRB- OOPs -RRB-", "start": 53, "end": 95}}, {"head": {"text": "recall", "start": 955, "end": 961}, "tail": {"text": "it", "start": 624, "end": 626}}, {"head": {"text": "PASCAL VOC 2007 dataset", "start": 982, "end": 1005}, "tail": {"text": "OOP", "start": 85, "end": 88}}, {"head": {"text": "object rotation scenarios", "start": 1272, "end": 1297}, "tail": {"text": "class-specific object detector", "start": 1150, "end": 1180}}, {"head": {"text": "object rotation scenarios", "start": 1272, "end": 1297}, "tail": {"text": "proposal generation methods", "start": 1234, "end": 1261}}, {"head": {"text": "general scenarios", "start": 1301, "end": 1318}, "tail": {"text": "class-specific object detector", "start": 1150, "end": 1180}}, {"head": {"text": "general scenarios", "start": 1301, "end": 1318}, "tail": {"text": "proposal generation methods", "start": 1234, "end": 1261}}], "compare": [{"head": {"text": "pixelwise object probability", "start": 246, "end": 274}, "tail": {"text": "objectness", "start": 302, "end": 312}}, {"head": {"text": "OOP", "start": 85, "end": 88}, "tail": {"text": "state-of-the-art fast methods", "start": 1049, "end": 1078}}, {"head": {"text": "class-specific object detector", "start": 1150, "end": 1180}, "tail": {"text": "proposal generation methods", "start": 1234, "end": 1261}}], "feature_of": [{"head": {"text": "shapes", "start": 466, "end": 472}, "tail": {"text": "object proposals", "start": 62, "end": 78}}], "hyponym_of": [{"head": {"text": "sizes", "start": 485, "end": 490}, "tail": {"text": "shapes", "start": 466, "end": 472}}, {"head": {"text": "orientations", "start": 144, "end": 156}, "tail": {"text": "shapes", "start": 466, "end": 472}}], "conjunction": [{"head": {"text": "sizes", "start": 485, "end": 490}, "tail": {"text": "orientations", "start": 144, "end": 156}}, {"head": {"text": "object rotation scenarios", "start": 1272, "end": 1297}, "tail": {"text": "general scenarios", "start": 1301, "end": 1318}}]}}, "schema": []}
{"input": "This paper describes three relatively domain-independent capabilities recently added to the Paramax spoken language understanding system: non-monotonic reasoning, implicit reference resolution, and database query paraphrase. In addition, we discuss the results of the February 1992 ATIS benchmark tests. We describe a variation on the standard evaluation metric which provides a more tightly controlled measure of progress. Finally, we briefly describe an experiment which we have done in extending the n-best speech/language integration architecture to improving OCR accuracy.", "output": {"entities": {"generic": [{"text": "domain-independent capabilities", "start": 38, "end": 69}], "method": [{"text": "Paramax spoken language understanding system", "start": 92, "end": 136}, {"text": "n-best speech/language integration architecture", "start": 503, "end": 550}], "task": [{"text": "non-monotonic reasoning", "start": 138, "end": 161}, {"text": "implicit reference resolution", "start": 163, "end": 192}, {"text": "database query paraphrase", "start": 198, "end": 223}], "material": [{"text": "February 1992 ATIS benchmark tests", "start": 268, "end": 302}], "metric": [{"text": "OCR accuracy", "start": 564, "end": 576}]}, "relations": {"part_of": [{"head": {"text": "domain-independent capabilities", "start": 38, "end": 69}, "tail": {"text": "Paramax spoken language understanding system", "start": 92, "end": 136}}], "hyponym_of": [{"head": {"text": "non-monotonic reasoning", "start": 138, "end": 161}, "tail": {"text": "domain-independent capabilities", "start": 38, "end": 69}}, {"head": {"text": "implicit reference resolution", "start": 163, "end": 192}, "tail": {"text": "domain-independent capabilities", "start": 38, "end": 69}}, {"head": {"text": "database query paraphrase", "start": 198, "end": 223}, "tail": {"text": "domain-independent capabilities", "start": 38, "end": 69}}], "evaluate_for": [{"head": {"text": "OCR accuracy", "start": 564, "end": 576}, "tail": {"text": "n-best speech/language integration architecture", "start": 503, "end": 550}}]}}, "schema": []}
{"input": "We investigate the problem of fine-grained sketch-based image retrieval -LRB- SBIR -RRB-, where free-hand human sketches are used as queries to perform instance-level retrieval of images. This is an extremely challenging task because -LRB- i -RRB- visual comparisons not only need to be fine-grained but also executed cross-domain, -LRB- ii -RRB- free-hand -LRB- finger -RRB- sketches are highly abstract, making fine-grained matching harder, and most importantly -LRB- iii -RRB- annotated cross-domain sketch-photo datasets required for training are scarce, challenging many state-of-the-art machine learning techniques. In this paper, for the first time, we address all these challenges, providing a step towards the capabilities that would underpin a commercial sketch-based image retrieval application. We introduce a new database of 1,432 sketch-photo pairs from two categories with 32,000 fine-grained triplet ranking annotations. We then develop a deep triplet-ranking model for instance-level SBIR with a novel data augmentation and staged pre-training strategy to alleviate the issue of insufficient fine-grained training data. Extensive experiments are carried out to contribute a variety of insights into the challenges of data sufficiency and over-fitting avoidance when training deep networks for fine-grained cross-domain ranking tasks.", "output": {"entities": {"task": [{"text": "fine-grained sketch-based image retrieval -LRB- SBIR -RRB-", "start": 30, "end": 88}, {"text": "instance-level retrieval of images", "start": 152, "end": 186}, {"text": "fine-grained matching", "start": 413, "end": 434}, {"text": "sketch-based image retrieval application", "start": 765, "end": 805}, {"text": "instance-level SBIR", "start": 986, "end": 1005}, {"text": "fine-grained cross-domain ranking tasks", "start": 1310, "end": 1349}], "other_scientific_term": [{"text": "free-hand human sketches", "start": 96, "end": 120}, {"text": "visual comparisons", "start": 248, "end": 266}, {"text": "free-hand -LRB- finger -RRB- sketches", "start": 347, "end": 384}, {"text": "fine-grained triplet ranking annotations", "start": 895, "end": 935}, {"text": "data sufficiency", "start": 1234, "end": 1250}, {"text": "over-fitting avoidance", "start": 1255, "end": 1277}], "material": [{"text": "annotated cross-domain sketch-photo datasets", "start": 480, "end": 524}, {"text": "insufficient fine-grained training data", "start": 1096, "end": 1135}], "method": [{"text": "machine learning techniques", "start": 593, "end": 620}, {"text": "deep triplet-ranking model", "start": 955, "end": 981}, {"text": "data augmentation", "start": 1019, "end": 1036}, {"text": "staged pre-training strategy", "start": 1041, "end": 1069}, {"text": "deep networks", "start": 1292, "end": 1305}]}, "relations": {"used_for": [{"head": {"text": "free-hand human sketches", "start": 96, "end": 120}, "tail": {"text": "instance-level retrieval of images", "start": 152, "end": 186}}, {"head": {"text": "annotated cross-domain sketch-photo datasets", "start": 480, "end": 524}, "tail": {"text": "machine learning techniques", "start": 593, "end": 620}}, {"head": {"text": "deep triplet-ranking model", "start": 955, "end": 981}, "tail": {"text": "instance-level SBIR", "start": 986, "end": 1005}}, {"head": {"text": "deep triplet-ranking model", "start": 955, "end": 981}, "tail": {"text": "insufficient fine-grained training data", "start": 1096, "end": 1135}}, {"head": {"text": "data augmentation", "start": 1019, "end": 1036}, "tail": {"text": "deep triplet-ranking model", "start": 955, "end": 981}}, {"head": {"text": "staged pre-training strategy", "start": 1041, "end": 1069}, "tail": {"text": "deep triplet-ranking model", "start": 955, "end": 981}}, {"head": {"text": "deep networks", "start": 1292, "end": 1305}, "tail": {"text": "fine-grained cross-domain ranking tasks", "start": 1310, "end": 1349}}], "conjunction": [{"head": {"text": "data augmentation", "start": 1019, "end": 1036}, "tail": {"text": "staged pre-training strategy", "start": 1041, "end": 1069}}, {"head": {"text": "data sufficiency", "start": 1234, "end": 1250}, "tail": {"text": "over-fitting avoidance", "start": 1255, "end": 1277}}]}}, "schema": []}
{"input": "In this paper we target at generating generic action proposals in unconstrained videos. Each action proposal corresponds to a temporal series of spatial bounding boxes, i.e., a spatio-temporal video tube, which has a good potential to locate one human action. Assuming each action is performed by a human with meaningful motion, both appearance and motion cues are utilized to measure the ac-tionness of the video tubes. After picking those spatiotem-poral paths of high actionness scores, our action proposal generation is formulated as a maximum set coverage problem, where greedy search is performed to select a set of action proposals that can maximize the overall actionness score. Compared with existing action proposal approaches, our action proposals do not rely on video segmentation and can be generated in nearly real-time. Experimental results on two challenging datasets, MSRII and UCF 101, validate the superior performance of our action proposals as well as competitive results on action detection and search.", "output": {"entities": {"other_scientific_term": [{"text": "generic action proposals", "start": 38, "end": 62}, {"text": "action proposal", "start": 46, "end": 61}, {"text": "temporal series of spatial bounding boxes", "start": 126, "end": 167}, {"text": "spatio-temporal video tube", "start": 177, "end": 203}, {"text": "human action", "start": 246, "end": 258}, {"text": "appearance and motion cues", "start": 334, "end": 360}, {"text": "action proposals", "start": 46, "end": 62}, {"text": "action proposals", "start": 622, "end": 638}, {"text": "action proposals", "start": 742, "end": 758}], "material": [{"text": "unconstrained videos", "start": 66, "end": 86}, {"text": "video tubes", "start": 408, "end": 419}, {"text": "MSRII", "start": 885, "end": 890}, {"text": "UCF 101", "start": 895, "end": 902}], "metric": [{"text": "ac-tionness", "start": 389, "end": 400}, {"text": "actionness scores", "start": 471, "end": 488}, {"text": "actionness score", "start": 471, "end": 487}], "task": [{"text": "action proposal generation", "start": 494, "end": 520}, {"text": "maximum set coverage problem", "start": 540, "end": 568}, {"text": "action detection and search", "start": 996, "end": 1023}], "method": [{"text": "greedy search", "start": 576, "end": 589}, {"text": "action proposal approaches", "start": 710, "end": 736}, {"text": "video segmentation", "start": 774, "end": 792}], "generic": [{"text": "datasets", "start": 875, "end": 883}]}, "relations": {"used_for": [{"head": {"text": "unconstrained videos", "start": 66, "end": 86}, "tail": {"text": "generic action proposals", "start": 38, "end": 62}}, {"head": {"text": "spatio-temporal video tube", "start": 177, "end": 203}, "tail": {"text": "human action", "start": 246, "end": 258}}, {"head": {"text": "appearance and motion cues", "start": 334, "end": 360}, "tail": {"text": "ac-tionness", "start": 389, "end": 400}}, {"head": {"text": "maximum set coverage problem", "start": 540, "end": 568}, "tail": {"text": "action proposal generation", "start": 494, "end": 520}}, {"head": {"text": "greedy search", "start": 576, "end": 589}, "tail": {"text": "action proposals", "start": 46, "end": 62}}], "hyponym_of": [{"head": {"text": "spatio-temporal video tube", "start": 177, "end": 203}, "tail": {"text": "temporal series of spatial bounding boxes", "start": 126, "end": 167}}, {"head": {"text": "MSRII", "start": 885, "end": 890}, "tail": {"text": "datasets", "start": 875, "end": 883}}, {"head": {"text": "UCF 101", "start": 895, "end": 902}, "tail": {"text": "datasets", "start": 875, "end": 883}}], "evaluate_for": [{"head": {"text": "ac-tionness", "start": 389, "end": 400}, "tail": {"text": "video tubes", "start": 408, "end": 419}}, {"head": {"text": "actionness score", "start": 471, "end": 487}, "tail": {"text": "action proposals", "start": 46, "end": 62}}, {"head": {"text": "datasets", "start": 875, "end": 883}, "tail": {"text": "action proposals", "start": 742, "end": 758}}, {"head": {"text": "action detection and search", "start": 996, "end": 1023}, "tail": {"text": "action proposals", "start": 742, "end": 758}}], "compare": [{"head": {"text": "action proposal approaches", "start": 710, "end": 736}, "tail": {"text": "action proposals", "start": 622, "end": 638}}], "conjunction": [{"head": {"text": "MSRII", "start": 885, "end": 890}, "tail": {"text": "UCF 101", "start": 895, "end": 902}}]}}, "schema": []}
{"input": "This paper reports recent research into methods for creating natural language text. A new processing paradigm called Fragment-and-Compose has been created and an experimental system implemented in it. The knowledge to be expressed in text is first divided into small propositional units, which are then composed into appropriate combinations and converted into text. KDS -LRB- Knowledge Delivery System -RRB-, which embodies this paradigm, has distinct parts devoted to creation of the propositional units, to organization of the text, to prevention of excess redundancy, to creation of combinations of units, to evaluation of these combinations as potential sentences, to selection of the best among competing combinations, and to creation of the final text. The Fragment-and-Compose paradigm and the computational methods of KDS are described.", "output": {"entities": {"generic": [{"text": "methods", "start": 40, "end": 47}, {"text": "processing paradigm", "start": 90, "end": 109}, {"text": "paradigm", "start": 101, "end": 109}, {"text": "computational methods", "start": 802, "end": 823}], "task": [{"text": "creating natural language text", "start": 52, "end": 82}], "method": [{"text": "Fragment-and-Compose", "start": 117, "end": 137}, {"text": "KDS -LRB- Knowledge Delivery System -RRB-", "start": 367, "end": 408}, {"text": "Fragment-and-Compose paradigm", "start": 764, "end": 793}, {"text": "KDS", "start": 367, "end": 370}]}, "relations": {"used_for": [{"head": {"text": "methods", "start": 40, "end": 47}, "tail": {"text": "creating natural language text", "start": 52, "end": 82}}, {"head": {"text": "computational methods", "start": 802, "end": 823}, "tail": {"text": "KDS", "start": 367, "end": 370}}], "part_of": [{"head": {"text": "paradigm", "start": 101, "end": 109}, "tail": {"text": "KDS -LRB- Knowledge Delivery System -RRB-", "start": 367, "end": 408}}]}}, "schema": []}
{"input": "This paper explores the issue of using different co-occurrence similarities between terms for separating query terms that are useful for retrieval from those that are harmful. The hypothesis under examination is that useful terms tend to be more similar to each other than to other query terms. Preliminary experiments with similarities computed using first-order and second-order co-occurrence seem to confirm the hypothesis. Term similarities could then be used for determining which query terms are useful and best reflect the user's information need. A possible application would be to use this source of evidence for tuning the weights of the query terms.", "output": {"entities": {"other_scientific_term": [{"text": "co-occurrence similarities", "start": 49, "end": 75}, {"text": "first-order and second-order co-occurrence", "start": 352, "end": 394}, {"text": "Term similarities", "start": 427, "end": 444}], "generic": [{"text": "query terms", "start": 105, "end": 116}, {"text": "those", "start": 152, "end": 157}, {"text": "useful terms", "start": 217, "end": 229}, {"text": "query terms", "start": 282, "end": 293}, {"text": "similarities", "start": 63, "end": 75}], "task": [{"text": "retrieval", "start": 137, "end": 146}]}, "relations": {"used_for": [{"head": {"text": "co-occurrence similarities", "start": 49, "end": 75}, "tail": {"text": "query terms", "start": 105, "end": 116}}, {"head": {"text": "query terms", "start": 105, "end": 116}, "tail": {"text": "retrieval", "start": 137, "end": 146}}, {"head": {"text": "first-order and second-order co-occurrence", "start": 352, "end": 394}, "tail": {"text": "similarities", "start": 63, "end": 75}}], "compare": [{"head": {"text": "those", "start": 152, "end": 157}, "tail": {"text": "query terms", "start": 105, "end": 116}}, {"head": {"text": "useful terms", "start": 217, "end": 229}, "tail": {"text": "query terms", "start": 282, "end": 293}}]}}, "schema": []}
{"input": "We propose a new phrase-based translation model and decoding algorithm that enables us to evaluate and compare several, previously proposed phrase-based translation models. Within our framework, we carry out a large number of experiments to understand better and explain why phrase-based models outperform word-based models. Our empirical results, which hold for all examined language pairs, suggest that the highest levels of performance can be obtained through relatively simple means: heuristic learning of phrase translations from word-based alignments and lexical weighting of phrase translations. Surprisingly, learning phrases longer than three words and learning phrases from high-accuracy word-level alignment models does not have a strong impact on performance. Learning only syntactically motivated phrases degrades the performance of our systems.", "output": {"entities": {"method": [{"text": "phrase-based translation model", "start": 17, "end": 47}, {"text": "decoding algorithm", "start": 52, "end": 70}, {"text": "phrase-based translation models", "start": 140, "end": 171}, {"text": "phrase-based models", "start": 275, "end": 294}, {"text": "word-based models", "start": 306, "end": 323}, {"text": "heuristic learning of phrase translations", "start": 488, "end": 529}, {"text": "word-based alignments", "start": 535, "end": 556}, {"text": "lexical weighting of phrase translations", "start": 561, "end": 601}, {"text": "high-accuracy word-level alignment models", "start": 684, "end": 725}], "generic": [{"text": "framework", "start": 184, "end": 193}, {"text": "means", "start": 481, "end": 486}, {"text": "systems", "start": 850, "end": 857}]}, "relations": {"conjunction": [{"head": {"text": "phrase-based translation model", "start": 17, "end": 47}, "tail": {"text": "decoding algorithm", "start": 52, "end": 70}}], "compare": [{"head": {"text": "phrase-based models", "start": 275, "end": 294}, "tail": {"text": "word-based models", "start": 306, "end": 323}}], "hyponym_of": [{"head": {"text": "heuristic learning of phrase translations", "start": 488, "end": 529}, "tail": {"text": "means", "start": 481, "end": 486}}, {"head": {"text": "lexical weighting of phrase translations", "start": 561, "end": 601}, "tail": {"text": "means", "start": 481, "end": 486}}], "used_for": [{"head": {"text": "word-based alignments", "start": 535, "end": 556}, "tail": {"text": "heuristic learning of phrase translations", "start": 488, "end": 529}}]}}, "schema": []}
{"input": "Color is known to be highly discriminative for many object recognition tasks, but is difficult to infer from uncontrolled images in which the illuminant is not known. Traditional methods for color constancy can improve surface re-flectance estimates from such uncalibrated images, but their output depends significantly on the background scene. In many recognition and retrieval applications, we have access to image sets that contain multiple views of the same object in different environments; we show in this paper that correspondences between these images provide important constraints that can improve color constancy. We introduce the multi-view color constancy problem, and present a method to recover estimates of underlying surface re-flectance based on joint estimation of these surface properties and the illuminants present in multiple images. The method can exploit image correspondences obtained by various alignment techniques, and we show examples based on matching local region features. Our results show that multi-view constraints can significantly improve estimates of both scene illuminants and object color -LRB- surface reflectance -RRB- when compared to a baseline single-view method.", "output": {"entities": {"task": [{"text": "object recognition tasks", "start": 52, "end": 76}, {"text": "color constancy", "start": 191, "end": 206}, {"text": "recognition and retrieval applications", "start": 353, "end": 391}, {"text": "color constancy", "start": 607, "end": 622}, {"text": "multi-view color constancy problem", "start": 641, "end": 675}, {"text": "estimates of underlying surface re-flectance", "start": 709, "end": 753}, {"text": "estimates of both scene illuminants and object color -LRB- surface reflectance -RRB-", "start": 1076, "end": 1160}], "material": [{"text": "uncontrolled images", "start": 109, "end": 128}, {"text": "uncalibrated images", "start": 260, "end": 279}], "other_scientific_term": [{"text": "illuminant", "start": 142, "end": 152}, {"text": "background scene", "start": 327, "end": 343}, {"text": "surface properties", "start": 789, "end": 807}, {"text": "illuminants", "start": 816, "end": 827}, {"text": "image correspondences", "start": 879, "end": 900}, {"text": "matching local region features", "start": 973, "end": 1003}, {"text": "multi-view constraints", "start": 1027, "end": 1049}], "generic": [{"text": "methods", "start": 179, "end": 186}, {"text": "method", "start": 179, "end": 185}, {"text": "method", "start": 691, "end": 697}], "method": [{"text": "surface re-flectance estimates", "start": 219, "end": 249}, {"text": "alignment techniques", "start": 921, "end": 941}, {"text": "baseline single-view method", "start": 1180, "end": 1207}]}, "relations": {"used_for": [{"head": {"text": "methods", "start": 179, "end": 186}, "tail": {"text": "color constancy", "start": 191, "end": 206}}, {"head": {"text": "methods", "start": 179, "end": 186}, "tail": {"text": "surface re-flectance estimates", "start": 219, "end": 249}}, {"head": {"text": "uncalibrated images", "start": 260, "end": 279}, "tail": {"text": "surface re-flectance estimates", "start": 219, "end": 249}}, {"head": {"text": "method", "start": 179, "end": 185}, "tail": {"text": "estimates of underlying surface re-flectance", "start": 709, "end": 753}}, {"head": {"text": "method", "start": 691, "end": 697}, "tail": {"text": "image correspondences", "start": 879, "end": 900}}, {"head": {"text": "alignment techniques", "start": 921, "end": 941}, "tail": {"text": "image correspondences", "start": 879, "end": 900}}, {"head": {"text": "multi-view constraints", "start": 1027, "end": 1049}, "tail": {"text": "estimates of both scene illuminants and object color -LRB- surface reflectance -RRB-", "start": 1076, "end": 1160}}], "compare": [{"head": {"text": "baseline single-view method", "start": 1180, "end": 1207}, "tail": {"text": "multi-view constraints", "start": 1027, "end": 1049}}]}}, "schema": []}
{"input": "We describe a dialogue system that works with its interlocutor to identify objects. Our contributions include a concise, modular architecture with reversible processes of understanding and generation, an information-state model of reference, and flexible links between semantics and collaborative problem solving.", "output": {"entities": {"method": [{"text": "dialogue system", "start": 14, "end": 29}, {"text": "concise, modular architecture", "start": 112, "end": 141}, {"text": "information-state model of reference", "start": 204, "end": 240}], "task": [{"text": "understanding", "start": 171, "end": 184}, {"text": "generation", "start": 189, "end": 199}, {"text": "collaborative problem solving", "start": 283, "end": 312}], "other_scientific_term": [{"text": "semantics", "start": 269, "end": 278}]}, "relations": {"used_for": [{"head": {"text": "concise, modular architecture", "start": 112, "end": 141}, "tail": {"text": "understanding", "start": 171, "end": 184}}, {"head": {"text": "concise, modular architecture", "start": 112, "end": 141}, "tail": {"text": "generation", "start": 189, "end": 199}}], "conjunction": [{"head": {"text": "understanding", "start": 171, "end": 184}, "tail": {"text": "generation", "start": 189, "end": 199}}]}}, "schema": []}