_id stringlengths 4 10 | text stringlengths 0 18.4k | title stringlengths 0 8.56k |
|---|---|---|
d258486873 | The importance of multiword expressions (MWEs) for language learning is well established. While MWE research has been evaluated on various downstream tasks such as syntactic parsing and machine translation, its applications in computer-assisted language learning has been less explored. This paper investigates the selection of MWEs for graded vocabulary lists. Widely used by language teachers and students, these lists recommend a language acquisition sequence to optimize learning efficiency. We automatically generate these lists using difficulty-graded corpora and MWEs extracted based on semantic compositionality. We evaluate these lists on their ability to facilitate text comprehension for learners. Experimental results show that our proposed method generates higher-quality lists than baselines using collocation measures. | Automatic Generation of Vocabulary Lists with Multiword Expressions |
d49573952 | Shi, Huang, and Lee (2017a) obtained state-of-the-art results for English and Chinese dependency parsing by combining dynamic-programming implementations of transition-based dependency parsers with a minimal set of bidirectional LSTM features. However, their results were limited to projective parsing. In this paper, we extend their approach to support non-projectivity by providing the first practical implementation of the MH 4 algorithm, an Opn 4 q mildly nonprojective dynamic-programming parser with very high coverage on non-projective treebanks. To make MH 4 compatible with minimal transition-based feature sets, we introduce a transition-based interpretation of it in which parser items are mapped to sequences of transitions. We thus obtain the first implementation of global decoding for non-projective transition-based parsing, and demonstrate empirically that it is more effective than its projective counterpart in parsing a number of highly non-projective languages. | Global Transition-based Non-projective Dependency Parsing |
d1222754 | This paper examines two problems in document-level sentiment analysis: (1) determining whether a given document is a review or not, and (2) classifying the polarity of a review as positive or negative. We first demonstrate that review identification can be performed with high accuracy using only unigrams as features. We then examine the role of four types of simple linguistic knowledge sources in a polarity classification system. | Examining the Role of Linguistic Knowledge Sources in the Automatic Identification and Classification of Reviews |
d3171415 | Training a statistical machine translation starts with tokenizing a parallel corpus. Some languages such as Chinese do not incorporate spacing in their writing system, which creates a challenge for tokenization. Moreover, morphologically rich languages such as Korean present an even bigger challenge, since optimal token boundaries for machine translation in these languages are often unclear. Both rule-based solutions and statistical solutions are currently used. In this paper, we present unsupervised methods to solve tokenization problem. Our methods incorporate information available from parallel corpus to determine a good tokenization for machine translation. | Unsupervised Tokenization for Machine Translation |
d8488020 | At the request of the USG National Virtual Translation Center, the University of Maryland Center for Advanced Study of Language conducted a study that assessed the role of several factors mediating transcript usefulness during translation tasks. These factors included source language (Mandarin or Modern Standard Arabic), native speaker status of the translators, transcript quality (low or moderate word error rate), and transcript functionality (static or dynamic). Using 54 Mandarin and 54 Arabic translators (half native speakers in each language) and broadcast news clips for input, the study demonstrated that translation environments that provide dynamic transcripts with low or moderate word error rates are likely to improve performance (measured as integrated speed and accuracy scores) among non-native speakers without decreasing performance among native speakers. | The Use of Machine-Generated Transcripts During Human Translation |
d227905505 | ||
d10388687 | ANALYSING DICTIONARY DEFINITIONS OF MOTION VERBS | |
d9574081 | Nowadays many researches focus on the automatic recognition of sign language. High recognition rates are achieved using lot of training data. This data is, generally, collected by manual annotating SL video corpus. However this is time consuming and the results depend on the annotators knowledge. In this work we intend to assist the annotation in terms of glosses which consist on writing down the sign meaning sign for sign thanks to automatic video processing techniques. In this case using learning data is not suitable since at the first step it will be needed to manually annotate the corpus. Also the context dependency of signs and the co-articulation effect in continuous SL make the collection of learning data very difficult. Here we present a novel approach which uses lexical representations of sign to overcome these problems and image processing techniques to match sign performances to sign representations. Signs are described using Zeebede (ZBD) which is a descriptor of signs that considers the high variability of signs. A ZBD database is used to stock signs and can be queried using several characteristics. From a video corpus sequence features are extracted using a robust body part tracking approach and a semi-automatic sign segmentation algorithm. Evaluation has shown the performances and limitation of the proposed approach. | Semi-Automatic Sign Language Corpora Annotation using Lexical Representations of Signs |
d227230701 | ||
d6019059 | In this paper we address the problem of extracting key pieces of information from voicemail messages, such as the identity and phone number of the caller. This task differs from the named entity task in that the information we are interested in is a subset of the named entities in the message, and consequently, the need to pick the correct subset makes the problem more difficult. Also, the caller's identity may include information that is not typically associated with a named entity. In this work, we present three information extraction methods, one based on hand-crafted rules, one based on maximum entropy tagging, and one based on probabilistic transducer induction. We evaluate their performance on both manually transcribed messages and on the output of a speech recognition system. | Information Extraction From Voicemail |
d250390539 | The task of modal dependency parsing aims to parse a text into its modal dependency structure, which is a representation for the factuality of events in the text. We design a modal dependency parser that is based on priming pre-trained language models, and evaluate the parser on two data sets. Compared to baselines, we show an improvement of 2.6% in F-score for English and 4.6% for Chinese. To the best of our knowledge, this is also the first work on Chinese modal dependency parsing. | Modal Dependency Parsing via Language Model Priming |
d46481745 | The Corpus de Referência do Português Contemporâneo (CRPC) is being developed in the Centro de Linguística da Universidade de Lisboa (CLUL) since 1988 under a perspective of research data enlargement, in the sense of concepts and hypothesis verification by rejecting the sole use of intuitive data. The intention of creating this open corpus is to establish an on-line representative sample collection of general usage contemporary Portuguese: a main corpus of great dimension as well as several specialized corpora. The CRPC has nowadays around 92 million words. Following the use in this area, the CRPC project intends to establish a linguistic database accessible to everyone interested in making theoretical and practical studies or applications. The Dialectal oral corpus of the Atlas Linguístico-Etnográfico de Portugal e da Galiza (ALEPG) is constituted by approximately 3500 hours of speech collected by the CLUL Dialectal Studies Research Group and recorded in analogic audio tape. This corpus contains mainly directed speech: answers to a linguistic questionnaire essentially lexical, but also focusing on some phonetic and morpho-phonological phenomena. An important part of spontaneous speech enables other kind of studies such as syntactic, morphological or phonetic ones. | Portuguese corpora at CLUL |
d534638 | Cognitive linguists contend that learners' awareness of motivations is the key in not only second language acquisition but also figurative language learning. Two cognitive-oriented methods are proposed to raise L2 learners' awareness on metaphoric/metonymic expressions and to enhance retention: instruction involving conceptual metaphors (CM) and instruction involving metaphoric mappings (MM). The present study aims to examine their effectiveness in an EFL context. The results show favorable influences on learners' awareness and retention, which confirm that cognitive-oriented instructions indeed can assist learners to make better sense of figurative language. Moreover, the instruction on metaphoric mappings seems to result in better awareness of expressions which involve more complicated and abstract mapping relationships. The findings of the study can shed light on the application of metaphor and metonymy to EFL teaching and learning of figurative language | The Effects of EFL Learners' Awareness and Retention in Learning Metaphoric and Metonymic Expressions |
d587115 | This paper demonstrates that machine learning is a suitable approach for rapid parser development. From 1000 newly treebanked Korean sentences we generate a deterministic shift-reduce parser. The quality of the treebank, particularly crucial given its small size, is supported by a consistency checker. | Rapid Parser Development: A Machine Learning Approach for Korean |
d5385888 | This paper describes a new automatic approach for extracting conceptual distinctions from dictionary definitions. A broad-coverage dependency parser is first used to extract the lexical relations from the definitions. Then the relations are disambiguated using associations learned from tagged corpora. This contrasts with earlier approaches using manually developed rules for disambiguation. | Empirical Acquisition of Differentiating Relations from Definitions |
d209516094 | We present AETHEL, a semantic compositionality dataset for written Dutch. AETHEL consists of two parts. First, it contains a lexicon of supertags for about 900 000 words in context. The supertags correspond to types of the simply typed linear lambda-calculus, enhanced with dependency decorations that capture grammatical roles supplementary to function-argument structures. On the basis of these types, AETHEL further provides 72 192 validated derivations, presented in four equivalent formats: natural-deduction and sequent-style proofs, linear logic proofnets and the associated programs (lambda terms) for meaning composition. AETHEL's types and derivations are obtained by means of an extraction algorithm applied to the syntactic analyses of Lassy Small, the gold standard corpus of written Dutch. We discuss the extraction algorithm and show how 'virtual elements' in the original Lassy annotation of unbounded dependencies and coordination phenomena give rise to higher-order types. We suggest some example usecases highlighting the benefits of a type-driven approach at the syntax semantics interface. The following resources are open-sourced with AETHEL: the lexical mappings between words and types, a subset of the dataset consisting of 7 924 semantic parses, and the Python code that implements the extraction algorithm. | AETHEL: Automatically Extracted Typelogical Derivations for Dutch |
d14198180 | This paper summarises the contributions of the teams at the University of Helsinki, Uppsala University and the University of Turku to the news translation tasks for translating from and to Finnish. Our models address the problem of treating morphology and data coverage in various ways. We introduce a new efficient tool for word alignment and discuss factorisations, gappy language models and reinflection techniques for generating proper Finnish output. The results demonstrate once again that training data is the most effective way to increase translation performance. | Shared Task Papers |
d9596742 | This paper talks about the deciding practical sense boundary of homonymous words. The important problem in dictionaries or thesauri is the confusion of the sense boundary by each resource. This also becomes a bottleneck in the practical language processing systems. This paper proposes the method about discovering sense boundary using the collocation from the large corpora and the clustering methods. In the experiments, the proposed methods show the similar results with the sense boundary from a corpus-based dictionary and sense-tagged corpus. | Automatic clustering of collocation for detecting practical sense boundary |
d252395296 | Cued Speech is a communication system developed for deaf people to complement speechreading at the phonetic level with hands. This visual communication mode uses handshapes in different placements near the face in combination with the mouth movements of speech to make the phonemes of spoken language look different from each other. This paper describes CLeLfPC -Corpus de Lecture en Langue française Parlée Complétée, a corpus of French Cued Speech. It consists in about 4 hours of audio and HD video recordings of 23 participants. The recordings are 160 different isolated 'CV' syllables repeated 5 times, 320 words or phrases repeated 2-3 times and about 350 sentences repeated 2-3 times. The corpus is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. It can be used for any further research or teaching purpose. The corpus includes orthographic transliteration and other phonetic annotations on 5 of the recorded topics, i.e. syllables, words, isolated sentences and a text. The early results are encouraging: it seems that 1/ the hand position has a high influence on the key audio duration; and 2/ the hand shape has not. | CLeLfPC: a Large Open Multi-Speaker Corpus of French Cued Speech |
d243865277 | Most of the existing Knowledge-based Question Answering (KBQA) methods first learn to map the given question to a query graph, and then convert the graph to an executable query to find the answer. The query graph is typically expanded progressively from the topic entity based on a sequence prediction model. In this paper, we propose a new solution to query graph generation that works in the opposite manner: we start with the entire knowledge base and gradually shrink it to the desired query graph. This approach improves both the efficiency and the accuracy of query graph generation, especially for complex multi-hop questions. Experimental results show that our method achieves state-of-the-art performance on ComplexWebQuestion (CWQ) dataset. | Improving Query Graph Generation for Complex Question Answering over Knowledge Base |
d9431465 | We present an approach for opinion role induction for verbal predicates. Our model rests on the assumption that opinion verbs can be divided into three different types where each type is associated with a characteristic mapping between semantic roles and opinion holders and targets. In several experiments, we demonstrate the relevance of those three categories for the task. We show that verbs can easily be categorized with semi-supervised graphbased clustering and some appropriate similarity metric. The seeds are obtained through linguistic diagnostics. We evaluate our approach against a new manuallycompiled opinion role lexicon and perform in-context classification. | Opinion Holder and Target Extraction based on the Induction of Verbal Categories |
d225062651 | Joint entity and relation extraction has received increasing interests recently, due to the capability of utilizing the interactions between both steps. Among existing studies, the Multi-Head Selection (MHS) framework is efficient in extracting entities and relations simultaneously. However, the method is weak for its limited performance. In this paper, we propose several effective insights to address this problem. First, we propose an entity-specific Relative Position Representation (eRPR) to allow the model to fully leverage the distance information between entities and context tokens. Second, we introduce an auxiliary Global Relation Classification (GRC) to enhance the learning of local contextual features. Moreover, we improve the semantic representation by adopting a pre-trained language model BERT as the feature encoder. Finally, these new keypoints are closely integrated with the multi-head selection framework and optimized jointly. Extensive experiments on two benchmark datasets demonstrate that our approach overwhelmingly outperforms previous works in terms of all evaluation metrics, achieving significant improvements for relation F1 by +2.40% on CoNLL04 and +1.90% on ACE05, respectively. * Work done during an internship at Tencent. | Entity Relative Position Representation based Multi-head Selection for Joint Entity and Relation Extraction |
d219308971 | ||
d17119732 | We examine the problem of content selection in statistical novel sentence generation. Our approach models the processes performed by professional editors when incorporating material from additional sentences to support some initially chosen key summary sentence, a process we refer to as Sentence Augmentation. We propose and evaluate a method called "Seed and Grow" for selecting such auxiliary information. Additionally, we argue that this can be performed using schemata, as represented by word-pair co-occurrences, and demonstrate its use in statistical summary sentence generation. Evaluation results are supportive, indicating that a schemata model significantly improves over the baseline.Human-Authored SummarySentence: Repeated [poor seasonal rains] 1 [in 2004] 2 , culminating in [food insecurity] 3 , indicate [another year] 4 of crisis, the scale of which is larger than last year's and is further [exacerbated by diminishing coping assets] 5 [in both rural and urban areas] 6 . Key Source Sentence: The consequences of [another year] 4 of [poor rains] 1 on [food security] 3 are severe. Auxiliary Source Sentence(s): However in addition to the needs of economic recovery activities for IDPs, [food insecurity] 3 [over the majority of 2004] 2 [has created great stress] 5 on the poorest families in the country, [both within the urban and rural settings] 6 . | Seed and Grow: Augmenting Statistically Generated Summary Sentences using Schematic Word Patterns |
d21720529 | The paper presents a method for parsing low-resource languages with very small training corpora using multilingual word embeddings and annotated corpora of larger languages. The study demonstrates that specific language combinations enable improved dependency parsing when compared to previous work, allowing for wider reuse of pre-existing resources when parsing low-resource languages. The study also explores the question of whether contemporary contact languages or genetically related languages would be the most fruitful starting point for multilingual parsing scenarios. | Multilingual Dependency Parsing for Low-Resource Languages: Case Studies on North Saami and Komi-Zyrian |
d5719973 | We address the problem of unknown word sense detection: the identification of corpus occurrences that are not covered by a given sense inventory. We model this as an instance of outlier detection, using a simple nearest neighbor-based approach to measuring the resemblance of a new item to a training set. In combination with a method that alleviates data sparseness by sharing training data across lemmas, the approach achieves a precision of 0.77 and recall of 0.82. | Unknown word sense detection as outlier detection |
d252819026 | This paper addresses a deficiency in existing cross-lingual information retrieval (CLIR) datasets and provides a robust evaluation of CLIR systems' disambiguation ability. CLIR is commonly tackled by combining translation and traditional IR. Due to translation ambiguity, the problem of ambiguity is worse in CLIR than in monolingual IR. But existing auto-generated CLIR datasets are dominated by searches for named entity mentions, which does not provide a good measure for disambiguation performance, as named entity mentions can often be transliterated across languages and tend not to have multiple translations. Therefore, we introduce a new evaluation dataset (MuSe-CLIR) to address this inadequacy. The dataset focusses on polysemous common nouns with multiple possible translations. MuSeCLIR is constructed from multilingual Wikipedia and supports searches on documents written in European (French, German, Italian) and Asian (Chinese, Japanese) languages. We provide baseline statistical and neural model results on MuSeCLIR which show that MuSeCLIR has a higher requirement on the ability of systems to disambiguate query terms. | MuSeCLIR: A Multiple Senses and Cross-lingual Information Retrieval dataset |
d252624766 | The smiling synchrony of the French audio-video conversational corpora "PACO" and "Cheese!" is investigated. The two corpora merged altogether last 6 hours and are made of 25 face-to-face dyadic interactions annotated following the 5 levels Smiling Intensity Scale proposed by Gironzetti et al. (2016). After introducing new indicators for characterizing synchrony phenomena, we find that almost all the 25 interactions of PACO-CHEESE show a strong and significant smiling synchrony behavior. We investigate in a second step the evolution of the synchrony parameters throughout the interaction. No effect is found and it appears rather that the smiling synchrony is present at the very start of the interaction and remains unchanged throughout the conversation. | A Measure of the Smiling Synchrony in the Conversational Face-to-face Interaction Corpus PACO-CHEESE |
d236486177 | ||
d18366179 | In this paper we present a complete framework for the annotation of negation in Italian, which accounts for both negation scope and negation focus, and also for language-specific phenomena such as negative concord. In our view, the annotation of negation complements more comprehensive Natural Language Processing tasks, such as temporal information processing and sentiment analysis. We applied the proposed framework and the guidelines built on top of it to the annotation of written texts, namely news articles and tweets, thus producing annotated data for a total of over 36,000 tokens. | The Scope and Focus of Negation: A Complete Annotation Framework for Italian |
d208332148 | ||
d1259347 | This paper introduces the Alligator theorem prover for Dependent Type Systems (dts). We start with highlighting a number of properties of dts that make them specifically suited for computational semantics. We then briefly introduce dts and our implementation. The paper concludes with an example of a dts proof that illustrates the suitability of dts for modelling anaphora resolution. | The ALLIGATOR Theorem Prover for Dependent Type Systems: Description and Proof Sample |
d204912182 | ||
d1583857 | This paper presents the building procedure of a Chinese sense annotated corpus. A set of software tools is designed to help human annotator to accelerate the annotation speed and keep the consistency. The software tools include 1) a tagger for word segmentation and POS tagging, 2) an annotating interface responsible for the sense describing in the lexicon and sense annotating in the corpus, 3) a checker for consistency keeping, 4) a transformer responsible for the transforming from text file to XML format, and 5) a counter for sense frequency distribution calculating. | Building Chinese Sense Annotated Corpus with the Help of Software Tools |
d8390088 | We present a localisation component that supports the generation of cross-modal deictic expressions in the knowledge-based presentation system WIP. We deal with relative localisations (e.g., "The object to the left, of object X."), absolute localisations (e.g., "The object in the upl)er left part of the l)icture.") and corner localisations (e.g., "The object in the lower right corner of the l)icture"). In addition, we distinguish two localisation granularities, one less detailed (e.g., "the object to the left. of object X.") and one more detailed (e.g., "the object above and to the left. of object X."). We consider corner localisations to be similar to absolute localisations and in turn absolute localisations to be specialisations of relative localisations. This allows us to compute all three localisation types with one generic localisation procedure. As elementary localisations are derived from previously computed composite localisations, we can cope with both localisation granularities in a computationally efficient way. Based on these l)rimary localisation l)rocedures, we discuss how objects can be localised among several other objects. Finally we introduce group localisations (e.g., "The object to left, of the group of or, her objects.") and show how to deal with thern. | Generating Spatial Descriptions for Cross-modal References |
d2887802 | At present there is no publicly available data set to evaluate the performance of different summarization systems on the task of generating location-related extended image captions. In this paper we describe a corpus of human generated model captions in English and German. We have collected 932 model summaries in English from existing image descriptions and machine translated these summaries into German. We also performed post-editing on the translated German summaries to ensure high quality. Both English and German summaries are evaluated using a readability assessment as in DUC and TAC to assess their quality. Our model summaries performed similar to the ones reported inDang (2005)and thus are suitable for evaluating automatic summarization systems on the task of generating image descriptions for location related images. In addition, we also investigated whether post-editing of machine-translated model summaries is necessary for automated ROUGE evaluations. We found a high correlation in ROUGE scores between post-edited and non-post-edited model summaries which indicates that the expensive process of post-editing is not necessary. | Model Summaries for Location-related Images |
d16647406 | CLaC-CORE, an exhaustive feature combination system ranked 4th among 34 teams in the Semantic Textual Similarity shared task STS 2013. Using a core set of 11 lexical features of the most basic kind, it uses a support vector regressor which uses a combination of these lexical features to train a model for predicting similarity between sentences in a two phase method, which in turn uses all combinations of the features in the feature space and trains separate models based on each combination. Then it creates a meta-feature space and trains a final model based on that. This two step process improves the results achieved by singlelayer standard learning methodology over the same simple features. We analyze the correlation of feature combinations with the data sets over which they are effective. | CLaC-CORE: Exhaustive Feature Combination for Measuring Textual Similarity |
d2677577 | We propose a framework for generating an abstractive summary from a semantic model of a multimodal document. We discuss the type of model required, the means by which it can be constructed, how the content of the model is rated and selected, and the method of realizing novel sentences for the summary. To this end, we introduce a metric called information density used for gauging the importance of content obtained from text and graphical sources. | Towards a Framework for Abstractive Summarization of Multimodal Documents |
d257258158 | Building upon existing work on word order freedom and syntactic annotation, this paper investigates whether we can differentiate between findings that reveal inherent properties of natural languages and their syntax, and features dependent on annotations used in computing the measures. An existing quantifiable and linguistically interpretable measure of word order freedom in language is applied to take a closer look at the robustness of the basic measure (word order entropy) to variations in dependency corpora used in the analysis. Measures are compared at three levels of generality, applied to corpora annotated according to the Universal Dependencies v1 and v2 annotation guidelines, selecting 31 languages for analysis. Preliminary results show that certain measures, such as subject-object relation order freedom, are sensitive to slight changes in annotation guidelines, while simpler measures are more robust, highlighting aspects of these metrics that should be taken into consideration when using dependency corpora for linguistic analysis and generalisation. | What quantifying word order freedom can tell us about dependency corpora |
d12428726 | This paper presents a novel approach to improve reordering in phrase-based machine translation by using richer, syntactic representations of units of bilingual language models (BiLMs). Our method to include syntactic information is simple in implementation and requires minimal changes in the decoding algorithm. The approach is evaluated in a series of Arabic-English and Chinese-English translation experiments. The best models demonstrate significant improvements in BLEU and TER over the phrase-based baseline, as well as over the lexicalized BiLM byNiehues et al. (2011). Further improvements of up to 0.45 BLEU for Arabic-English and up to 0.59 BLEU for Chinese-English are obtained by combining our dependency BiLM with a lexicalized BiLM. An improvement of 0.98 BLEU is obtained for Chinese-English in the setting of an increased distortion limit. | Dependency-Based Bilingual Language Models for Reordering in Statistical Machine Translation |
d44090119 | Detecting irony is an important task to mine fine-grained information from social web messages. Therefore, the Semeval-2018 task 3 is aimed to detect the ironic tweets (subtask A) and their irony types (subtask B). In order to address this task, we propose a system based on a densely connected LSTM network with multi-task learning strategy. In our dense LSTM model, each layer will take all outputs from previous layers as input. The last LSTM layer will output the hidden representations of texts, and they will be used in three classification task. In addition, we incorporate several types of features to improve the model performance. Our model achieved an F-score of 70.54 (ranked 2/43) in the subtask A and 49.47 (ranked 3/29) in the subtask B. The experimental results validate the effectiveness of our system. | THU NGN at SemEval-2018 Task 3: Tweet Irony Detection with Densely Connected LSTM and Multi-task Learning |
d245838272 | ||
d14901938 | Due to instant availability and low cost, machine translation is becoming popular. Machine translation mediated communication plays a more and more important role in international collaboration. However, machine translators cannot guarantee high quality translation. In a multilingual communication task, many in-domain resources, for example domain dictionaries, are needed to promote translation quality. This raises the problem of how to help communication task designers provide higher quality translation systems, systems that can take advantage of various in-domain resources. The Language Grid, a service-oriented collective intelligent platform, allows in-domain resources to be wrapped into language services. For task-oriented translation, we propose service composition scenarios for the composition of different language services, where various in-domain resources are utilized effectively. We design the architecture, provide a script language as the interface for the task designer, which is easy for describing the composition scenario, and make a case study of a Japanese-English campus orientation task. Based on the case study, we analyze the increase in translation quality possible and the usage of in-domain resources. The results demonstrate a clear improvement in translation accuracy when the in-domain resources are used. | Service Composition Scenarios for Task-Oriented Translation |
d2288422 | Since the 1950s, linguists have been using short lists (40-200 items) of basic vocabulary as the central component in a methodology which is claimed to make it possible to automatically calculate genetic relationships among languages. In the last few years these methods have experienced something of a revival, in that more languages are involved, different distance measures are systematically compared and evaluated, and methods from computational biology are used for calculating language family trees. In this paper, we explore how this methodology can be extended in another direction, by using larger word lists automatically extracted from a parallel corpus using word alignment software. We present preliminary results from using the Europarl parallel corpus in this way for estimating the distances between some languages in the Indo-European language family. | Estimating Language Relationships from a Parallel Corpus. A Study of the Europarl Corpus |
d236486293 | ||
d18133122 | IntroductionSubstantial formal grammatical and lexical resources exist in various NLP systems and in the form of textbook specifications. In the present paper we report on experimental results obtained in manual, semi-antomatic and automatic migration of entire computational or textbook descriptions (as opposed to a more informal reuse of ideas or the design of a single "polytheoretic" representation) from a variety of formalisms into the ALEP formalism. 1 The choice of ALEP (a comparatively lean, typed feature structure formalism based on rewrite rules) was motivated by the assumption that the study would be most interesting if the target formalism is relatively mainstream without overt ideological commitments to particular grammatical theories. As regards the source formalisms we have attempted migrations of descriptions in HPSG (which uses fullytyped feature structures and has a strong 'non-derivational' flavour), ETS (an untyped stratificational formalism which essentially uses rewrite rules for feature structures and has run-time non-monotonic devices) and LFG (which is an un-typed constraint and CF-PSG based formalism with extensions such as existential, negative and global well-formedness constraints).1 The work reported in this paper was supported by the CEC as part of the project ET10/52. Reusability of grammatical resources is an important idea. Practically, it has obvious economic benefits in allowing grammars to be developed cheaply; for theoreticians it is important in allowing new formalisms to be tested out, quickly and in depth, by providing large-scale grammars. It is timely since substantial computational grammatical resources exist in various NLP systems, and large scale descriptions must be quickly produced if applications are to succeed. Meanwhile, in the CL community, there is a perceptible paradigm shift towards typed feature structure and constraint based systems and, if successful, migration allows such systems to be equipped with large bodies of descriptions drawn from existing resources.In principle, there are two approaches to achieving the reuse of grammatical and lexical resources. The first involves storing or developing resources in some theory neutral representation language, and is probably impossible in the current state of knowledge. In this paper, we focus on reusability through migration--the transfer of linguistic resources (grammatical and lexical descriptions) from one computational formalism into another (a target computational formalism). Migration can be completely manual (as when a linguist attempts to encode the analyses of a particular linguistic theory in some computationally interpreted formalism), semi-automatic or automatic. The starting resource can be a paper description or an implemented, runnable grammar.The literature on migration is thin, and practical experience is episodic at best. Shieber's work (e.g.[Shieber 1988]) is relevant, but this was concerned with relations between formalisms, rather than on migrating grammars per se. He studied the extent to which the formalisms of FUG, LFG and GPSG could be reduced to PATlt-II. Although these stud-12 | Experiments in Reusability of Grammatical Resources |
d18860232 | Informal language is actively used in network-mediated communication, e.g. chat room, BBS, email and text message. We refer the anomalous terms used in such context as network informal language (NIL) expressions. For example, " (ou3)" is used to replace " (wo3)" in Chinese ICQ. Without unconventional resource, knowledge and techniques, the existing natural language processing approaches exhibit less effectiveness in dealing with NIL text. We propose to study NIL expressions with a NIL corpus and investigate techniques in processing NIL expressions. Two methods for Chinese NIL expression recognition are designed in NILER system. The experimental results show that pattern matching method produces higher precision and support vector machines method higher F-1 measure. These results are encouraging and justify our future research effort in NIL processing. | NIL Is Not Nothing: Recognition of Chinese Network Informal Language Expressions |
d250390876 | This paper presents the transformer model built to participate in the SIGTYP 2022 Shared Task on the Prediction of Cognate Reflexes. It consists of an encoder-decoder architecture with multi-head attention mechanism. Its output is concatenated with the one hot encoding of the language label of an input character sequence to predict a target character sequence. The results show that the transformer outperforms the baseline rule-based system only partially. | A Transformer Architecture for the Prediction of Cognate Reflexes |
d226239285 | ||
d566648 | This paper presents a corpus search system utilizing lexical dependency structure. The user's query consists of a sequence of keywords. For a given query, the system automatically generates the dependency structure patterns which consist of keywords in the query, and returns the sentences whose dependency structures match the generated patterns. The dependency structure patterns are generated by using two operations: combining and interpolation, which utilize dependency structures in the searched corpus. The operations enable the system to generate only the dependency structure patterns that occur in the corpus. The system achieves simple and intuitive corpus search and it is enough linguistically sophisticated to utilize structural information. | A Corpus Search System Utilizing Lexical Dependency Structure |
d24711023 | Cet article présente un système d'identification des relations discursives dites « implicites » (à savoir, non explicitement marquées par un connecteur) pour le français. Etant donné le faible volume de données annotées disponibles, notre système s'appuie sur des données étiquetées automatiquement en supprimant les connecteurs non ambigus pris comme annotation d'une relation, une méthode introduite par(Marcu et Echihabi, 2002). Comme l'ont montré(Sporleder et Lascarides, 2008)pour l'anglais, cette approche ne généralise pas très bien aux exemples de relations implicites tels qu'annotés par des humains. Nous arrivons au même constat pour le français et, partant du principe que le problème vient d'une différence de distribution entre les deux types de données, nous proposons une série de méthodes assez simples, inspirées par l'adaptation de domaine, qui visent à combiner efficacement données annotées et données artificielles. Nous évaluons empiriquement les différentes approches sur le corpus ANNODIS : nos meilleurs résultats sont de l'ordre de 45.6% d'exactitude, avec un gain significatif de 5.9% par rapport à un système n'utilisant que les données annotées manuellement.ABSTRACTAutomatically identifying implicit discourse relations using annotated data and raw corporaThis paper presents a system for identifying « implicit » discourse relations (that is, relations that are not marked by a discourse connective). Given the little amount of available annotated data for this task, our system also resorts to additional automatically labeled data wherein unambiguous connectives have been suppressed and used as relation labels, a method introduced by (Marcu etEchihabi, 2002). As shown by(Sporleder et Lascarides, 2008)for English, this approach doesn't generalize well to implicit relations as annotated by humans. We show that the same conclusion applies to French due to important distribution differences between the two types of data. In consequence, we propose various simple methods, all inspired from work on domain adaptation, with the aim of better combining annotated data and artificial data. We evaluate these methods through various experiments carried out on the ANNODIS corpus : our best system reaches a labeling accuracy of 45.6%, corresponding to a 5.9% significant gain over a system solely trained on manually labeled data. MOTS-CLÉS : analyse du discours, relations implicites, apprentissage automatique. | Identification automatique des relations discursives « implicites » à partir de données annotées et de corpus bruts |
d1003611 | Semantic similarity is a central concept that extends across numerous fields such as artificial intelligence, natural language processing, cognitive science and psychology. Accurate measurement of semantic similarity between words is essential for various tasks such as, document clustering, information retrieval, and synonym extraction. We propose a novel model of semantic similarity using the semantic relations that exist among words. Given two words, first, we represent the semantic relations that hold between those words using automatically extracted lexical pattern clusters. Next, the semantic similarity between the two words is computed using a Mahalanobis distance measure. We compare the proposed similarity measure against previously proposed semantic similarity measures on Miller-Charles benchmark dataset and WordSimilarity-353 collection. The proposed method outperforms all existing web-based semantic similarity measures, achieving a Pearson correlation coefficient of 0.867 on the Millet-Charles dataset. | A Relational Model of Semantic Similarity between Words using Automatically Extracted Lexical Pattern Clusters from the Web |
d34293248 | Stylistic variations of language, such as formality, carry speakers' intention beyond literal meaning and should be conveyed adequately in translation. We propose to use lexical formality models to control the formality level of machine translation output. We demonstrate the effectiveness of our approach in empirical evaluations, as measured by automatic metrics and human assessments. | A Study of Style in Machine Translation: Controlling the Formality of Machine Translation Output |
d8806715 | We exploit and extend the Generative Lexicon Theory to develop a formal description of adnominal constituents in a lexicon which can deal with linguistic phenomena found in Japanese adnominal constituents. We classify the problematic behavior into "static disambiguation" and "dynamic disambiguation" tasks. Static disambiguation can be done using lexical information in a dictionary, whereas dynamic disambiguation requires inferences at the knowledge representation level. | Lexical Semantics to Disambiguate Polysemous Phenomena of Japanese Adnominal Constituents |
d13513838 | This paper presents a new general supervised word sense disambiguation method based on a relatively small syntactically parsed and semantically tagged training corpus. The method exploits a full sentential context and all the explicit semantic relations in a sentence to identify the senses of all of that sentence's content words. In spite of a very small training corpus, we report an overall accuracy of 80.3% (85.7, 63.9, 83.6 and 86.5%, for nouns, verbs, adjectives and adverbs, respectively), which exceeds the accuracy of a statistical sense-frequency based semantic tagging, the only really applicable general disambiguating technique. | I I I I I I 1 General Word Sense Disambiguation Method Based on a Full Sentential Context |
d59641 | We i)rcsent a method to r(:aliz(: th:xil)le mix(;(linitiative dialogue, in which the syst(:m can mak(, etti:ctive COlflirmation mad guidmn(:(: using (-oncel)t-leve,1 confidcn('e mcmsur(,s (CMs) derived from st)eech recognizer output in ord(:r to handl(: sl)eech recognition errors. W(: d(:tine two con('et)t-level CMs, which are oil COllt(~,Iltwords and on semantic-attrilmtes, using 10-best outtmts of the Sl)e(:ch r(:cognizt:r and l)arsing with t)hrmse-level grammars. Content-word CM is useflll for s(:lecting 1)]ausible int(:rl)retati(ms.Less contid(:nt illt(:rl)r(:tmtions arc given to confirmation 1)roc(:ss. The strat(:gy iml)roved the interpr(:tmtion accuracy l)y 11.5(/0. Moreover, th(: semanti(:-mttrilmt(: CM ix us(:d to (:stimmtc user's intention and generates syst(mi-initiative guidances (:v(,,n wh(:n suc(-(:sstSfl int(:rl)r(:tmtiol~ is not o|)tain(:(1. | Flexible Mixed-Initiative Dialogue Management using Concept-Level Confidence Measures of Speech Recognizer Output |
d2221473 | Dans cet article, nous présentons la campagne 2012 du défi fouille de texte (DEFT). Cette édition traite de l'indexation automatique par des mots-clés d'articles scientifiques au travers de deux pistes. La première fournit aux participants la terminologie des mots-clés employés dans les documents à indexer tandis que la seconde ne fournit pas cette terminologie, rendant la tâche plus complexe. Le corpus se compose d'articles scientifiques parus dans des revues de sciences humaines, indexés par leurs auteurs. Cette indexation sert de référence pour l'évaluation. Les résultats ont été évalués en termes de micro-mesures sur les rappel, précision et F-mesure calculés après lemmatisation de chaque mot-clé. Dans la piste fournissant la terminologie des mots-clés employés, la F-mesure moyenne est de 0,3575, la médiane de 0,3321 et l'écart-type de 0,2985 ; sur la seconde piste, en l'absence de terminologie, la F-mesure moyenne est de 0,2055, la médiane de 0,1901 et l'écart-type de 0,1516.ABSTRACTControlled and free indexing of scientific papers Presentation and results of the DEFT2012 text-mining challengeIn this paper, we present the 2012 edition of the DEFT text-mining challenge. This edition addresses the automatic, keyword-based indexing of scientific papers through two tracks. The first gives to the participants the terminology of keywords used to index the documents, while the second does not provide this terminology. The corpus is composed of scientific papers published in humanities journals, indexed by their authors. This indexing is used as a reference for the evaluation. The results have been evaluated in terms of micro-measures on the recall, precision and F-measure computed after keyword lemmatization. In the track giving the terminology of used keywords, the mean F-measure is 0.3575, the median is 0.3321 and the standard deviation is 0.2985 ; in the second track, the mean F-measure is 0.2055, the median is 0.1901 and the standard deviation is 0.1516. | Indexation libre et contrôlée d'articles scientifiques Présentation et résultats du défi fouille de textes DEFT2012 |
d257154231 | This study is the first likelihood ratio (LR)based forensic text comparison study in which each text is mapped onto an embedding vector using RoBERTa as the pre-trained model. The scores obtained with Cosine distance and probabilistic linear discriminant analysis (PLDA) were calibrated to LRs with logistic regression; the quality of the LRs was assessed by log LR cost ( ). Although the documents in the experiments were very short (maximum 100 words), the systems reached the values of 0.55595 and 0.71591 for the Cosine and PLDA systems, respectively. The effectiveness of deep-learning-based text representation is discussed by comparing the results of the current study to those of the previous studies of systems based on conventional feature engineering tested with longer documents. | |
d9351556 | We compare the effect of joint modeling of phonological features to independent feature detectors in a Conditional Random Fields framework. Joint modeling of features is achieved by deriving phonological feature posteriors from the posterior probabilities of the phonemes. We find that joint modeling provides superior performance to the independent models on the TIMIT phone recognition task. We explore the effects of varying relationships between phonological features, and suggest that in an ASR system, phonological features should be handled as correlated, rather than independent. | JOINT VERSUS INDEPENDENT PHONOLOGICAL FEATURE MODELS WITHIN CRF PHONE RECOGNITION |
d16020183 | The application of two-level morphology to non-concatenative German morphology | |
d225062656 | ||
d17678822 | Most of the world languages are resource-poor for statistical machine translation; still, many of them are actually related to some resource-rich language. Thus, we propose three novel, language-independent approaches to source language adaptation for resource-poor statistical machine translation. Specifically, we build improved statistical machine translation models from a resource-poor language POOR into a target language TGT by adapting and using a large bitext for a related resource-rich language RICH and the same target language TGT. We assume a small POOR-TGT bitext from which we learn word-level and phrase-level paraphrases and cross-lingual morphological variants between the resource-rich and the resource-poor language. Our work is of importance for resource-poor machine translation because it can provide a useful guideline for people building machine translation systems for resource-poor languages.Our experiments for Indonesian/Malay-English translation show that using the large adapted resource-rich bitext yields 7.26 BLEU points of improvement over the unadapted one and 3.09 BLEU points over the original small bitext. Moreover, combining the small POOR-TGT bitext with the adapted bitext outperforms the corresponding combinations with the unadapted bitext by 1.93-3.25 BLEU points. We also demonstrate the applicability of our approaches to other languages and domains. *Computational LinguisticsVolume 42, Number 2 bitexts. Unfortunately, collecting sufficiently large, high-quality bitexts is difficult, and thus most of the 6,500+ world languages are resource-poor for SMT. Fortunately, many of these resource-poor languages are related to some resource-rich language, with whom they overlap in vocabulary and share cognates, which offers opportunities for bitext reuse.Example pairs of such resource rich-poor languages include Spanish-Catalan, Finnish-Estonian, Swedish-Norwegian, Russian-Ukrainian, Irish-Gaelic Scottish, Standard German-Swiss German, Modern Standard Arabic-Dialectical Arabic (e.g., Gulf, Egyptian), Turkish-Azerbaijani, and so on.Previous work has already demonstrated the benefits of using a bitext for a related resource-rich language to X (e.g., X = English) to improve machine translation from a resource-poor language to X (Nakov and Ng 2009, 2012). Here we take a different, orthogonal approach: We adapt the resource-rich language to get closer to the resourcepoor one.We assume two bitexts: (1) a small bitext for a resource-poor source language S 1 and some target language T, and (2) a large bitext for a related resource-rich source language S 2 and the same target language T. We use these bitexts to learn word-level and phrase-level paraphrases and cross-lingual morphological variants between the resource-poor and resource-rich languages, S 1 and S 2 . We propose three approaches to adapt (the source side of) the large bitext for S 2 -T: word-level paraphrasing, phraselevel paraphrasing, and text rewriting using a specialized decoder. The first two approaches were proposed in our previous work(Wang, Nakov, and Ng 2012), and the third approach is novel and outperforms the other two in our experiments.Training on the adapted large bitext S 2 -T yields very significant improvements in translation quality compared with both training on the unadapted large bitext S 2 -T, and training on the small bitext for the resource-poor language S 1 -T. We further achieve very sizable improvements when combining the small bitext S 1 -T with the large adapted bitext S 2 -T, compared with combining the former with the unadapted bitext S 2 -T.Although here we focus on adapting Malay to look like Indonesian, we also demonstrate the applicability of our approach to another language pair, Bulgarian-Macedonian, which is also from a different domain.The remainder of this article is organized as follows. Section 2 presents an overview of related work. Section 3 introduces our target resource rich-poor language pair: Malay-Indonesian. Then, Section 4 presents our three approaches for source language adaptation. Section 5 describes the experimental set-up, after which we present the experimental results and discussions in Section 6. Section 7 contains deeper analysis of the obtained results. Finally, Section 8 concludes and points to possible directions for future work.Related WorkOne relevant line of research is on machine translation between closely related languages, which is arguably simpler than general SMT, and thus can be handled using word-for-word translation, manual language-specific rules that take care of the necessary morphological and syntactic transformations, or character-level translation/ transliteration. This has been tried for a number of language pairs including Czech- | Source Language Adaptation Approaches for Resource-Poor Machine Translation |
d430897 | Phonetic string transduction problems, such as letter-to-phoneme conversion and name transliteration, have recently received much attention in the NLP community. In the past few years, two methods have come to dominate as solutions to supervised string transduction: generative joint n-gram models, and discriminative sequence models. Both approaches benefit from their ability to consider large, flexible spans of source context when making transduction decisions. However, they encode this context in different ways, providing their respective models with different information. To combine the strengths of these two systems, we include joint n-gram features inside a state-of-the-art discriminative sequence model. We evaluate our approach on several letter-to-phoneme and transliteration data sets. Our results indicate an improvement in overall performance with respect to both the joint n-gram approach and traditional feature sets for discriminative models. | Integrating Joint n-gram Features into a Discriminative Training Framework |
d12638278 | We present a model for the inclusion of semantic role annotations in the framework of confidence estimation for machine translation. The model has several interesting properties, most notably: 1) it only requires a linguistic processor on the (generally well-formed) source side of the translation; 2) it does not directly rely on properties of the translation model (hence, it can be applied beyond phrase-based systems). These features make it potentially appealing for system ranking, translation re-ranking and user feedback evaluation. Preliminary experiments in pairwise hypothesis ranking on five confidence estimation benchmarks show that the model has the potential to capture salient aspects of translation quality. | Automatic Projection of Semantic Structures: an Application to Pairwise Translation Ranking |
d1291352 | Event recognitionWe present a program for segmenting texts according to the separate events they describe. A modular architecture is described that allows us to examine the contributions made by particular aspects of natural language to event structuring. This is applied in the context of terrorist news articles, and a technique is suggested for evaluating the resulting segmentations. We also examine the usefulness of various heuristics in forming these segmentations. | CONSTRAINT-BASED EVENT RECOGNITION FOR INFORMATION EXTRACTION |
d9296465 | We describe the framework and present detailed results of an evaluation of 1.500 dialogues recorded during a three-months field-trial of the ACCeSS Dialogue System. The system was routing incoming calls to agents of a call-center and handled about 100 calls per day. | Issues in the Evaluation of Spoken Dialogue Systems -Experience from the ACCeSS Project |
d8509375 | Importance weighting is a generalization of various statistical bias correction techniques. While our labeled data in NLP is heavily biased, importance weighting has seen only few applications in NLP, most of them relying on a small amount of labeled target data. The publication bias toward reporting positive results makes it hard to say whether researchers have tried. This paper presents a negative result on unsupervised domain adaptation for POS tagging. In this setup, we only have unlabeled data and thus only indirect access to the bias in emission and transition probabilities. Moreover, most errors in POS tagging are due to unseen words, and there, importance weighting cannot help. We present experiments with a wide variety of weight functions, quantilizations, as well as with randomly generated weights, to support these claims. | Importance weighting and unsupervised domain adaptation of POS taggers: a negative result |
d226239219 | ||
d237217077 | ||
d43449978 | In an early model of generative phonology the lexicon of a language contained entries with as few feature specifications as possible in the interest of economy. The blank feature specifications representing both nondistinctive features and those rendered redundant by sequential constraints were filled in by the same 9honological rules. At this point, the concept of ~ rules changing feature values was unclear. When the distinction between rules that fill in blanks and those that change feature values became clear, it was zmbodied in the concept of morpheme structure rules and P rules. The MS rules were further split into feature redundancy (segment structure) rules and sequ~tial constraint rules. The MS component bore a striking resemblence to the earlier "pkonotactic" sections of autonomous phonemic analyses, but the claim was made for I~S ~les that they explained what phonotactiee merely described. The MS rules formed a major part of Chomsky's "readjustment component" which rendered th~ output of the syntactic component fit to be the input to the phonological component. A fairly current version of ~his model is the following one from Harms' Introduction t__oo Phonological c~) | A NOTE ON MOR2H~E STRUCTURE IN GENF~TIVE |
d14359949 | We propose a new integrated approach based on Markov logic networks (MLNs), an effective combination of probabilistic graphical models and firstorder logic for statistical relational learning, to extracting relations between entities in encyclopedic articles from Wikipedia. The MLNs model entity relations in a unified undirected graph collectively using multiple features, including contextual, morphological, syntactic, semantic as well as Wikipedia characteristic features which can capture the essential characteristics of relation extraction task. This model makes simultaneous statistical judgments about the relations for a set of related entities. More importantly, implicit relations can also be identified easily. Our experimental results showed that, this integrated probabilistic and logic model significantly outperforms the current stateof-the-art probabilistic model, Conditional Random Fields (CRFs), for relation extraction from encyclopedic articles. | An Integrated Probabilistic and Logic Approach to Encyclopedia Relation Extraction with Multiple Features * |
d3031101 | We demonstrate that the bidirectionality of deep grammars, allowing them to generate as well as parse sentences, can be used to automatically and effectively identify errors in the grammars. The system is tested on two implemented HPSG grammars: Jacy for Japanese, and the ERG for English. Using this system, we were able to increase generation coverage in Jacy by 18% (45% to 63%) with only four weeks of grammar development. | Using Generation for Grammar Analysis and Error Detection |
d5558321 | This paper describes our system for subtask-A: SDQC for RumourEval, task-8 of SemEval 2017. Identifying rumours, especially for breaking news events as they unfold, is a challenging task due to the absence of sufficient information about the exact rumour stories circulating on social media. Determining the stance of Twitter users towards rumourous messages could provide an indirect way of identifying potential rumours. The proposed approach makes use of topic independent features from two categories, namely cue features and message specific features to fit a gradient boosting classifier. With an accuracy of 0.78, our system achieved the second best performance on subtask-A of Ru-mourEval. | UWaterloo at SemEval-2017 Task 8: Detecting Stance towards Rumours with Topic Independent Features |
d7126603 | Sentence alignment is the problem of making explicit the relations that exist between the sentences of two texts that are known to be mutual translations. Automatic sentence alignment methods typically face two kinds of difficulties. First, there is the question of robustness. In real life, discrepancies between the source-text and its translation are quite common: differences in layout, omissions, inversions, etc. Sentence alignment programs must be ready to deal with such phenomena. Then, there is the question of accuracy. Even when translations are "clean", alignment is still not a trivial matter: some decisions are hard to make, even for humans. We report here on the current state of our ongoing efforts to produce a sentence alignment program that is both robust and accurate. The method that we propose relies on two new alignment engines, and combines the robustness of so-called "character-based" methods with the accuracy of stochastic translation models. Experimental results are presented, that demonstrate the method's effectiveness, and highlight where problems remain to be solved. | BILINGUAL SENTENCE ALIGNMENT: BALANCING ROBUSTNESS AND ACCURACY |
d259370501 | In this work, we address the NER problem by splitting it into two logical sub-tasks: (1) Span Detection which simply extracts mention spans of entities, irrespective of entity type;(2) Span Classification which classifies the spans into their entity types. Further, we formulate both sub-tasks as question-answering (QA) problems and produce two leaner models which can be optimized separately for each sub-task. Experiments with four crossdomain datasets demonstrate that this two-step approach is both effective and time efficient. Our system, SplitNER outperforms baselines on OntoNotes5.0, WNUT17 and a cybersecurity dataset and gives on-par performance on BioNLP13CG. In all cases, it achieves a significant reduction in training time compared to its QA baseline counterpart. The effectiveness of our system stems from fine-tuning the BERT model twice, separately for span detection and classification. The source code can be found at github.com/c3sr/split-ner. | Split-NER: Named Entity Recognition via Two Question-Answering-based Classifications |
d236486091 | This paper describes Edinburgh's submissions to the IWSLT2021 multilingual speech translation (ST) task. We aim at improving multilingual translation and zero-shot performance in the constrained setting (without using any extra training data) through methods that encourage transfer learning and larger capacity modeling with advanced neural components. We build our end-to-end multilingual ST model based on Transformer, integrating techniques including adaptive speech feature selection, language-specific modeling, multitask learning, deep and big Transformer, sparsified linear attention and root mean square layer normalization. We adopt data augmentation using machine translation models for ST which converts the zero-shot problem into a zero-resource one. Experimental results show that these methods deliver substantial improvements, surpassing the official baseline by > 15 average BLEU and outperforming our cascading system by > 2 average BLEU. Our final submission achieves competitive performance (runner up). 1 | Edinburgh's End-to-End Multilingual Speech Translation System for IWSLT 2021 |
d220047314 | Exploiting sentence-level labels, which are easy to obtain, is one of the plausible methods to improve low-resource named entity recognition (NER), where token-level labels are costly to annotate. Current models for jointly learning sentence and token labeling are limited to binary classification. We present a joint model that supports multi-class classification and introduce a simple variant of self-attention that allows the model to learn scaling factors. Our model produces 3.78%, 4.20%, 2.08% improvements in F1 over the BiLSTM-CRF baseline on e-commerce product titles in three different low-resource languages: Vietnamese, Thai, and Indonesian, respectively. | Improving Low-Resource Named Entity Recognition using Joint Sentence and Token Labeling |
d14972216 | Dialogue systems typically follow a rigid pace of interaction where the system waits until the user has finished speaking before producing a response. Interpreting user utterances before they are completed allows a system to display more sophisticated conversational behavior, such as rapid turn-taking and appropriate use of backchannels and interruptions. We demonstrate a natural language understanding approach for partial utterances, and its use in a virtual human dialogue system that can often complete a user's utterances in real time. | Interpretation of Partial Utterances in Virtual Human Dialogue Systems |
d201671273 | Context gates are effective to control the contributions from the source and target contexts in the recurrent neural network (RNN) based neural machine translation (NMT). However, it is challenging to extend them into the advanced Transformer architecture, which is more complicated than RNN. This paper first provides a method to identify source and target contexts and then introduce a gate mechanism to control the source and target contributions in Transformer. In addition, to further reduce the bias problem in the gate mechanism, this paper proposes a regularization method to guide the learning of the gates with supervision automatically generated using pointwise mutual information. Extensive experiments on 4 translation datasets demonstrate that the proposed model obtains an averaged gain of 1.0 BLEU score over a strong Transformer baseline. | Regularized Context Gates on Transformer for Machine Translation |
d102471 | Information graphics (non-pictorial graphics such as bar charts or line graphs) are an important component of multimedia documents. Often such graphics convey information that is not contained elsewhere in the document. Thus document summarization must be extended to include summarization of information graphics. This paper addresses our work on graphic summarization. It argues that the message that the graphic designer intended to convey must play a major role in determining the content of the summary, and it outlines our approach to identifying this intended message and using it to construct the summary. | Extending Document Summarization to Information Graphics |
d5963937 | In word sense disambiguation (WSD), the heuristic of choosing the most common sense is extremely powerful because the distribution of the senses of a word is often skewed. The first (or predominant) sense heuristic assumes the availability of handtagged data. Whilst there are hand-tagged corpora available for some languages, these are relatively small in size and many word forms either do not occur, or occur infrequently. In this paper we investigate the performance of an unsupervised first sense heuristic where predominant senses are acquired automatically from raw text. We evaluate on both the SENSEVAL-2 and SENSEVAL-3 English allwords data. For accurate WSD the first sense heuristic should be used only as a back-off, where the evidence from the context is not strong enough. In this paper however, we examine the performance of the automatically acquired first sense in isolation since it turned out that the first sense taken from SemCor outperformed many systems in SENSEVAL-2. | Using Automatically Acquired Predominant Senses for Word Sense Disambiguation |
d174800420 | Tweets are short messages that often include specialized language such as hashtags and emojis. In this paper, we present a simple strategy to process emojis: replace them with their natural language description and use pretrained word embeddings as normally done with standard words. We show that this strategy is more effective than using pretrained emoji embeddings for tweet classification. Specifically, we obtain new state-of-the-art results in irony detection and sentiment analysis despite our neural network is simpler than previous proposals. | Incorporating Emoji Descriptions Improves Tweet Classification |
d18172062 | Temporal annotation is a complex task characterized by low markup speed and low inter-annotator agreements scores. Tango is a graphical annotation tool for temporal relations. It is developed for the TimeML annotation language and allows annotators to build a graph that resembles a timeline. Temporal relations are added by selecting events and drawing labeled arrows between them. Tango is integrated with a temporal closure component and includes features like SmartLink, user prompting and automatic linking of time expressions. Tango has been used to create two corpora with temporal annotation, TimeBank and the AQUAINT Opinion corpus. | Annotation of Temporal Relations with Tango |
d9996436 | This project aims to build an ontology authoring interface in which the user is engaged in a dialogue with the system in controlled natural language. To investigate what such a dialogue might be like, a layered annotation scheme is being developed for interactions between ontology authors and the Protégé ontology authoring environment. A pilot experiment has been conducted with ontology authors, which reveals the complexity of mapping between user-interface actions and acts that appear in natural language dialogues; it also suggests the addition of some unanticipated types of dialogue acts and points the way to some possible enhancements of the authoring interface.Questions | A Pilot Experiment in Knowledge Authoring as Dialogue * |
d16342306 | Identifying documents that describe a specific type of event is challenging due to the high complexity and variety of event descriptions. We propose a multi-faceted event recognition approach, which identifies documents about an event using event phrases as well as defining characteristics of the event. Our research focuses on civil unrest events and learns civil unrest expressions as well as phrases corresponding to potential agents and reasons for civil unrest. We present a bootstrapping algorithm that automatically acquires event phrases, agent terms, and purpose (reason) phrases from unannotated texts. We use the bootstrapped dictionaries to identify civil unrest documents and show that multi-faceted event recognition can yield high accuracy. | Multi-faceted Event Recognition with Bootstrapped Dictionaries |
d9214231 | Names of people, places, and organizations have unique linguistic properties, and they typically require special treatment in automatic processes. Appropriate processing of names is essential to achieve highquality information extraction, speech recognition, machine translation, and information management, yet most HLT applications provide limited specialized processing of names. Variation in the forms of names can make it difficult to retrieve names from data sources, to perform co-reference resolution across documents, or to associate instances of names with their representations in gazetteers and lexicons. Name matching has become critical in government contexts for checking watchlists and maintaining tax, health, and Social Security records. In commercial contexts, name matching is essential in credit, insurance, and legal applications.This tutorial will focus on personal names, with special attention given to Arabic names, though it will be clear that much of the material applies to other languages and to names of places and organizations. Case studies will be used to illustrate problems and approaches to solutions. Arabic names illustrate many of the issues encountered in multilingual name matching, among which are complex name structures and spelling variation due to morphophonemic alternation and competing transliteration conventions. | . What's in a Name: Current Methods, Applications, and Evaluation in Multilingual Name Search and Matching |
d935515 | ||
d219299773 | ||
d7521858 | Traditional approaches to Sentiment Analysis (SA) rely on large annotated data sets or wide-coverage sentiment lexica, and as such often perform poorly on under-resourced languages. This paper presents empirical evidence of an efficient SA approach using freely available machine translation (MT) systems to translate Arabic tweets to English, which we then label for sentiment using a state-of-theart English SA system. We show that this approach significantly outperforms a number of standard approaches on a gold-standard heldout data set, and performs equally well compared to more cost-intense methods with 76% accuracy. This confirms MT-based SA as a cheap and effective alternative to building a fully fledged SA system when dealing with under-resourced languages. | Benchmarking Machine Translated Sentiment Analysis for Arabic Tweets |
d42113570 | In this note, a straw man is destroyed, optimism is expressed, an existing system is sketched, and some issues are laid out.A Direct Approach. | Yes! NLP-based FL-ITS will be Important |
d8237760 | Many implementations of Latent Dirichlet Allocation (LDA), including those described inBlei et al. (2003), rely at some point on the removal of stopwords, words which are assumed to contribute little to the meaning of the text. This step is considered necessary because otherwise high-frequency words tend to end up scattered across many of the latent topics without much rhyme or reason. We show, however, that the 'problem' of high-frequency words can be dealt with more elegantly, and in a way that to our knowledge has not been considered in LDA, through the use of appropriate weighting schemes comparable to those sometimes used in Latent Semantic Indexing (LSI). Our proposed weighting methods not only make theoretical sense, but can also be shown to improve precision significantly on a non-trivial cross-language retrieval task. | Term Weighting Schemes for Latent Dirichlet Allocation |
d219300494 | We are developing COMET, an interactive system that generates multimedia explanations of how to operate, maintain, and repair equipment. Our research stresses the dynamic generation of the content and form of all material presented, addressing issues in the generation of text and graphics, and in coordinating text and graphics in an integrated presentation.COMET contains a static knowledge base describing objects and plans for maintenance and repair, and a dynamic knowledge source for diagnosing failures. Explanations are produced using a content planner that determines what information should be communicated, a media coordinator that determines which informarion should be realized in graphics and which in text, and separate text and graphics generators. The graphics and text for a single explanation are laid out on the screen by a media layout component. A menu interface allows users to request explanations of specific procedures or to specify failure symptoms that will invoke a diagnostic component. The diagnostic component can ask the user to carry out procedures that COMET will explain if requested. In contrast to hypermedia systems that present previously authored material, COMET has underlying models of the user and context that allow each aspect of the explanation generated to be based on the current situation.Our focus in the text generation component has been on the development of the Functional Unification Formalism (FFUF) for non-syntactic tasks, of a large syntactic grammar in FUF, of lexical choice in FUF using constraints from underlying knowledge sources and from past discourse, and of models of constraints on several classes of word• choice. Important results in knowledge-based graphics generation include the automated design of 3D technical illustrations that contain nested insets, algorithms for and rule-based application of illustrative techniques such as cutaway views, a design-grid--based methodology • for display layout, and development of a testbed for knowledgebased animation.Finally, we have had significant results in the development of our media coordinator which, unlike other systems, features a common description language that allows a fine-grained division of information between text and graphics. The media coordinator maps information to media specific resources, and allows informarion expressed in one media to influence realization in the other. This allows for tight integration and coordination between different media.RECENT RESULTS• Incorporated user model constraints on word selection in order to use words appropriate to user's vocabulary level. This includes both word substitution and replanning of sentence content when there is no word that can be substituted for unknown word (e.g., "Check the polarity." is replaced by "Make sure the plus lines up with the plus.")• Completed sentence-picture coordination, allowing longer sentences to be broken into shorter ones that can separately accompany each generated picture when necessal T .• Added all m&r procedures for the radio from the manual to the knowledge base and augmented the lexicon to inelude new words for the procedures.• Continued implementation of cross-references between text and graphics, including query facilities for the graphics representation that allow the text generator to determine where and how an object is displayed, use of these facilities along with the underlying knowledge base to construct cross-references (e.g., "The battery is shown in the cutaway view of the radio."), and development of a lexicon for such cross-references.• Extended the graphics generator to support the maintenanee of visibility constraints through a set of illustrative techniques modeled after those used by technical illustrators. These involve detecting objects that obscure those that must remain visible and rendering the obscuring objects using transparency, cutaway views, and "ghosting" effects. The effects are invoked automatically as the graphics generator designs its illustrations.• Developed facilities for dynamic illustrations that are inerementally redesigned to allow users to explore the generated pictures by choosing viewpoints different from those selected by the system.PLANS FOR THE COMING YEARWe plan to finish implementation of cross references between text and graphics, to increase the ways in which the user model can influence lexical choice, and to incorporate all extensions as part of our demo system. Following that, we will move to a new contract, where we will begin work on identifying usage constraints on a variety of lexical classes through automatic and manual examination of large text corpora.413 | Interactive Multimedia Explanation for Equipment Maintenance and Repair |
d250390510 | Spoken 'grammatical error correction' (SGEC) is an important process to provide feedback for second language learning. Due to a lack of end-to-end training data, SGEC is often implemented as a cascaded, modular system, consisting of speech recognition, disfluency removal, and grammatical error correction (GEC). This cascaded structure enables efficient use of training data for each module. It is, however, difficult to compare and evaluate the performance of individual modules as preceeding modules may introduce errors. For example the GEC module input depends on the output of nonnative speech recognition and disfluency detection, both challenging tasks for learner data. This paper focuses on the assessment and development of SGEC systems. We first discuss metrics for evaluating SGEC, both individual modules and the overall system. The systemlevel metrics enable tuning for optimal system performance. A known issue in cascaded systems is error propagation between modules. To mitigate this problem semi-supervised approaches and self-distillation are investigated. Lastly, when SGEC system gets deployed it is important to give accurate feedback to users. Thus, we apply filtering to remove edits with low-confidence, aiming to improve overall feedback precision. The performance metrics are examined on a Linguaskill multi-level data set, which includes the original non-native speech, manual transcriptions and reference grammatical error corrections, to enable system analysis and development. | On Assessing and Developing Spoken 'Grammatical Error Correction' Systems |
d18335639 | INTRODUCTIO NThis paper describes the Unisys MUC-3 text understanding system, a system based upon a threetiered approach to text processing in which a powerful knowledge-based form of information retrieva l plays a central role . This knowledge-based form of information retrieval makes it possible to define a n effective level of text analysis that falls somewhere between what is possible with standard keyword-base d information retrieval techniques and deep linguistic analysis .The Unisys Center for Advanced Information Technology (CAIT) has a long-standing commitment to NLP research and development, with the Pundit NLP system developed at CAIT serving as the Center' s primary research vehicle[3]. The Unisys MUC-3 system, however, consists primarily of components that are less than 7 months old and still in a developmental stage . Although the three-tiered processing approach that the MUC-3 system's architecture is based upon includes Pundit as its third level of (linguistic ) analysis, the incorporation of Pundit into the MUC-3 system was not achieved in time for the final MUC-3 test in May, 1991 . A decision was made to focus on the development of a knowledge-based informatio n retrieval component, and this precluded the integration of Pundit into the prototype) The Unisys MUC-3 system without its linguistic analysis component is depicted inFigure 1. This is the version of the syste m that was actually used in the MUC-3 test .APPROACH AND SYSTEM DESCRIPTIO NThe Unisys MUC-3 system's architecture consists of five main processing components, three of whic h represent levels of text understanding . An initial preprocessing component transforms texts into a n appropriate format for the text understanding components to manipulate . The three text understandin g components engage in (1) standard keyword-based information retrieval, (2) knowledge-based informatio n retrieval, and (3) linguistic analysis . 2 A final, template generation component gathers together all th e facts extracted from a given text and builds template data structures out of them . These five components are described in more detail below . | UNISYS : DESCRIPTION OF THE UNISYS SYSTEM USED FOR MUC-3 INTRODUCTIO N |
d252091136 | This paper summarizes the current state of development in improving an open source subtitling tool. This includes improvements to the speech recognition model for German, the replacement for the punctuation reconstruction architecture and the addition of an audio segmentation. The goal of these adjustments is an overall better subtitle quality. The most crucial part of the existing pipeline, the German speech recognition, is replaced by a new Kaldi TDNN-HMM model trained on 70% of additional audio data, resulting in a word error rate of 6.9% on Tuda-De. The punctuation reconstruction model for German texts is replaced by a Transformer-based approach that is also trained on new data. English is added as a fully supported second language, including speech recognition and punctuation reconstruction models. Furthermore, to improve speech recognition in long videos, audio segmentation was also added into the pipeline to support long videos flawlessly without quality issues. | Improved Open Source Automatic Subtitling for Lecture Videos |
d62435274 | This paper presents a methodology that aims at building knowledge models from a natural language description of a domain. Our methodology is based on the establishment of a dialogue with the knowledge engineer of an application. This dialogue is motivated by the Semantic Differentiation Process, which solves problems related to acquisition and modelling. Moreover, the dialogue can be naturally formalised within a theory of communicating rational agents. We can thus consider a more complete automation of the process of modelling and show how to integrate our methodology into this type of theory. | A Rational Agent for the Construction of a Semantic Model |
d1440229 | We classify and review current approaches to software infrastructure for research, development and delivery of NLP systems. The task is motivated by a discussion of current trends in the field of NLP and Language Engineering. We describe a system called GATE (a General Architecture for Text Engineering) that provides a software infrastructure on top of which heterogeneous NLP processing modules may be evaluated and refined individually, or may be combined into larger application systems. GATE aims to support both researchers and developers working on component technologies (e.g. parsing, tagging, morphological analysis) and those working on developing end-user applications (e.g. information extraction, text summarisation, document generation, machine translation, and second language learning). GATE promotes reuse of component technology, permits specialisation and collaboration in large-scale projects, and allows for the comparison and evaluation of alternative technologies. The first release of GATE is now available. | Software Infrastructure for Natural Language Processing |
d8276051 | This paper summarizes the current status of version 2 of the Open Lexicon Interchange Format (OLIF). As a natural extension of the OLIF prototype (OLIF version 1), version 2 has been modified with respect to content and formalization (e.g., it is now XMLcompliant). These enhancements now make it possible to use OLIF in a variety of Natural Language Processing applications and general language technology environments (e.g., terminology management systems). At the time of writing, several industrial partners of the OLIF Consortium had already started work on implementing OLIF support. Details on OLIF can be found on www.olif.net. | The Open Lexicon Interchange Format (OLIF) Comes of Age |
d202624472 | We describe our submissions to the 4th edition of the Social Media Mining for Health Applications (SMM4H) shared task. Our team (UZH) participated in two sub-tasks: Automatic classifications of adverse effects mentions in tweets (Task 1) and Generalizable identification of personal health experience mentions (Task 4). For our submissions, we exploited ensembles based on a pretrained language representation with a neural transformer architecture (BERT) (Tasks 1 and 4) and a CNN-BiLSTM(-CRF) network within a multi-task learning scenario (Task 1). These systems are placed on top of a carefully crafted pipeline of domain-specific preprocessing steps. | Approaching SMM4H with Merged Models and Multi-task Learning |
d441443 | An efficient bit-vector-based CKY-style parser for context-free parsing is presented. The parser computes a compact parse forest representation of the complete set of possible analyses for large treebank grammars and long input sentences. The parser uses bit-vector operations to parallelise the basic parsing operations. The parser is particularly useful when all analyses are needed rather than just the most probable one. | Efficient Parsing of Highly Ambiguous Context-Free Grammars with Bit Vectors |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.