_id stringlengths 4 10 | text stringlengths 0 18.4k | title stringlengths 0 8.56k |
|---|---|---|
d248780450 | The influence of fake news in the perception of reality has become a mainstream topic in the last years due to the fast propagation of misleading information. In order to help in the fight against misinformation, automated solutions to fact-checking are being actively developed within the research community. In this context, the task of Automated Claim Verification is defined as assessing the truthfulness of a claim by finding evidence about its veracity. In this work we empirically demonstrate that enriching a BERT model with explicit semantic information such as Semantic Role Labelling helps to improve results in claim verification as proposed by the FEVER benchmark. Furthermore, we perform a number of explainability tests that suggest that the semantically-enriched model is better at handling complex cases, such as those including passive forms or multiple propositions. | A Semantics-Aware Approach to Automated Claim Verification |
d16290774 | We present a model to perform authorship attribution of tweets using Convolutional Neural Networks (CNNs) over character n-grams. We also present a strategy that improves model interpretability by estimating the importance of input text fragments in the predicted classification. The experimental evaluation shows that text CNNs perform competitively and are able to outperform previous methods. | Convolutional Neural Networks for Authorship Attribution of Short Texts |
d459520 | In this paper, we encode topic dependencies in hierarchical multi-label Text Categorization (TC) by means of rerankers. We represent reranking hypotheses with several innovative kernels considering both the structure of the hierarchy and the probability of nodes. Additionally, to better investigate the role of category relationships, we consider two interesting cases: (i) traditional schemes in which node-fathers include all the documents of their child-categories; and (ii) more general schemes, in which children can include documents not belonging to their fathers. The extensive experimentation on Reuters Corpus Volume 1 shows that our rerankers inject effective structural semantic dependencies in multi-classifiers and significantly outperform the state-of-the-art. | Modeling Topic Dependencies in Hierarchical Text Categorization |
d196205549 | Existing datasets for scoring text pairs in terms of semantic similarity contain instances whose resolution differs according to the degree of difficulty. This paper proposes to distinguish obvious from non-obvious text pairs based on superficial lexical overlap and ground-truth labels. We characterise existing datasets in terms of containing difficult cases and find that recently proposed models struggle to capture the non-obvious cases of semantic similarity. We describe metrics that emphasise cases of similarity which require more complex inference and propose that these are used for evaluating systems for semantic similarity. | Aiming beyond the Obvious: Identifying Non-Obvious Cases in Semantic Similarity Datasets |
d27057113 | A Discrete-cepstrum Based Spectrum-envelope Estimation Scheme and Its Application to Voice Transformation | |
d11054881 | In the field of empirical natural language processing, researchers constantly deal with large amounts of marked-up data; whether the markup is done by the researcher or someone else, human nature dictates that it will have errors in it. This paper will more fully characterise the problem and discuss whether and when (and how) to correct the errors. The discussion is illustrated with specific examples involving function tagging in the Penn treebank. | Handling noisy training and testing data |
d248780233 | By sharing parameters and providing taskindependent shared features, multi-task deep neural networks are considered one of the most interesting ways for parallel learning from different tasks and domains. However, fine-tuning on one task may compromise the performance of other tasks or restrict the generalization of the shared learned features. To address this issue, we propose to use task uncertainty to gauge the effect of the shared feature changes on other tasks and prevent the model from overfitting or over-generalizing. We conducted an experiment on 16 text classification tasks, and findings showed that the proposed method consistently improves the performance of the baseline, facilitates the knowledge transfer of learned features to unseen data, and provides explicit control over the generalization of the shared model. | Uncertainty Regularized Multi-Task Learning |
d9588761 | The new trend in sentiment classification is to use semantic features for representation of documents. We propose a semantic space based on WordNet senses for a supervised document-level sentiment classifier. Not only does this show a better performance for sentiment classification, it also opens opportunities for building a robust sentiment classifier. We examine the possibility of using similarity metrics defined on WordNet to address the problem of not finding a sense in the training corpus. Using three popular similarity metrics, we replace unknown synsets in the test set with a similar synset from the training set. An improvement of 6.2% is seen with respect to baseline using this approach. | Robust Sense-Based Sentiment Classification |
d10206811 | In this paper, we propose to introduce syntactic classes in a lexicalized dependency formalism. Subcategories of words are organized hierarchically from a general, abstract level (syntactic categories) to a word-specific level (single lexical items). The formalism is parsimonious, and useful for processing. We also sketch a parsing model that uses the hierarchical mixed-grain representation to make predictions on the structure of the input. | Integration of syntactic and lexical information in a hierarchical dependency grammar |
d18522898 | The ready availability of free online machine translation (MT) systems has given rise to a problem in the world of language teaching in that students -especially weaker ones -use free online MT to do their translation homework. Apart from the pedagogic implications, one question of interest is whether we can devise any techniques for automatically detecting such use. This paper reports an experiment which aims to address this particular problem, using methods from the broader world of computational stylometry, plagiarism detection, text reuse, and MT evaluation. A pilot experiment comparing 'honest' and 'derived' translations produced by 25 intermediate learners of Spanish, Italian and German is reported. | Detecting Inappropriate Use of Free Online Machine Translation by Language Students -A Special Case of Plagiarism Detection |
d6849510 | This paper describes the UPC submissions to the WMT14 Metrics Shared Task: UPC-IPA and UPC-STOUT. These metrics use a collection of evaluation measures integrated in ASIYA, a toolkit for machine translation evaluation. In addition to some standard metrics, the two submissions take advantage of novel metrics that consider linguistic structures, lexical relationships, and semantics to compare both source and reference translation against the candidate translation. The new metrics are available for several target languages other than English. In the the official WMT14 evaluation, UPC-IPA and UPC-STOUT scored above the average in 7 out of 9 language pairs at the system level and 8 out of 9 at the segment level. | IPA and STOUT: Leveraging Linguistic and Source-based Features for Machine Translation Evaluation |
d251747 | This paper reports on the annotation of all English verbs included in WordNet 2.0 with TimeML event classes. Two annotators assign each verb present in WordNet the most relevant event class capturing most of that verb's meanings. At the end of the annotation process, inter-annotator agreement is measured using kappa statistics, yielding a kappa value of 0.87. The cases of disagreement between the two independent annotations are clarified by obtaining a third, and in some cases, a fourth opinion, and finally each of the 11,306 WordNet verbs is mapped to a unique event class. The resulted annotation is then employed to automatically assign the corresponding class to each occurrence of a finite or non-finite verb in a given text. The evaluation performed on TimeBank reveals an F-measure of 86.43% achieved for the identification of verbal events, and an accuracy of 85.25% in the task of classifying them into TimeML event classes. | Annotation of WordNet Verbs with TimeML Event Classes |
d15033136 | This paper presents an algorithm for automatic word forms inflection. We use the method of longest common subsequence to extract abstract paradigms from given pairs of basic and inflected word forms, as well as suffix and prefix features to predict this paradigm automatically. We elaborate this algorithm using combination of affix feature-based and character ngram models, which substantially enhances performance especially for the languages possessing nonlocal phenomena such as vowel harmony. Our system took part in SIGMORPHON 2016 Shared Task and took 3rd place in 17 of 30 subtasks and 4th place in 7 substasks among 7 participants. | Using longest common subsequence and character models to predict word forms |
d219309416 | ||
d227231404 | ||
d196172425 | ||
d16970678 | Our goal is to generate reading lists for students that help them optimally learn technical material. Existing retrieval algorithms return items directly relevant to a query but do not return results to help users read about the concepts supporting their query. This is because the dependency structure of concepts that must be understood before reading material pertaining to a given query is never considered. Here we formulate an information-theoretic view of concept dependency and present methods to construct a "concept graph" automatically from a text corpus. We perform the first human evaluation of concept dependency edges (to be published as open data), and the results verify the feasibility of automatic approaches for inferring concepts and their dependency relations. This result can support search capabilities that may be tuned to help users learn a subject rather than retrieve documents based on a single query. | Modeling Concept Dependencies in a Scientific Corpus |
d11561366 | In this work we present a novel technique to rescore fragments in the Data-Oriented Translation model based on their contribution to translation accuracy. We describe three new rescoring methods, and present the initial results of a pilot experiment on a small subset of the Europarl corpus. This work is a proof-of-concept, and is the first step in directly optimizing translation decisions solely on the hypothesized accuracy of potential translations resulting from those decisions. | Accuracy-Based Scoring for DOT: Towards Direct Error Minimization for Data-Oriented Translation |
d13452846 | We introduce a new multi-threaded parsing algorithm on unification grammars designed specifically for multimodal interaction and noisy environments. By lifting some traditional constraints, namely those related to the ordering of constituents, we overcome several difficulties of other systems in this domain. We also present several criteria used in this model to constrain the search process using dynamically loadable scoring functions. Some early analyses of our implementation are discussed. | Clavius: Bi-Directional Parsing for Generic Multimodal Interaction |
d2988527 | We use words to talk about the world. Therefore, to understand what words mean, we must have a prior explication of how we view the world. In a sense, efforts in the past to decompose words into semantic primitives were attempts to link word meaning to a theory of the world, where the set of semantic primitives constituted the theory of the world. With the advent of naive physics and research programs to formalize commonsense knowledge in a number of areas in predicate calculus or some other formal language, we now have at our disposal means for building much richer theories of varlous aspects of the world, and consequently, we are in a much better position to address the problems of le~cal semantics.In the TACITUS project for using commonsense.knowledge in the understanding of texts about mechanical devices and their failures, we have been developing various commonsense theories that are needed to mediate between the way we talk about the behavior of such devices and causal models of their operation (Hobbs et M., 1986). The theories cover a number of areas that figure in virtually every domain of discourse, such as scalar notions, granularity, structured systems, time, space, material, physical objects, causality, functionality, force, and shape. Our approach has been to construct core theories of each of these areas. These core theories may use English words as their predicates, but the principal criterion for adequacy of the core theory is elegance, whatever that is, and this can usually be achieved better using predicates that are not lexically realized. It is easier to achieve elegance if one does not have to be held responsible to linguistic evidence. Predicates that are lexically realized are then pushed to the periphery of the theory. A large number of lexical items can be defined, or at least characterized, in terms provided by the core theories. The hypothesis is that once these core theories have been formulated in the right way, it will be straightforward to explicate the meanings of a great many words.The phrase "in the right way" is key in this strategy. The world is 20 | World Knowledge and Word Meaning |
d4350 | We present results from the February '92 evaluation on the ATIS travel planning domain for HARC, the BBN spoken language system (SLS). In addition, we discuss in detail the individual perfor-2. mance of BYBLOS, the speech recognition (SPREC) component.In the official scoring, conducted by NIST, BBN's HARC system 3. produced a weighted SLS score of 43.7 on all 687 evaluable utterances in the test set. This was the lowest error achieved by any of the 7 systems evaluated.4. For the SPREC evaluation BBN's BYBLOS system achieved a word error rate of 6.2% on the same 687 utterances and 9.4% on the entire test set of 971 utterances. These results were significantly better than any other speech system evaluated. | BBN BYBLOS and HARC February 1992 ATIS Benchmark Results |
d233365134 | ||
d5320074 | Lexicalized Tree Adjoining Grammar (LTAG) is an attractive formalism for linguistic description mainly because cff its extended domain of locality and its factoring recursion out from the domain of local dependencies(Joshi, 1985, Kroch and Joshi, 1985, Abeilld, 1988. LTAG's extended domain of locality enables one to localize syntactic dependencies (such as filler-gap), as well as semantic dependencies (such as predicate-arguments). The aim of this paper is to show that these properties combined with the lexicalized property of LTAG are especially attractive for machine translation.The transfer between two languages, such as French and English, can be done by putting directly into correspondence large elementary units without going through some interlingual representation and without major changes to the source and target grammars. The underlying formalism for the transfer is "synchronous Tree Adjoining Grammars" (Shieber and ) 1. Transfer rules are stated as correspondences between nodes of trees of large domain of locality which are associated with words. We can thus define lexical transfer rules that avoid the defects of a mere word-to-word approach but still benefit from the simplicity and elegance of a lexical approach.We rely on the French and English LTAG grammars (Abeille [1988], Abeille [1.990 (b)], Abeilld et al. [1990], Abeill6 and Schabes [1989, 1990]) that have been designed over the past two years jointly at | Using Lexicalized Tags for Machine Translation * |
d252377412 | The manner in which gender is portrayed in materials used to teach children conveys messages about people's roles in society. In this paper, we measure the gendered depiction of central domains of social life in 100 years of highly influential children's books. We make two main contributions: (1) we find that the portrayal of gender in these books reproduces traditional gender norms in society, and (2) we publish StoryWords 1.0, the first word embeddings trained on such a large body of children's literature. We find that, relative to males, females are more likely to be represented in relation to their appearance than in relation to their competence; second, they are more likely to be represented in relation to their role in the family than their role in business. Finally, we find that non-binary or gender-fluid individuals are rarely mentioned. Our analysis advances understanding of the different messages contained in content commonly used to teach children, with immediate applications for practice, policy, and research. | Tales and Tropes: Gender Roles from Word Embeddings in a Century of Children's Books Tales and Tropes: Gender Roles from Word Embeddings in a Century of Children's Books * |
d8940890 | Recent dysarthric speech recognition studies using mixed data from a collection of neurological diseases suggested articulatory data can help to improve the speech recognition performance. This project was specifically designed for the speakerindependent recognition of dysarthric speech due to amyotrophic lateral sclerosis (ALS) using articulatory data. In this paper, we investigated three across-speaker normalization approaches in acoustic, articulatory, and both spaces: Procrustes matching (a physiological approach in articulatory space), vocal tract length normalization (a data-driven approach in acoustic space), and feature space maximum likelihood linear regression (a model-based approach for both spaces), to address the issue of high degree of variation of articulation across different speakers. A preliminary ALS data set was collected and used to evaluate the approaches. Two recognizers, Gaussian mixture model (GMM) -hidden Markov model (HMM) and deep neural network (DNN) -HMM, were used. Experimental results showed adding articulatory data significantly reduced the phoneme error rates (PERs) using any or combined normalization approaches. DNN-HMM outperformed GMM-HMM in all configurations. The best performance (30.7% PER) was obtained by triphone DNN-HMM + acoustic and articulatory data + all three normalization approaches, a 15.3% absolute PER reduction from the baseline using triphone GMM-HMM + acoustic data. | Recognizing Dysarthric Speech due to Amyotrophic Lateral Sclerosis with Across-Speaker Articulatory Normalization |
d49563344 | We propose to study the evolution of concepts by learning to complete diachronic analogies between lists of terms which relate to the same concept at different points in time. We present a number of models based on operations on word embedddings that correspond to different assumptions about the characteristics of diachronic analogies and change in concept vocabularies. These are tested in a quantitative evaluation for nine different concepts on a corpus of Dutch newspapers from the 1950s and 1980s. We show that a model which treats the concept terms as analogous and learns weights to compensate for diachronic changes (weighted linear combination) is able to more accurately predict the missing term than a learned transformation and two baselines for most of the evaluated concepts. We also find that all models tend to be coherent in relation to the represented concept, but less discriminative in regard to other concepts. Additionally, we evaluate the effect of aligning the time-specific embedding spaces using orthogonal Procrustes, finding varying effects on performance, depending on the model, concept and evaluation metric. For the weighted linear combination, however, results improve with alignment in a majority of cases. All related code is released publicly. This work is licensed under a Creative Commons Attribution 4.0 International License.License details: | Learning Diachronic Analogies to Analyze Concept Change |
d232021849 | ||
d242168 | To alleviate data sparsity in spoken Uyghur machine translation, we proposed a log-linear based morphological segmentation approach. Instead of learning model only from monolingual annotated corpus, this approach optimizes Uyghur segmentation for spoken translation based on both bilingual and monolingual corpus. Our approach relies on several features such as traditional conditional random field (CRF) feature, bilingual word alignment feature and monolingual suffixword co-occurrence feature. Experimental results shown that our proposed segmentation model for Uyghur spoken translation achieved 1.6 BLEU score improvements compared with the state-of-the-art baseline. | Log-linear Models for Uyghur Segmentation in Spoken Language Translation |
d4660878 | This paper explores a new TM-based CAT tool entitled CATaLog. New features have been integrated into the tool which aim to improve post-editing both in terms of performance and productivity. One of the new features of CAT-aLog is a color coding scheme that is based on the similarity between a particular input sentence and the segments retrieved from the TM. This color coding scheme will help translators to identify which part of the sentence is most likely to require post-editing thus demanding minimal effort and increasing productivity. We demonstrate the tool's functionalities using an English -Bengali dataset. | CATaLog: New Approaches to TM and Post Editing Interfaces |
d390089 | The Regesta Imperii (RI) are an important source for research in European-medieval history. Sources spread over many centuries of medieval history -mainly charters of German-Roman Emperors -are summarized as "Regests" and pooled in the RI. Interesting medieval demographic groups and players are i.a. cities, citizens or spiritual institutions (e.g. bishops or monasteries). Themes of historical interest are i.a. peace and war or the endowment of new privileges. We investigate the RI for important players and themes, applying state-of-the-art text classification methods from computational linguistics. We examine the performance of different classification methods in view of the linguistically very heterogeneous RI, including a Neural Network approach that is designed to capture complex interactions between players and themes. | Deriving Players & Themes in the Regesta Imperii using SVMs and Neural Networks |
d6023976 | We describe a supervised approach to predicting the set of all inflected forms of a lexical item. Our system automatically acquires the orthographic transformation rules of morphological paradigms from labeled examples, and then learns the contexts in which those transformations apply using a discriminative sequence model. Because our approach is completely data-driven and the model is trained on examples extracted from Wiktionary, our method can extend to new languages without change. Our end-to-end system is able to predict complete paradigms with 86.1% accuracy and individual inflected forms with 94.9% accuracy, averaged across three languages and two parts of speech. * Research conducted during an internship at Google. 1 | Supervised Learning of Complete Morphological Paradigms |
d18836940 | Based on simple methods such as observing word and part of speech tag co-occurrence and clustering, we generate syntactic parses of sentences in an entirely unsupervised and self-inducing manner. The parser learns the structure of the language in question based on measuring 'breaking points' within sentences. The learning process is divided into two phases, learning and application of learned knowledge. The basic learning works in an iterative manner which results in a hierarchical constituent representation of the sentence. Part-of-Speech tags are used to circumvent the data sparseness problem for rare words. The algorithm is applied on untagged data, on manually assigned tags and on tags produced by an unsupervised part of speech tagger. The results are unsurpassed by any self-induced parser and challenge the quality of trained parsers with respect to finding certain structures such as noun phrases. | UnsuParse: Unsupervised Parsing with unsupervised Part of Speech tagging |
d236477650 | Identifying relations from dialogues is more challenging than traditional sentence-level relation extraction (RE), since the difficulties of speaker information representation and the long-range semantic reasoning. Despite the successful efforts, existing methods do not fully consider the particularity of dialogues, making them difficult to truly understand the semantics between conversational arguments. In this paper, we propose two beneficial tasks, speaker prediction and trigger words prediction, to enhance the extraction of dialoguebased relations. Specifically, speaker prediction captures the characteristics of speakerrelated entities, and the trigger words prediction provides supportive contexts for relations between arguments. Extensive experiments on the DialogRE dataset show noticeable improvements compared to the baseline models, which achieves a new state-of-the-art performance with a 65.5% of F1 score and a 60.5% of F1 c score, respectively. | Enhancing Dialogue-based Relation Extraction by Speaker and Trigger Words Prediction |
d64758482 | Supervised approaches for text summarisation suffer from the problem of mismatch between the target labels/scores of individual sentences and the evaluation score of the final summary. Reinforcement learning can solve this problem by providing a learning mechanism that uses the score of the final summary as a guide to determine the decisions made at the time of selection of each sentence. In this paper we present a proof-of-concept approach that applies a policy-gradient algorithm to learn a stochastic policy using an undiscounted reward. The method has been applied to a policy consisting of a simple neural network and simple features. The resulting deep reinforcement learning system is able to learn a global policy and obtain encouraging results. | Towards the Use of Deep Reinforcement Learning with Global Policy For Query-based Extractive Summarisation ⇤ |
d3921598 | This paper presents a data-driven model of eye movement control in reading that builds on earlier work using machine learning methods to model saccade behavior. We extend previous work by modeling the time course of eye movements, in addition to where the eyes move. In this model, the initiation of eye movements is delayed as a function of on-line processing difficulty, and the decision of where to move the eyes is guided by past reading experience, approximated using machine learning methods. In benchmarking the model against held-out previously unseen data, we show that it can predict gaze durations and skipping probabilities with good accuracy. | Towards a Data-Driven Model of Eye Movement Control in Reading |
d1902688 | Probabilistic Parsing of Unrestricted English Text, With a Highly-Detailed Grammar | |
d53080145 | Recent progress in dialogue generation has inspired a number of studies on dialogue systems that are capable of accomplishing tasks through natural language interactions. A promising direction among these studies is the use of reinforcement learning techniques, such as self-play, for training dialogue agents. However, current datasets are limited in size, and the environment for training agents and evaluating process is relatively unsophisticated. We present AirDialogue, a large dataset that contains 402,038 goal-oriented conversations. To collect this dataset, we create a contextgenerator which provides travel and flight restrictions. We then ask human annotators to play the role of a customer or an agent and interact with the goal of successfully booking a trip given the restrictions. Key to our environment is the ease of evaluating the success of the dialogue, which is achieved by using ground-truth states (e.g., the flight being booked) generated by the restrictions. Any dialogue agent that does not generate the correct states is considered to fail. Our experimental results indicate that state-of-the-art dialogue models on the test dataset can only achieve a scaled score of 0.22 and an exact match score of 0.1 while humans can reach a score of 0.94 and 0.93 respectively, which suggests significant opportunities for future improvement. | AirDialogue: An Environment for Goal-Oriented Dialogue Research |
d8841532 | Word sense induction and discrimination (WSID) identifies the senses of an ambiguous word and assigns instances of this word to one of these senses. We have build a WSID system that exploits syntactic and semantic features based on the results of a natural language parser component. To achieve high robustness and good generalization capabilities, we designed our system to work on a restricted, but grammatically rich set of features. Based on the results of the evaluations our system provides a promising performance and robustness. | KCDC: Word Sense Induction by Using Grammatical Dependencies and Sentence Phrase Structure |
d1329333 | Most Arabic Named Entity Recognition (NER) systems have been developed using either of two approaches: a rule-based or Machine Learning (ML) based approach, with their strengths and weaknesses. In this paper, the problem of Arabic NER is tackled through integrating the two approaches together in a pipelined process to create a hybrid system with the aim of enhancing the overall performance of NER tasks. The proposed system is capable of recognizing 11 different types of named entities (NEs): Person, Location, Organization, Date, Time, Price, Measurement, Percent, Phone Number, ISBN and File Name. Extensive experiments are conducted using three different ML classifiers to evaluate the overall performance of the hybrid system. The empirical results indicate that the hybrid approach outperforms both the rule-based and the ML-based approaches. Moreover, our system outperforms the state-of-the-art of Arabic NER in terms of accuracy when applied to ANERcorp dataset, with f-measures 94.4% for Person, 90.1% for Location, and 88.2% for Organization. Abstract in Arabic لة، ا تعلم على المبنية المنھجية تبني أو القواعد منھجية تبني ل خ من العربية سماء ا أنماط على التعرف أنظمة معظم بناء تم دمج ل خ من معالجتھا يتم العربية اللغة في سماء ا أنماط على التعرف عملية الورقة، ھذه في وضعف. قوة نقاط من فيھما بما ًا مع المنھجيتين ٍ تنسيق في لتشكيل ٍ متتال المنھج ُقترح الم النظام سماء. ا أنماط على التعرف مھام أداء لتحسين محاولة في الھجين على التعرف على قادر 11 والتواريخ، والمنظمات، ماكن، وا شخاص، ا أسماء ذلك في بما سماء ا أنماط من ا ً مختلف ًا نوع والنسب القياسية(، )المقادير والمقاييس موال(، )ا سعار وا وقات، وا الدولي )الرقم وردمك الھواتف، وأرقام المئوية، تعلم ّق َب ُط ت مختلفة مصنفات ث ث باستخدام مكثفة تجارب إجراء تم وقد الملفات. وأسماء للكتاب(، المعياري لتقييم لة ا أداء القوا على المبني المنھج من ٍ كل على الھجين المنھج تفوق التجريبية النتائج ُظھر ت الھجين. النظام تعلم على المبني والمنھج عد من العربية سماء ا أنماط على التعرف مجال في العلمية الدوريات في المنشورة نظمة ا أفضل على الھجين نظامنا يتفوق لة. ا قدرھا: توافقية ت معد بنتيجة "أنيركورب" البيانات مجموعة على نظامنا تطبيق عند الدقة حيث 94.4 أسماء حالة في % شخ ا اص، 90.1 و ماكن، ا أسماء حالة في % 88.2 المنظمات. أسماء حالة في % KEYWORDS in Arabic لة ا تعلم سماء، ا أنماط على التعرف الطبيعية، اللغات معالجة | A Pipeline Arabic Named Entity Recognition Using a Hybrid Approach |
d14033047 | In this paper, we present and evaluate an approach for the compositional alignment of compound nouns using comparable corpora from technical domains. The task of term alignment consists in relating a source language term to its translation in a list of target language terms with the help of a bilingual dictionary. Compound splitting allows to transform a compound into a sequence of components which can be translated separately and then related to multi-word target language terms. We present and evaluate a method for compound splitting, and compare two strategies for term alignment (bag-of-word vs. pattern-based). The simple word-based approach leads to a considerable amount of erroneous alignments, whereas the pattern-based approach reaches a decent precision. We also assess the reasons for alignment failures: in the comparable corpora used for our experiments, a substantial number of terms has no translation in the target language data; furthermore, the non-isomorphic structures of source and target language terms cause alignment failures in many cases. | Analyzing and Aligning German Compound Nouns |
d6793318 | In 2004 a new, comprehensive Dutch-Flemish HLT programme will be launched by a number of ministries and organizations in the Netherlands and Flanders. To guarantee its Dutch-Flemish character, this large-scale programme will be carried out under the auspices of the Dutch Language Union (NTU). The aim of this new initiative, which is a continuation of the previous HLT Platform project, is to contribute to the further progress of HLT for the Dutch language. In trying to achieve this goal the project partners will make a concerted effort aimed at raising awareness of HLT results, , promoting innovation oriented strategic research in HLT, stimulating the demand of HLT products, developing HLT resources that are essential and are known to be missing, organising the management, maintenance and distribution of HLT resources. | The new Dutch-Flemish HLT Programme: a concerted effort to stimulate the HLT sector |
d9587160 | We study the problem of ontology population for a domain ontology and present solutions based on semi-automatic techniques. A domain ontology for an organization, often consists of classes whose instances are either specific to, or independent of the organization. E.g. in an academic domain ontology, classes like Professor, Department could be organization (university) specific, while Conference, Programming languages are organization independent. This distinction allows us to leverage data sources both -within the organization and those in the Internet -to extract entities and populate an ontology. We propose techniques that build on those for open domain IE. Together with user input, we show through comprehensive evaluation, how these semi-automatic techniques achieve high precision. We experimented with the academic domain and built an ontology comprising of over 220 classes. Intranet documents from five universities formed our organization specific corpora and we used open domain knowledge bases like Wikipedia, Linked Open Data, and web pages from the Internet as the organization independent data sources. The populated ontology that we built for one of the universities comprised of over 75,000 instances. We adhere to the semantic web standards and tools and make the resources available in the OWL format. These could be useful for applications such as information extraction, text annotation, and information retrieval. | Efficient reuse of structured and unstructured resources for ontology population |
d7314969 | The task of machine translation (MT) evaluation is closely related to the task of sentence-level semantic equivalence classification. This paper investigates the utility of applying standard MT evaluation methods (BLEU, NIST, WER and PER) to building classifiers to predict semantic equivalence and entailment. We also introduce a novel classification method based on PER which leverages part of speech information of the words contributing to the word matches and non-matches in the sentence. Our results show that MT evaluation techniques are able to produce useful features for paraphrase classification and to a lesser extent entailment. Our technique gives a substantial improvement in paraphrase classification accuracy over all of the other models used in the experiments. | Using Machine Translation Evaluation Techniques to Determine Sentence-level Semantic Equivalence |
d671799 | This paper suggests the efficient indexing method based on a concept vector space that is capable of representing the semantic content of a document. The two information measure, namely the information quantity and the information ratio, are defined to represent the degree of the semantic importance within a document. The proposed method is expected to compensate the limitations of term frequency based methods by exploiting related lexical items. Furthermore, with information ratio, this approach is independent of document length. | A Novel Approach to Semantic Indexing Based on Concept |
d208325499 | Defined as the intentional or unintentional spread of false information (K et al., 2019) through context and/or content manipulation, fake news has become one of the most serious problems associated with online information(Waldrop, 2017). Consequently, it comes as no surprise that Fake News Detection has become one of the major foci of various fields of machine learning and while machine learning models have allowed individuals and companies to automate decision-based processes that were once thought to be only doable by humans, it is no secret that the real-life applications of such models are not viable without the existence of an adequate training dataset. In this paper we describe the Veritas Annotator, a web application for manually identifying the origin of a rumour. These rumours, often referred as claims, were previously checked for validity by Fact-Checking Agencies. | Veritas Annotator: Discovering the Origin of a Rumour |
d1586259 | We present an approach to grammar development where the task is decomposed into two separate subtasks. The first task is hnguistic, with the goal of producing a set of rules that have a large coverage (in the sense that the correct parse is among the proposed parses) on a bhnd test set of sentences. The second task is statistical, with the goal of developing a model of the grammar which assigns maximum probability for the correct parse. We give parsing results on text from computer manuals. | Development and Evaluation of a Broad-Coverage Probabilistic Grammar of English-Language Computer Manuals |
d220064250 | AUTHOR INDEX | |
d221373821 | Nous présentons ici les résultats d'un travail de réplication et d'extension pour l'alsacien d'une expérience concernant l'étiquetage en parties du discours de langues peu dotées par spécialisation des plongements lexicaux(Magistry et al., 2018). Ce travail a été réalisé en étroite collaboration avec les auteurs de l'article d'origine. Cette interaction riche nous a permis de mettre au jour les éléments manquants dans la présentation de l'expérience, de les compléter, et d'étendre la recherche à la robustesse à la variation.ABSTRACTReplicating and extending for Alsatian : "POS tagging for low-resource languages by adapting word embeddings"We present here the results of our efforts in replicating and extending for Alsatian an experiment concerning the POS tagging of low-resourced languages by adapting word embeddings(Magistry et al., 2018). This work was performed in close collaboration with the authors of the original article. This rich interaction allowed us to identify the missing elements in the presentation of the experiment, to add them and to extend the experiment to robustness to variation. MOTS-CLÉS : réplicabilité, étiquetage en parties du discours, langues peu dotées, variation.Les corpus utilisés et le tagger entraîné ont été développés dans le cadre du projet RESTAURE porté Actes de la 6e conférence conjointe Journées d'Études sur la Parole (JEP, 33e édition), Traitement Automatique des Langues Naturelles (TALN, 27e édition), Rencontre des Étudiants Chercheurs en Informatique pour le Traitement Automatique des Langues (RÉCITAL, 22e édition) Nancy, France, 08-19 juin 2020 2e atelier Éthique et TRaitemeNt Automatique des Langues (ETeRNAL), pages 29-37. hal : hal-02750224.Cette oeuvre est mise à disposition sous licence Attribution 4.0 International. | Répliquer et étendre pour l'alsacien « Étiquetage en parties du discours de langues peu dotées par spécialisation des plongements lexicaux » |
d16451235 | CONSTRUCTIBLE REPRESENTATIONS FOR TWO SEMANTIC RELATIONS | |
d54458698 | Text-based adventure games provide a platform on which to explore reinforcement learning in the context of a combinatorial action space, such as natural language. We present a deep reinforcement learning architecture that represents the game state as a knowledge graph which is learned during exploration. This graph is used to prune the action space, enabling more efficient exploration. The question of which action to take can be reduced to a question-answering task, a form of transfer learning that pre-trains certain parts of our architecture. In experiments using the TextWorld framework, we show that our proposed technique can learn a control policy faster than baseline alternatives. We have also open-sourced our code at https://github.com/rajammanabrolu/KG-DQN. | Playing Text-Adventure Games with Graph-Based Deep Reinforcement Learning |
d16883813 | This paper shows that the semantics of shenme 'what' exhibits a double quantification phenomenon in Chinese bare conditionals. I show that such double quantification can be nicely accounted for if one adopts Carlosn's semantics of bare plurals and verb meanings as well as the assumption that shenme 'what' may denote kinds of things as bare plurals do. | On the Meaning of Shenme 'what' in Chinese Bare Conditionals and its Implications for Carlson's Semantics of Bare Plurals |
d412164 | We present in this paper methods to improve HMM-based part-of-speech (POS) tagging of Mandarin. We model the emission probability of an unknown word using all the characters in the word, and enrich the standard left-to-right trigram estimation of word emission probabilities with a right-to-left prediction of the word by making use of the current and next tags. In addition, we utilize the RankBoost-based reranking algorithm to rerank the N-best outputs of the HMMbased tagger using various n-gram, morphological, and dependency features. Two methods are proposed to improve the generalization performance of the reranking algorithm. Our reranking model achieves an accuracy of 94.68% using n-gram and morphological features on the Penn Chinese Treebank 5.2, and is able to further improve the accuracy to 95.11% with the addition of dependency features. | Mandarin Part-of-Speech Tagging and Discriminative Reranking |
d17763313 | Keyword-matching systems based on simple models of semantic relatedness are inadequate at modelling the ambiguities in natural language text, and cannot reliably address the increasingly complex information needs of users. In this paper we propose novel methods for computing semantic relatedness by spreading activation energy over the hyperlink structure of Wikipedia. We demonstrate that our techniques can approach state-of-the-art performance, while requiring only a fraction of the background data. | Measuring Conceptual Similarity by Spreading Activation over Wikipedia's Hyperlink Structure |
d11826603 | Morph~ is a Common Lisp compiler for reversible inflectional morphology rules developed at the Center for Machine Translation at Carnegie Mellon University. This paper describes the Morph~ processing model, its implementation, and how it handles some common morphological processes. | MORPHIa: A Practical Compiler for Reversible Morphology Rules |
d6181440 | On a multi-dimensional text categorization task, we compare the effectiveness of a feature based approach with the use of a stateof-the-art sequential learning technique that has proven successful for tasks such as "email act classification". Our evaluation demonstrates for the three separate dimensions of a well established annotation scheme that novel thread based features have a greater and more consistent impact on classification performance. | A Feature Based Approach to Leveraging Context for Classifying Newsgroup Style Discussion Segments |
d7236507 | We investigate a combination of a traditional linear sparse feature model and a multi-layer neural network model for deterministic transition-based dependency parsing, by integrating the sparse features into the neural model. Correlations are drawn between the hybrid model and previous work on integrating word embedding features into a discrete linear model. By analyzing the results of various parsers on web-domain parsing, we show that the integrated model is a better way to combine traditional and embedding features compared with previous methods. | Combining Discrete and Continuous Features for Deterministic Transition-based Dependency Parsing |
d233429873 | In this paper, we tackle the task of Word Sense Disambiguation (WSD). We present our system submitted to the Word-in-Context Target Sense Verification challenge, part of the SemDeep workshop at IJCAI 2020(Breit et al., 2020). That challenge asks participants to predict if a specific mention of a word in a text matches a pre-defined sense. Our approach uses pre-trained transformer models such as BERT that are fine-tuned on the task using different architecture strategies. Our model achieves the best accuracy and precision on Subtask 1 -make use of definitions for deciding whether the target word in context corresponds to the given sense or not. We believe the strategies we explored in the context of this challenge can be useful to other Natural Language Processing tasks. | Word Sense Disambiguation with Transformer Models |
d231821642 | Algorithmic-based decision making powered via AI and (big) data has already penetrated into almost all spheres of human life, from content recommendation and healthcare to predictive policing and autonomous driving, deeply affecting everyone, anywhere, anytime. While technology allows previously unthinkable optimizations in the automation of expensive human decision making, the risks that the technology can pose are also high, leading to an ever increasing public concern about the impact of the technology in our lives. The area of responsible AI has recently emerged in an attempt to put humans at the center of AI-based systems by considering aspects, such as fairness, reliability and privacy of decision-making systems. | Invited talk Bias in AI-systems: A multi-step approach |
d252763289 | This paper is aiming at the problems of lack of annotated data and inadequate sentiment semantic representation in existing domain sentiment lexicon construction methods. In this paper, the joint weight is calculated by multi-source data. Combining prior emotional knowledge and Fasttext word vector representation learning, the sentiment semantic knowledge is mapped to a new word vector space, and the domain sentiment dictionary is automatically constructed from unlabeled data to adapt to the multi-domain and multi-language environment. The comparative experiments on Chinese and English multi-domain public data sets show that, compared with sentiment dictionary and pretrained language model, the proposed multi-source knowledge fusion method of domain sentiment dictionary representation learning has significantly improved the classification accuracy on public data sets, and has good robustness on various algorithms, multi-language, multi-domain and multi-data sets. This paper also verifies the role of each module of the proposed model in improving the effect of sentiment classification through ablation experiments. | Domain Sentiment Lexicon Representation Learning Based on Multi-source Knowledge Fusion |
d196174722 | ||
d10838059 | This paper describes recent work aimed at relating multi-level dialog annotations with meta-data annotations for a corpus of real humanhuman dialogs. This work is carried out in the context of the AMITIES project in which spoken dialog systems for call center services are being developed. A corpus of 100 agent-client dialogs have been annotated with three types of annotations. The first are utterance-level DAMSL-style dialogic labels. The second set of annotations applies to exchanges and takes into account of the dynamic aspect of dialog progress. Finally, 5 emotions types are annotated at the utterance level. Some of these multi-style annotations were used in a multiple linear regression analysis to predict dialog quality. The predictive factors are able to explain about 80% of the dialog accidents. | Annotations for Dynamic Diagnosis of the Dialog State |
d221097289 | ||
d227062345 | This paper presents the study of sentiment analysis for Amharic social media texts. As the number of social media users is ever-increasing, social media platforms would like to understand the latent meaning and sentiments of a text to enhance decision-making procedures. However, lowresource languages such as Amharic have received less attention due to several reasons such as lack of well-annotated datasets, unavailability of computing resources, and fewer or no expert researchers in the area. This research addresses three main research questions. We first explore the suitability of existing tools for the sentiment analysis task. Annotation tools are scarce to support large-scale annotation tasks in Amharic. Also, the existing crowdsourcing platforms do not support Amharic text annotation. Hence, we build a social-network-friendly annotation tool called 'ASAB' using the Telegram bot. We collect 9.4k tweets, where each tweet is annotated by three Telegram users. Moreover, we explore the suitability of machine learning approaches for Amharic sentiment analysis. The FLAIR deep learning text classifier, based on network embeddings that are computed from a distributional thesaurus, outperforms other supervised classifiers. We further investigate the challenges in building a sentiment analysis system for Amharic and we found that the widespread usage of sarcasm and figurative speech are the main issues in dealing with the problem. To advance the sentiment analysis research in Amharic and other related lowresource languages, we release the dataset, the annotation tool, source code, and models publicly under a permissive. This work is licensed under a Creative Commons Attribution 4.0 International License. License details: http:// creativecommons.org/licenses/by/4.0/. | Exploring Amharic Sentiment Analysis from Social Media Texts: Building Annotation Tools and Classification Models |
d12160209 | We carried out a study on monolingual translators with no knowledge of the source language, but aided by post-editing and the display of translation options. On Arabic-English and Chinese-English, using standard test data and current statistical machine translation systems, 10 monolingual translators were able to translate 35% of Arabic and 28% of Chinese sentences correctly on average, with some of the participants coming close to professional bilingual performance on some of the documents. | Enabling Monolingual Translators: Post-Editing vs. Options |
d146621 | The critical issues involved in speech-to-speech translation are obtaining proper source segments and synthesizing accurate target speech. Therefore, this article develops a novel multiple-translation spotting method to deal with these issues efficiently. Term multiple-translation spotting refers to the task of extracting target-language synthesis patterns that correspond to a given set of source-language spotted patterns in conditional multiple pairs of speech patterns known to be translation patterns. According to the extracted synthesis patterns, the target speech can be properly synthesized by using a waveform segment concatenation-based synthesis method. Experiments were conducted with the languages of Mandarin and Taiwanese. The results reveal that the proposed approach can achieve translation understanding rates of 80% and 76% on average for Mandarin/Taiwanese translation and Taiwanese/Mandarin translation, respectively. | Multiple-Translation Spotting for Mandarin-Taiwanese Speech-to-Speech Translation |
d226284011 | In recent years, large pre-trained models have demonstrated state-of-the-art performance in many NLP tasks. However, the deployment of these models on devices with limited resources is challenging due to the models' large computational consumption and memory requirements. Moreover, the need for a considerable amount of labeled training data also hinders real-world deployment scenarios. Model distillation has shown promising results for reducing model size, computational load and data efficiency. In this paper we test the boundaries of BERT model distillation in terms of model compression, inference efficiency and data scarcity. We show that classification tasks that require the capturing of general lexical semantics can be successfully distilled by very simple and efficient models and require relatively small amount of labeled training data. We also show that the distillation of large pretrained models is more effective in real-life scenarios where limited amounts of labeled training are available. | Exploring the Boundaries of Low-Resource BERT Distillation |
d226283780 | We present preliminary results on investigating the benefits of coreference resolution features for neural RST discourse parsing by considering different levels of coupling of the discourse parser with the coreference resolver. In particular, starting with a strong baseline neural parser unaware of any coreference information, we compare a parser which utilizes only the output of a neural coreference resolver, with a more sophisticated model, where discourse parsing and coreference resolution are jointly learned in a neural multitask fashion. Results indicate that these initial attempts to incorporate coreference information do not boost the performance of discourse parsing in a statistically significant way. | Coreference for Discourse Parsing: A Neural Approach |
d1222779 | INTRODUCTIONWhereas in the United States work in machine translation (MT) has only recently been reinstated as a 'respectable' natural language processing (NLP) application, it has long been considered a worthwhile and interesting topic for research and development in both Europe and Japan. In terms of number of projects in one sub-field of computational linguistics, MT is currently perhaps the most important application. 1 One obvious reason for this is simply the daily awareness that people communicate in languages other than English, a situation that naturally encourages an interest in translation. On a practical level, for example, every television cable system in Europe broadcasts stations from numerous countries, and on the political level, the European Community (EC) is committed to protecting the language of each of the Member States, which implies providing numerous translation services. From an economic viewpoint, every company knows that in order to market its products, the documentation must be in the language of the target country. And a last motivation for interest in MT, which was also the origin of MT activities in the US and an important concern for Japan, is the desire for better access to information--important documents often exist in some foreign language.Yet MT in Europe is not viewed as just a matter of developing working MT systems for commercial and practical needs--it is also accepted as a legitimate topic of research. The view of MT as a test bed for NLP work has long been defended in the United States (Kay,1980). Reasons why this position has only recently gained favor can be attributed to Bar-Hillel's strong view on the impossibility of high-quality MT coupled with the far-reaching effects the ALPAC report (1966) had on funding in the US. All direct funding for translation was withdrawn and redirected to more basic research and thus linguistics and AI work prospered. Though practical work continued, as well as a real need for translation, 2 MT fell into disrepute as an academically respectable enterprise. While there is consensus that fully automatic high quality MT of unrestricted text is impossible, it is nevertheless an attractive long-term goal, similar to pursuits in artificial intelligence. In Europe, a growing number of researchers in computational linguistics regard translation as a challenging field of application. Eased on developments in the field such as a more rigorous formalization of semantics (e.g., MontagueAt the two most recent Coling conferences, for example, the number of papers devoted to issues in MT constituted the largest single topic; and this figure does not take into account all the general NLP papers presented by the MT projects. Ironically, the Georgetown system, on which the ALPAC report was based, continued to be used in Europe, until well into the 70s and Systran, a direct descendant, is still the most widely used commercial MT system. grammar), the attention paid to formal and computational properties of linguistic theories (e.g., LFG and GPSG) and the definition and implementation of linguistically problemoriented computational methods (e.g., unification), it is quite natural that attempts are being made to test their adequacy with regard to problems of translation.The multilingual setting of Europe, where translation is a fact of life, along with its varied and decentralized funding agencies (including EC, national and regional programs), as opposed to the more centralized nature of US federal agencies, helps explain why the ALPAC report had less of an impact overseas. Machine translation has a long and relatively stable tradition in Europe. Similar to the early work with computers and language in the United States where CL and MT were synonymous, MT projects in Europe have served as a vehicle for developing expertise in computational linguistics in centers which had little experience in the field. This latter point is particularly true in the Eurotra project; Greece, for example, had no tradition in computational linguistics.The historical and socio-political references have been introduced as background material, given the rather strong positions taken up by members in and out of the community over the last decades. The distinction between research and development or theoretical vs. practical, though somewhat artificial (and definitely a touchy issue in the community), serves as a means of clarifying and motivating what people are working on and why. The extreme view repeatedly put forward by M. Kay that "all machine translation of natural languages is experimental ..." (Kay, 1984:75) is, in my view, correct. There are nevertheless things that we can accomplish, albeit imperfectly, and from which we can learn both about language and about translafion--a situation similar to all NLP work. My purpose here is to distinguish major topics currently popular in MT work and to identify the projects and centers active in the field. | MACHINE TRANSLATION IN EUROPE |
d18614158 | Multi-sense word embeddings (MSEs) model different meanings of word forms with different vectors. We propose two new methods for evaluating MSEs, one based on monolingual dictionaries, and the other exploiting the principle that words may be ambiguous as far as the postulated senses translate to different words in some other language. | Evaluating multi-sense embeddings for semantic resolution monolingually and in word translation |
d1577616 | A useful enhancement of an NLG system for verbalising ontologies would be a module capable of explaining undesired entailments of the axioms encoded by the developer. This task raises interesting issues of content planning. One approach, useful as a baseline, is simply to list the subset of axioms relevant to inferring the entailment; however, in many cases it will still not be obvious, even to OWL experts, why the entailment follows. We suggest an approach in which further statements are added in order to construct a proof tree, with every step based on a relatively simple deduction rule of known difficulty; we also describe an empirical study through which the difficulty of these simple deduction patterns has been measured. | Planning Accessible Explanations for Entailments in OWL Ontologies |
d258486929 | Pre-trained Transformer-based models have become immensely popular amongst NLP practitioners. We present TrelBERT -the first Polish language model suited for application in the social media domain. TrelBERT is based on an existing general-domain model and adapted to the language of social media by pre-training it further on a large collection of Twitter data. We demonstrate its usefulness by evaluating it in the downstream task of cyberbullying detection, in which it achieves state-of-the-art results, outperforming larger monolingual models trained on general-domain corpora, as well as multilingual in-domain models, by a large margin. We make the model publicly available. We also release a new dataset for the problem of harmful speech detection. | TrelBERT: A pre-trained encoder for Polish Twitter |
d258486952 | A Simplified Training Pipeline for Low-Resource and Unsupervised Machine Translation | |
d51871763 | Turning NMT research into commercial products Dragos Munteanu and Adrià de Gispert | |
d115850728 | CESTA, the first European Campaign dedicated to MT Evaluation, is a project labelled by the French Technolangue action. CESTA provides an evaluation of six commercial and academic MT systems using a protocol set by an international panel of experts. CESTA aims at producing reusable resources and information about reliability of the metrics. Two runs will be carried out: one using the system's basic dictionary, another after terminological adaptation. Evaluation task, test material, resources, evaluation measures, metrics, will be detailed in the full paper. The protocol is the combination of a contrastive reference to: IBM "BLEU" protocol (Papineni, K., S. Roukos, T. Ward | Work-in-Progress project report : CESTA -Machine Translation Evaluation Campaign |
d2279676 | We examine the problem of choosing word order for a set of dependency trees so as to minimize total dependency length. We present an algorithm for computing the optimal layout of a single tree as well as a numerical method for optimizing a grammar of orderings over a set of dependency types. A grammar generated by minimizing dependency length in unordered trees from the Penn Treebank is found to agree surprisingly well with English word order, suggesting that dependency length minimization has influenced the evolution of English. | Optimizing Grammars for Minimum Dependency Length |
d26746915 | In the history of Artificial Intelligence (AI), Turing Test, a question answering imitation game was proposed to determine whether the computer system has intelligence. It becomes the ultimate goal to answer all the natural language questions for generations of AI researchers. In the past century, AI changed tremendously from its theories to its applications, while with this goal unchanged. Especially in the past 20 years, along with the development of the Internet, computers have the ability to acquire, store and process huge volumes of data, which makes the AI-related techniques deeply involve themselves in the domain of intelligent information processing. On one hand, Question Answering develops in theories, models and methods with the combination of the large scale data processing. On the other hand, the next-generation information service engines are expected to integrate Question Answering as an important part to retrieve and display information, where knowledge is important for information accumulation, understanding and serving. This presentation will present the history and development of the Question Answering, its related key technologies and applications in the background of big data and AI. | QA: from Turing Test to Intelligent Information Service About the Speaker |
d202787258 | In this paper we present a novel approach to simultaneously representing multiple languages in a common space. Procrustes Analysis (PA) is commonly used to find the optimal orthogonal word mapping in the bilingual case. The proposed Multi Pairwise Procrustes Analysis (MPPA) is a natural extension of the PA algorithm to multilingual word mapping. Unlike previous PA extensions that require a k-way dictionary, this approach requires only pairwise bilingual dictionaries that are much easier to construct in either a supervised or an unsupervised way. The improved performance of the MPPA algorithm is demonstrated on two standard multilingual tasks. | A Multi-Pairwise Extension of Procrustes Analysis for Multilingual Word Translation |
d16497823 | Statistical measures of word similarity have application in many areas of natural language processing, such as language modeling and information retrieval. We report a comparative study of two methods for estimating word cooccurrence frequencies required by word similarity measures. Our frequency estimates are generated from a terabyte-sized corpus of Web data, and we study the impact of corpus size on the effectiveness of the measures. We base the evaluation on one TOEFL question set and two practice questions sets, each consisting of a number of multiple choice questions seeking the best synonym for a given target word. For two question sets, a context for the target word is provided, and we examine a number of word similarity measures that exploit this context. Our best combination of similarity measure and frequency estimation method answers 6-8% more questions than the best results previously reported for the same question sets. | Frequency Estimates for Statistical Word Similarity Measures |
d218973735 | ||
d219302901 | We investigate the differences between language models compiled from original target-language texts and those compiled from texts manually translated to the target language. Corroborating established observations of Translation Studies, we demonstrate that the latter are significantly better predictors of translated sentences than the former, and hence fit the reference set better. Furthermore, translated texts yield better language models for statistical machine translation than original texts. | Language Models for Machine Translation: Original vs. Translated Texts |
d218610661 | Pre-trained visually grounded language models such as ViLBERT, LXMERT, and UNITER have achieved significant performance improvement on vision-and-language tasks but what they learn during pre-training remains unclear. In this work, we demonstrate that certain attention heads of a visually grounded language model actively ground elements of language to image regions. Specifically, some heads can map entities to image regions, performing the task known as entity grounding. Some heads can even detect the syntactic relations between non-entity words and image regions, tracking, for example, associations between verbs and regions corresponding to their arguments. We denote this ability as syntactic grounding. We verify grounding both quantitatively and qualitatively, using Flickr30K Entities as a testbed. . 2015. VQA: Visual question answering. In ICCV.Xinlei Chen, Hao Fang, Tsung-Yi Lin, Ramakrishna Vedantam, Saurabh Gupta, Piotr Dollár, and C Lawrence Zitnick. 2015. Microsoft COCO captions: Data collection and evaluation server. arXiv preprint arXiv:1504.00325. | What Does BERT with Vision Look At? |
d27663067 | This paper will describe a way to organize the salient objects, their attributes, and relationships between the objects in a given domain. This organization allows us to assign an information value to each collection, and to the domain as a whole, which corresponds to the number of things to "talk about" in the domain. This number gives a measure of semantic complexity; that is, it will correspond to the number of objects, attributes, and relationships in the domain, but not to the level of syntactic diversity allowed when conveying these meanings.Defining a measure of semantic complexity for a dialog system domain will give an insight towards making a complexity measurement standard. With such a standard, natural language programmers can measure the feasibility of making a natural language interface, compare different language processors' ability to handle more and more complex domains, and quantify the abilities of the current state of the art in natural language processors. | A Measure of Semantic Complexity for Natural Language Systems |
d15616683 | The paper describes correction modules for AutoPat, -an authoring system for patent claims. The specificity of AutoPat is that it relies on human-computer content specification in controlled language. The quality of the textual input is a crucial point in getting a high quality AutoPat output. Our correction modules handle both the quality of human input and the final system output. The human input is passed through a spellchecker, a grammar checker and a content checker. An application-tuned grammar checker is run over the system final output of the generation module. | Integration of Correction Modules in a Controlled Language Application |
d51876481 | Building systems able to provide a semantic representation of texts has long been an objective, both in linguistics and in applied NLP. Although advances in machine learning sometimes seem to diminish the need to use as input sophisticated structured representations of sentences, the enthusiasm for interpreting trained neural networks somewhat seems to reaffirm that need.Because they represent schematic situations, semantic frames (Fillmore, 1982), as intantiated into FrameNet (Baker, Fillmore and Petruck, 1983) are an appealing level of generalization over the eventualities described in texts.In this talk, I will present some feedback from the development of a French FrameNet, including analysis of the main difficulties we faced during annotation. I will describe how linking generalizations can be extracted from the frame-annotated data, using deep syntactic annotations. I will then investigate what kind of input is most effective for FrameNet parsing, from no syntax at all to deep syntactic representations. | Annotating and parsing to semantic frames: feedback from the French FrameNet project Invited talk |
d18047795 | ||
d31298398 | Building a persona-based conversation agent is challenging owing to the lack of large amounts of speaker-specific conversation data for model training. This paper addresses the problem by proposing a multi-task learning approach to training neural conversation models that leverages both conversation data across speakers and other types of data pertaining to the speaker and speaker roles to be modeled. Experiments show that our approach leads to significant improvements over baseline model quality, generating responses that capture more precisely speakers' traits and speaking styles. The model offers the benefits of being algorithmically simple and easy to implement, and not relying on large quantities of data representing specific individual speakers. * This work was performed at Microsoft.User input: I am getting a loop back to login page. Baseline model: Ah, ok. Thanks for the info. Our model: I'm sorry to hear that. Have you tried clearing your cache and cookies? | Multi-Task Learning for Speaker-Role Adaptation in Neural Conversation Models |
d64726655 | This paper presents a method to compute similarity of folktales based on conceptual overlap at various levels of abstraction as defined in Dutch WordNet. The method is applied on a corpus of Dutch folktales and evaluated using a comparison to traditional folktale similarity analysis based on the Aarne-Thompson-Uther (ATU) classification system. Document similarity computed by the presented method is in agreement with traditional analysis for a certain amount of folktale pairs, but differs for other pairs. However, it can be argued that the current approach computes an alternative, data-driven type of similarity. Using WordNet instead of a domainspecific ontology or classification system ensures applicability of the method outside of the folktale domain. | Folktale similarity based on ontological abstraction |
d258378141 | In simultaneous translation, the retranslation approach has the advantage of requiring no modifications to the inference engine. However, in order to reduce the undesirable flicker in the output, previous work has resorted to increasing the latency through masking, and introducing specialised inference, thus losing the simplicity of the approach. In this work, we show that self-training improves the flickerlatency tradeoff, while maintaining similar translation quality to the original. Our analysis indicates that self-training reduces flicker by controlling monotonicity. Furthermore, selftraining can be combined with biased beam search to further improve the flicker-latency tradeoff. | Self-training Reduces Flicker in Retranslation-based Simultaneous Translation |
d14933987 | The Computing Research laboratory (CRL) is developing a machine translation toolkit that allows rapid deployment of translation capabilities. This toolkit has been used to develop several machine translation systems, including a Persian-English and a Turkish-English system, which will be demonstrated. We present the architecture of these systems as well as the development methodology. | Rapid Development of Translation Tools: Application to Persian and Turkish |
d5734707 | In this paper, we describe a new, publicly available corpus intended to stimulate research into language modeling techniques which are sensitive to overall sentence coherence. The task uses the Scholastic Aptitude Test's sentence completion format. The test set consists of 1040 sentences, each of which is missing a content word. The goal is to select the correct replacement from amongst five alternates. In general, all of the options are syntactically valid, and reasonable with respect to local N-gram statistics. The set was generated by using an N-gram language model to generate a long list of likely words, given the immediate context. These options were then hand-groomed, to identify four decoys which are globally incoherent, yet syntactically correct. To ensure the right to public distribution, all the data is derived from out-of-copyright materials from Project Gutenberg. The test sentences were derived from five of Conan Doyle's Sherlock Holmes novels, and we provide a large set of Nineteenth and early Twentieth Century texts as training material. | A Challenge Set for Advancing Language Modeling |
d18341076 | Nowadays we are facing a growing demand for semantic knowledge in computational applications, particularly in Natural Language Processing (NLP). However, there aren't sufficient human resources to produce that knowledge at the same rate of its demand. Considering the Portuguese language, which has few resources in the semantic area, the situation is even more alarming. Aiming to solve that problem, this work investigates how some semantic relations can be automatically extracted from Portuguese texts. The two main approaches investigated here are based on (i) textual patterns and (ii) machine learning algorithms. Thus, this work investigates how and to which extent these two approaches can be applied to the automatic extraction of seven binary semantic relations (is-a, part-of, location-of, effect-of, property-of, made-of and used-for) in Portuguese texts. The results indicate that machine learning, in particular Support Vector Machines, is a promising technique for the task, although textual patterns presented better results for the used-for relation. | Automatic semantic relation extraction from Portuguese texts |
d250150639 | Social media are heavily used by many users to share their mental health concerns and diagnoses. This trend has turned social media into a large-scale resource for researchers focused on detecting mental health conditions. Social media usage varies considerably across individuals. Thus, classification of patterns, including detecting signs of depression, must account for such variation. We address the disparity in classification effectiveness for users with little activity (e.g., new users). Our evaluation, performed on a large-scale dataset, shows considerable detection discrepancy based on user posting frequency. For instance, the F1 detection score of users with an above-median versus below-median number of posts is greater than double (0.803 vs 0.365) using a conventional CNN-based model; similar results were observed on lexical and transformer-based classifiers. To complement this evaluation, we propose a dynamic thresholding technique that adjusts the classifier's sensitivity as a function of the number of posts a user has. This technique alone reduces the margin between users with many and few posts, on average, by 45% across all methods and increases overall performance, on average, by 33%. These findings emphasize the importance of evaluating and tuning natural language systems for potentially vulnerable populations. | TBD3: Thresholding-Based Dynamic Depression Detection from Social Media for Low-Resource Users |
d253762082 | Many NLP main tasks benefit from an accurate understanding of temporal expressions, e.g., text summarization, question answering, and information retrieval. This paper introduces Hengam, an adversarially trained transformer for Persian temporal tagging outperforming state-of-the-art approaches on a diverse and manually created dataset. We create Hengam in the following concrete steps: (1) we develop HengamTagger, an extensible rule-based tool that can extract temporal expressions from a set of diverse language-specific patterns for any language of interest. (2) We apply HengamTagger to annotate temporal tags in a large and diverse Persian text collection (covering both formal and informal contexts) to be used as weakly labeled data. (3) We introduce an adversarially trained transformer model on HengamCorpus that can generalize over the HengamTagger's rules. We create HengamGold, the first highquality gold standard for Persian temporal tagging. Our trained adversarial HengamTransformer not only achieves the best performance in terms of the F1-score (a type F1-Score of 95.42 and a partial F1-Score of 91.60) but also successfully deals with language ambiguities and incorrect spellings. Our code, data, and models are publicly available at https:// github.com/kargaranamir/Hengam. | Hengam: An Adversarially Trained Transformer for Persian Temporal Tagging |
d7012538 | The ability to make progress in Computational Linguistics depends on the availability of large annotated corpora, but creating such corpora by hand annotation is very expensive and time consuming; in practice, it is unfeasible to think of annotating more than one million words. However, the success of Wikipedia, the ESP game, and other projects shows that another approach might be possible: collaborative resource creation through the voluntary participation of thousands of Web users. ANAWIKI is a recently started project that will develop tools to allow and encourage large numbers of volunteers over the Web to collaborate in the creation of annotated corpora (in the first instance, of a corpus annotated with semantic information about anaphoric relations) through a variety of interfaces. | ANAWIKI: Creating Anaphorically Annotated Resources Through Web Cooperation |
d198232757 | This paper describes a dataset and baseline systems for linking paragraphs from court cases to clauses or amendments in the US Constitution. We implement a rule-based system, a linear model, and a neural architecture for matching pairs of paragraphs, taking training data from online databases in a distantly-supervised fashion. In experiments on a manually-annotated evaluation set, we find that our proposed neural system outperforms a rules-driven baseline. Qualitatively, this performance gap seems largest for abstract or indirect links between documents, which suggests that our system might be useful for answering political science and legal research questions or discovering novel links. We release the dataset along with the manuallyannotated evaluation set to foster future work. | Legal Linking: Citation Resolution and Suggestion in Constitutional Law |
d17503845 | Recent work has shown success in learning word embeddings with neural network language models (NNLM). However, the majority of previous NNLMs represent each word with a single embedding, which fails to capture polysemy. In this paper, we address this problem by representing words with multiple and sense-specific embeddings, which are learned from bilingual parallel data. We evaluate our embeddings using the word similarity measurement and show that our approach is significantly better in capturing the sense-level word similarities. We further feed our embeddings as features in Chinese named entity recognition and obtain noticeable improvements against single embeddings. | Learning Sense-specific Word Embeddings By Exploiting Bilingual Resources |
d219302566 | ||
d218974164 | ||
d61159418 | Letter-to-sound rules, also known as grapheme-to-phoneme rules, are important computational tools and have been used for a variety of purposes including word or name lookups for database searches and speech synthesis.These rules are especially useful when integrated into database searches on names and addresses, since they can complement orthographic search algorithms that make use of permutation, deletion, and insertion by allowing for a comparison with the phonetic equivalent. In databases, phonetics can help retrieve a word or a proper name without the user needing to know the correct spelling. A phonetic index is built with the vocabulary of the application. This could be an entire dictionary, or a list of proper names. The searched word is then converted into phonetics and retrieved with its information, if the word is in the phonetic index. This phonetic lookup can be used to retrieve a misspelled word in a dictionary or a database, or in a text editor to suggest corrections.Such rules are also necessary to formalize grapheme-phoneme correspondences in speech synthesis architecture. In text-to-speech systems, these rules are typically used to create phonemes from computer text. These phonemic symbols, in turn, are used to feed lower-level phonetic modules (such as timing, intonation, vowel formant trajectories, etc.) which, in turn,feed a vocal tract model and finally output a waveform and, via a digital-analogue converter, synthesized speech. Such rules are a necessary and integral part of a text-to-speech system since a database lookup (dictionary search) is not sufficient to handle derived forms, new words, nonce forms, proper nouns, low-frequency technical jargon, and the like; such forms typically are not included in the database. And while the use of a dictionary is more important now that denser and faster memory is available to smaller systems, letter-to-sound still plays a crucial and central role in speech synthesis technology.Grapheme-to-phoneme technology is also useful in speech recognition, as a way of generating pronunciations for new words that may be available in grapheme form, or for naive users to add new words more easily. In that case, the system must generate the multiple variations of the word.While there are different problems in languages that use non-alphabetic writing systems (syllabaries, as in Japanese, or logographic systems, as in Chinese)(DeFrancis 1984), all alphabetic systems have a structured set of correspondences. These range from the trivial in languages like Spanish or Swahili, to extremely complex in languages such as English and French. This paper Computational Linguistics Volume 23, Number 4 will outline some of the previous attempts to construct such rule sets and will describe new and successful approaches to the construction of letter-to-sound rules for English and French. | Algorithms for Grapheme-Phoneme Translation for English and French: Applications for Database Searches and Speech Synthesis |
d11209971 | NBSVM is one of the most popular methods for text classification and has been widely used as baselines for various text representation approaches. It uses Naive Bayes (NB) feature to weight sparse bag-of-n-grams representation. N-gram captures word order in short context and NB feature assigns more weights to those important words. However, NBSVM suffers from sparsity problem and is reported to be exceeded by newly proposed distributed (dense) text representations learned by neural networks. In this paper, we transfer the n-grams and NB weighting to neural models. We train n-gram embeddings and use NB weighting to guide the neural models to focus on important words. In fact, our methods can be viewed as distributed (dense) counterparts of sparse bag-of-n-grams in NBSVM. We discover that n-grams and NB weighting are also effective in distributed representations. As a result, our models achieve new strong baselines on 9 text classification datasets, e.g. on IMDB dataset, we reach performance of 93.5% accuracy, which exceeds previous state-of-the-art results obtained by deep neural models. All source codes are publicly available at https://github.com/zhezhaoa/neural_BOW_toolkit. * Equal contribution. † Corresponding author. This work is licensed under a Creative Commons Attribution 4.0 International Licence.Licence details: | Weighted Neural Bag-of-n-grams Model: New Baselines for Text Classification |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.