Word Sense Disambiguation

In this paper, we explore different strategies for implementing a crowdsourcing methodology for a single-step construction of an empirically-derived sense inventory and the corresponding sense-annotated corpus. We report on the crowdsourcing experiments using implementation strategies with different HIT costs, worker qualification testing, and locale restrictions. We describe multiple adjustments required to ensure successful HIT completion, given significant changes within the crowdsourcing community over the past three years.

Many researchers have used lexical networks and ontologies to mitigate synonymy and polysemy problems in Question Answering (QA), systems coupled with taggers, query classifiers, and answer extractors in complex and ad-hoc ways. We seek to make QA systems reproducible with shared and modest human effort, carefully separating knowledge from algorithms. To this end, we propose an aesthetically "clean" Bayesian inference scheme for exploiting lexical relations for passage-scoring for QA . The factors which contribute to the efficacy of Bayesian Inferencing on lexical relations are soft word sense disambiguation, parameter smoothing which ameliorates the data sparsity problem and estimation of joint probability over words which overcomes the deficiency of naive-bayes-like approaches. Our system is superior to vector-space ranking techniques from IR, and its accuracy approaches that of the top contenders at the TREC QA tasks in recent years.

L'étude de la langue française au nigérian n'a cessé d'étre confronter par des problèmes de transferts négatifs. Donc, l'objectif de cette étude vise à analyser des contraintes linguistiques de description et d'usage d'adjectifs qualificatifs français auxquelles font face les apprenants nigérians dans l'apprentissage du français langue étrangère. La méthodologie de cette étude est analytique et descriptive. Pour identifier cette interférence linguistique, un questionnaire a été employé pour cette enquête. L'étude se fonde sur la théorie de l'analyse contrastive de Robert Lado. L'étude a été faite auprès de (50) cinquante étudiants de quatrième année à l'

Vous vous apprêtez à lire un document retraçant plus de trois années de travaux de thèse. Une thèse, c'est à la fois une aventure scientifique et humaine. Dans les pages qui suivent, l'aspect scientifique de ce travail est présenté en détail. Si vous le permettez, j'aimerais prendre quelques lignes pour revenir sur son aspect humain. Je remercie Annie Foret et Christian Rétoré, les rapporteurs de cette thése, ainsi que Colin de La Higuera, Isabelle Tellier, Denis Béchet et Joseph Le Roux d'avoir accepté de faire partie du jury. Un grand merci également à mon directeur de thèse, Alexandre Dihovsky, et à mon encadrant, Denis Béchet, pour avoir cru en moi dès le début, puis pour la patience et le dévouement avec lesquels ils m'ont supporté (dans tous les sens du terme) tout au long de ces années. D'un point de vue plus personnel, je veux remercier mes amis et ma famille qui ont été un grand soutien. Merci tout particulièrement à mes parents (mon père Alsayed et ma mère Aïsha) qui m'ont fait confiance dans mes choix, à mon frère Ali et à mes soeurs. Je remercie également mon coeur Samira et mes chers enfants Mohamed, Namareq, Rahma et Abdelmouhaimen. En plus de certains collègues susmentionnés qui sont devenus des amis que j'ai eu plaisir à fréquenter hors du travail,

Ontologies have served as a knowledge representation about the whole world or some part of it. Building ontologies is a challenging and active research area. Manually constructed Ontologies often have higher quality than the ones created by automatic or semi-automatic approaches but they tend to be more applicable to small domains. Automatic approaches are considered more suitable for building large scale Ontologies where time and efforts of human experts become a bottleneck. For both paradigms, approaches to building Ontologies from Vietnamese texts are still very limited. In this paper, we propose a system that automatically builds Ontology from Vietnamese texts using cascades of annotationbased grammars. Obtained experimental results on a university organizational structure domain are very promising.

Alignment of words and multiword units plays an important role in many natural language processing applications, such as example-based machine translation, transfer rule learning for machine translation, bilingual lexicography, word sense disambiguation, etc. In this paper we describe LIHLA, a lexical aligner which uses bilingual probabilistic lexicons generated by a freely available set of tools (NATools) and language-independent heuristics to find links between single words and multiword units in Brazilian Portuguese, Spanish and English parallel texts. The method has achieved a precision of 92.48% and 84.35% and a recall of 88.32% and 76.39% on Brazilian Portuguese-Spanish and Brazilian Portuguese-English parallel texts, respectively.

Domain portability and adaptation of NLP components and Word Sense Disambiguation systems present new challenges. The difficulties found by supervised systems to adapt might change the way we assess the strengths and weaknesses of supervised and knowledge-based WSD systems. Unfortunately, all existing evaluation datasets for specific domains are lexical-sample corpora. This task presented all-words datasets on the environment domain for WSD in four languages (Chinese, Dutch, English, Italian). 11 teams participated, with supervised and knowledge-based systems, mainly in the English dataset. The results show that in all languages the participants where able to beat the most frequent sense heuristic as estimated from general corpora. The most successful approaches used some sort of supervision in the form of hand-tagged examples from the domain. 1

Nowadays the Web represents a growing collection of an enormous amount of contents where the need for better ways to find and organize the available data is becoming a fundamental issue, in order to deal with information overload. Keyword based Web searches are actually the preferred mean to seek for contents related to a specific topic. Search engines and collaborative tagging systems make possible the search for information thanks to the association of descriptive keywords to Web resources. All of them show problems of inconsistency and consequent reduction of recall and precision of searches, due to polysemy, synonymy and in general all the different lexical forms that can be used to refer to a particular meaning. A possible way to face or at least reduce these problems is represented by the introduction of semantics to characterize the contents of Web resources: each resource is described by one or more concepts instead of simple and often ambiguous keywords. To support these task the availability of a global semantic resource of reference is fundamental. On the basis of our past experience with the semantic tagging of Web resources and the SemKey Project, we are developing Tagpedia, a general-domain "encyclopedia" of tags, semantically structured for generating semantic descriptions of contents over the Web, created by mining Wikipedia. In this paper, starting from an analysis of the weak points of non-semantic keyword based Web searches, we introduce our idea of semantic characterization of Web resources describing the structure and organization of Tagpedia. We introduce our first realization of Tagpedia, suggesting all the possible improvements that can be carried out in order to exploit its full potential.

Abstract. At present tagging is experimenting a great diffusion as the most adopted way to collaboratively classify resources over the Web. In this paper, after a detailed analysis of the attempts made to improve the organization and structure of tagging systems as well as the usefulness of ...

In terms of natural language processing, Serbian belongs to low-resource languages, with a small number of available datasets and tools. In this paper, we present a novel poem classification corpus in the Serbian language, in multi-label and multiclass variants. We provide an analysis of this dataset and then use it to train and evaluate machine learning classifiers, ranging from simple linear ones to multiple large language model variants.

We describe here the principles underlying the automatic creation of a semantic map to support navigation in a lexicon, our target group being authors (speakers, writers) rather than readers. While machines can generally access information that it has stored, this does not always hold for people. A speaker may very well know a word, yet still be (occasionally) unable to access it. To help authors to overcome word-finding problems one could add to an existing electronic resource an index based on the (age-old) notion of association. Since ideas or their expressive forms (words) are related, they may evoke each other (lemon-yellow), but the likelihood for doing so varies over time and with the context. For example, the word 'piano' may prime 'instrument' or 'weight', but which of the two gets evoked depends on the context: 'concert' vs. 'house moving'. Given this dynamic aspect of the human brain, we should build the index automatically, computing the relation of terms and their weights on the fly. This dynamic creation of the index could be done via a corpus. This latter representing ideally the dictionary users' world knowledge, and the way how the prominence of words and ideas varies over time. Another important point are link-names, i.e. the type of relationship holding between two associates: [(rose) <--color (red)]. Given the fact that any query (e.g. 'India') may yield many hits, hits whose weights may be misleading, it makes sense to group the output according to some (other) category, for example, link names (color, city_of, instrument, ...). Yet, important as they may be, links or relations are hard to extract and to name. This is why we have decided to start with a very small sub-set, meronymic-, i.e. part-of relations (x is part of y, x has y, etc.).

In this paper, we applied a novel learning algorithm, namely, Deep Belief Networks (DBN) to word sense disambiguation (WSD). DBN is a probabilistic generative model composed of multiple layers of hidden units. DBN uses Restricted Boltzmann Machine (RBM) to greedily train layer by layer as a pretraining. Then, a separate fine tuning step is employed to improve the discriminative power. We compared DBN with various state-of-the-art supervised learning algorithms in WSD such as Support Vector Machine (SVM), Maximum Entropy model (MaxEnt), Naïve Bayes classifier (NB) and Kernel Principal Component Analysis (KPCA). We used all words in the given paragraph, surrounding context words and part-of-speech of surrounding words as our knowledge sources. We conducted our experiment on the SENSEVAL-2 data set. We observed that DBN outperformed all other learning algorithms.

The effort required to build a classifier for a task in a target language can be significantly reduced by utilizing the knowledge gained during an earlier effort of model building in a source language for a similar task. In this paper, we investigate whether unlabeled data in the target language can be labeled given the availability of labeled data for a similar domain in the source language. We view the problem of labeling unlabeled documents in the target language as that of clustering them such that the resulting partitioning has the best alignment with the classes provided in the source language. We develop a cross language guided clustering (CLGC) method to achieve this. We also propose a method to discover concept mapping between languages which is utilized by CLGC to transfer supervision across languages. Our experimental results show significant gains in the accuracy of labeling documents over the baseline methods.

It has repeatedly been found that very good predictive models can result from using Boolean features constructed by an an Inductive Logic Programming (ILP) system with access to relevant relational information. The process of feature construction by an ILP system, sometimes called "propositionalization", has been largely done either as a pre-processing step (in which a large set of possibly useful features are constructed first, and then a predictive model is constructed) or by tightly coupling feature construction and model construction (in which a predictive model is constructed with each new feature, and only those that result in a significant improvement in performance are retained). These represent two extremes, similar in spirit to filter and wrapperbased approaches to feature selection. An interesting, third perspective on the problem arises by taking search-based view of feature construction. In this, we conceptually view the task as searching through subsets of all possible features that can be constructed by the ILP system. Clearly an exhaustive search of such a space will usually be intractable. We resort instead to a randomised local search which repeatedly constructs randomly (but non-uniformly) a subset of features and then performs a greedy local search starting from this subset. The number of possible features usually prohibits an enumeration of all local moves. Consequently, the next move in the search-space is guided by the errors made by the model constructed using the current set of features. This can be seen as sampling non-uniformly from the set of all possible local moves, with a view of selecting only those capable of improving performance. The result is a procedure in which a feature subset is initially generated in the pre-processing style, but further alterations are guided actively by actual model predictions. We test this procedure on language processing task of word-sense disambiguation. Good models have previously been obtained for this task using an SVM in conjunction with ILP features constructed in the pre-processing style. Our results show an improvement on these previous results: predictive accuracies are usually higher, and substantially fewer features are needed.

Identifying the correct sense of a word in context is crucial for many tasks in natural language processing (machine translation is an example). State-of-the art methods for Word Sense Disambiguation (WSD) build models using hand-crafted features that usually capturing shallow linguistic information. Complex background knowledge, such as semantic relationships, are typically either not used, or used in specialised manner, due to the limitations of the feature-based modelling techniques used. On the other hand, empirical results from the use of Inductive Logic Programming (ILP) systems have repeatedly shown that they can use diverse sources of background knowledge when constructing models. In this paper, we investigate whether this ability of ILP systems could be used to improve the predictive accuracy of models for WSD. Specifically, we examine the use of a general-purpose ILP system as a method to construct a set of features using semantic, syntactic and lexical information. This feature-set is then used by a common modelling technique in the field Editors: Filip Zelezny and Nada Lavrac.

In this paper, we propose a novel approach to induce automatically a Part-Of-Speech (POS) tagger for resource-poor languages (languages that have no labeled training data). This approach is based on cross-language projection of linguistic annotations from parallel corpora without the use of word alignment information. Our approach does not assume any knowledge about foreign languages, making it applicable to a wide range of resource-poor languages. We use Recurrent Neural Networks (RNNs) as multilingual analysis tool. Our approach combined with a basic crosslingual projection method (using word alignment information) achieves comparable results to the state-of-the-art. We also use our approach in a weakly supervised context, and it shows an excellent potential for very lowresource settings (less than 1k training utterances). 1 We did not use incremental training did).

The present paper describes the current release of the Bochum English Countability Lexicon (BECL 2.1), a large empirical database consisting of lemmata from Open ANC (http://www.anc.org) with added senses from WordNet (Fellbaum 1998). BECL 2.1 contains ≈ 11,800 annotated noun-sense pairs, divided in four major countability classes and 18 fine-grained subclasses. In the current version, BECL also provides information on nouns whose senses occur in more than one class allowing a closer look on polysemy and homonymy with regard to countability. Further included are sets of similar senses using the Leacock and Chodorow (LCH) score for semantic similarity (Leacock & Chodorow 1998), information on orthographic variation, on the completeness of all WordNet senses in the database and an annotated representation of different types of proper names. The further development of BECL will investigate the different countability classes of proper names and the general relation between semantic simi...

In this paper we present a novel bilingual (Czech, English) dataset called ShadowSense developed for the purposes of word sense induction (WSI) evaluation. Unlike existing WSI datasets, ShadowSense is annotated by multiple annotators whose inter-annotator agreement represents key reliability score to be used for evaluation of systems automatically inducing word senses. In this paper we clarify the motivation for such an approach, describe the dataset in detail and provide evaluation of three neural WSI systems showing substantial differences compared to traditional evaluation paradigms.

The work presented in this paper deals with the construction of a large-vocabulary semantic network to assist computerised speech or text recognition. The semantic network is systematically constructed with semantic information about nouns and verbs from the Longman Dictionary of Contemporary English by the application of pattern matching rules. It is represented in the form of a directed graph where nodes represent word senses and links represent the types of conceptual relationships. A semantic score, for pairwise combinations of word candidates in a speech recognition lattice, can be derived by traversing the network and calculating the conceptual distance between the senses of these candidates.

There has been a recent spike in interest in multi-modal Language and Vision problems. On the language side, most of these models primarily focus on English since most multi-modal datasets are monolingual. We try to bridge this gap with a zero-shot approach for learning multi-modal representations using cross-lingual pretraining on the text side. We present a simple yet practical approach for building a cross-lingual image retrieval model which trains on a monolingual training dataset but can be used in a zero-shot crosslingual fashion during inference. We also introduce a new objective function which tightens the text embedding clusters by pushing dissimilar texts away from each other. For evaluation, we introduce a new 1K multi-lingual MSCOCO2014 caption test dataset (XTD10) in 7 languages that we collected using a crowdsourcing platform. We use this as the test set for zero-shot model performance across languages. We also demonstrate how a cross-lingual model can be used for downstream tasks like multi-lingual image tagging in a zero shot manner. XTD10 dataset is made publicly available here: • Computing methodologies → Neural networks.

Within a larger frame of facilitating human-robot interaction, we present here the creation of a core vocabulary to be learned by a robot. It is extracted from two tokenised and lemmatized scenarios pertaining to two imagined microworlds in which the robot is supposed to play an assistive role. We also evaluate two resources for their utility for expanding this vocabulary so as to better cope with the robot’s communication needs. The language under study is Romanian and the resources used are the Romanian wordnet and word embedding vectors extracted from the large representative corpus of contemporary Romanian, CoRoLa. The evaluation is made for two situations: one in which the words are not semantically disambiguated before expanding the lexicon, and another one in which they are disambiguated with senses from the Romanian wordnet. The appropriateness of each resource is discussed.

This paper outlines the methods of practical preautomatic and automatic verification of evolving lexicons and an ontology used to process natural language meaning, and several approaches that can be taken to speed up the process and decrease the cost of such verification. The main methodological direction of this type of verification is increased automation in conjunction with confirmation by native speakers with basic instructions rather than by fully trained linguist knowledge engineers. It is assumed that ontological and lexical verification based on mathematical logic has been completed prior to the outlined verification.

Semantic lexicons are indispensable to research in lexical semantics and word sense disambiguation (WSD). For the study of WSD for English text, researchers have been using different kinds of lexicographic resources, including machine readable dictionaries (MRDs), machine readable thesauri, and bilingual corpora. In recent years, WordNet has become the most widely used resource for the study of WSD and lexical semantics in general. This paper describes the Class-Based Translation Model and its application in assigning translations to nominal senses in WordNet in order to build a prototype Chinese WordNet. Experiments and evaluations show that the proposed approach can potentially be adopted to speed up the construction of WordNet for Chinese and other languages.

We present an unsupervised learning strategy for word sense disambiguation (WSD) that exploits multiple linguistic resources including a parallel corpus, a bilingual machine readable dictionary, and a thesaurus. The approach is based on Class Based Sense Definition Model (CBSDM) that generates the glosses and translations for a class of word senses. The model can be applied to resolve sense ambiguity for words in a parallel corpus. That sense tagging procedure, in effect, produces a semantic bilingual concordance, which can be used to train WSD systems for the two languages involved. Experimental results show that CBSDM trained on Longman Dictionary of Contemporary English, English-Chinese Edition (LDOCE E-C) and Longman Lexicon of Contemporary English (LLOCE) is very effectively in turning a Chinese-English parallel corpus into sense tagged data for development of WSD systems.

We present an unsupervised learning strategy for word sense disambiguation (WSD) that exploits multiple linguistic resources including a parallel corpus, a bilingual machine readable dictionary, and a thesaurus. The approach is based on Class Based Sense Definition Model (CBSDM) that generates the glosses and translations for a class of word senses. The model can be applied to resolve sense ambiguity for words in a parallel corpus. That sense tagging procedure, in effect, produces a semantic bilingual concordance, which can be used to train WSD systems for the two languages involved. Experimental results show that CBSDM trained on Longman Dictionary of Contemporary English, English-Chinese Edition (LDOCE E-C) and Longman Lexicon of Contemporary English (LLOCE) is very effectively in turning a Chinese-English parallel corpus into sense tagged data for development of WSD systems.

Semantic lexicons are indispensable to research in lexical semantics and word sense disambiguation (WSD). For the study of WSD for English text, researchers have been using different kinds of lexicographic resources, including machine readable dictionaries (MRDs), machine readable thesauri, and bilingual corpora. In recent years, WordNet has become the most widely used resource for the study of WSD and lexical semantics in general. This paper describes the Class-Based Translation Model and its application in assigning translations to nominal senses in WordNet in order to build a prototype Chinese WordNet. Experiments and evaluations show that the proposed approach can potentially be adopted to speed up the construction of WordNet for Chinese and other languages.

Dans cet article, nous presentons une approche pour mesurer la similarite semantique entre des textes heterogenes et de qualite differente provenant de differentes sources Web. Notre approche commence par extraire le contenu des textes par deux methodes : (i) utilisation d&#39;un systeme d&#39;extraction que nous avons implante et qui identifie tous les mots contenus dans un texte donne, (ii) utilisation d&#39;un thesaurus multilingue (AGROVOC). Ensuite, nous combinons les resultats des deux approches afin de mesurer la similarite entre les representations textuelles des documents. Afin d&#39;evaluer les resultats, nous nous appuyons sur deux ensembles de donnees heterogenes issus du Web (tweets et articles scientifiques). (Resume d&#39;auteur)

We propose CMSMs, a novel type of generic compositional models for syntac-tic and semantic aspects of natural lan-guage, based on matrix multiplication. We argue for the structural and cognitive plau-sibility of this model and show that it is able to cover and combine various ...

Identifying topics and concepts associated with a set of documents is a critical task for information retrieval systems. One approach is to associate a query with a set of topics selected from a fixed ontology or vocabulary of terms. The core idea of this research is using Wikipedia articles and associated pages to make a topic ontology for this purpose. The benefit of this method is that Wikipedia is an online free-content encyclopedia which is developed through a social process and kept current by the Wikipedia community. In this paper the Persian Wikipedia has been analyzed in accordance to its articles and the category link graphs to extract a Persian pseudo-ontology. Thereafter, the created ontology has been applied through a query expansion algorithm to improve the performance of an information retrieval system. Our experiments show that it is possible to improve the precision of the information retrieval system by queries expansion based on Wikipedia.

Semantic similarity is a confidence score that replicates semantic equivalence between the meanings of two sentences. Determining the similarity among sentences is one of the critical tasks which have a wide-ranging impact in recent NLP applications. This paper presents a method for identifying semantic sentence similarity among sentences using semantic relation of word senses across the different synsets using Wordnet for different part of speech of words. This method firstly detects all the semantic relations (hypernym, hyponym, holonym, meronymy etc.) considering the word as a noun and all the sense relations considering word as a verb from Wordnet respectively. Then it uses common senses between the two sets as Noun and Verb, for two input words for the calculation of semantic word similarity score. As sentence is made up of different words, these word similarity scores have been used for calculation of semantic sentence similarity among the sentences. It is difficult to achieve...

Word sense ambiguity is widely spread in all natural languages; a word may carry several distinct meanings. Human can figure out the suitable meaning according to the context in which the word occurs. The Arabic language is highly polysemous; in many situations we find it extremely necessary to disambiguate the word senses. This paper studies and compares the performance of a search engine before and after expanding the query through Interactive Word Sense Disambiguation (WSD). We found that expanding polysemous query terms by adding more specific synonyms will narrow the search into the specific targeted request and thus causes both precision and recall to increase; on the other hand, expanding the query with a more general (polysemous) synonym will broaden the search which would cause the precision to decrease.

Betingelser for brug af denne artikel Denne artikel er omfattet af ophavsretsloven, og der må citeres fra den. Følgende betingelser skal dog vaere opfyldt:  Citatet skal vaere i overensstemmelse med "god skik"  Der må kun citeres "i det omfang, som betinges af formålet"  Ophavsmanden til teksten skal krediteres, og kilden skal angives, jf. ovenstående bibliografiske oplysninger. Søgbarhed Artiklerne i de aeldre Nordiske studier i leksikografi (1-5) er skannet og OCR-behandlet. OCR står for 'optical character recognition' og kan ved tegngenkendelse konvertere et billede til tekst. Dermed kan man søge i teksten. Imidlertid kan der opstå fejl i tegngenkendelsen, og når man søger på fx navne, skal man vaere forberedt på at søgningen ikke er 100 % pålidelig.

We propose a statistical method for identifying words that have a novel sense in one corpus compared to another based on differences in their lexico-syntactic contexts in those corpora. In contrast to previous work on identifying semantic change, we focus specifically on infrequent word senses. Given the challenges of evaluation for this task, we further propose a novel evaluation method based on synthetic examples of semantic change that allows us to simulate differing degrees of sense change. Our proposed method is able to identify rather subtle simulated sense changes, and outperforms both a random baseline and a previously-proposed approach.

We replace the overlap mechanism of the Lesk algorithm with a simple, generalpurpose Naive Bayes model that measures many-to-many association between two sets of random variables. Even with simple probability estimates such as maximum likelihood, the model gains significant improvement over the Lesk algorithm on word sense disambiguation tasks. With additional lexical knowledge from Word-Net, performance is further improved to surpass the state-of-the-art results.

WordNet is probably the best known lexical resource in Natural Language Processing. While it is widely regarded as a high quality repository of concepts and semantic relations, updating and extending it manually is costly. One important type of relation which could potentially add enormous value to WordNet is the inclusion of collocational information, which is paramount in tasks such as Machine Translation, Natural Language Generation and Second Language Learning. In this paper, we present ColWordNet (CWN), an extended WordNet version with fine-grained collocational information, automatically introduced thanks to a method exploiting linear relations between analogous sense-level embeddings spaces. We perform both intrinsic and extrinsic evaluations, and release CWN for the use and scrutiny of the community.

Word Sense Disambiguation (WSD) is considered one of the most important problems in Natural Language Processing [1]. It is claimed that WSD is essential for those applications that require of language comprehension modules such as search engines, machine translation systems, automatic answer machines, second life agents, etc. Moreover, with the huge amounts of information in Internet and the fact that this information is continuosly growing in different languages, we are encourage to deal with cross-lingual scenarios where WSD systems are also needed. On the other hand, Lexical Substitution (LS) refers to the process of finding a substitute word for a source word in a given sentence. The LS task needs to be approached by firstly disambiguating the source word, therefore, these two tasks (WSD and LS) are somehow related. In this paper, we present a naïve approach to tackle the problem of cross-lingual WSD and cross-lingual lexical substitution. We use a bilingual statistical dictionary, which is calculated with Giza++ by using the EUROPARL parallel corpus, in order to calculate the probability of a source word to be translated to a target word (which is assumed to be the correct sense of the source word but in a different language). Two versions of the probabilistic model are tested: unweighted and weighted. The results were compared with those of an international competition, obtaining a good performance.

Notre travail se situe dans le cadre d'un projet d'annotation descriptive, conceptuelle et thématique de corpus textuel. Dans le présent article, nous focalisons notre attention sur l'annotation conceptuelle et plus précisément sur la tâche d'extraction des termes pertinents. Pour ceci, nous proposons trois méthodes d'extraction de termes. Ces méthodes diffèrent du point de vue sources d'information utilisées (texte intégral, résumé et/ou annotations descriptives) et de point de vue méthodes d'estimation de la pertinence des termes (calculs basés sur la fréquence des termes, calculs basés sur les relations d'association entre les termes). Afin d'évaluer ces trois méthodes nous avons eu recours à un ensemble de 19 classifieurs. Une discussion sur nos résultats met en évidence l'importance de l'exploitation des réseaux de cooccurrences dans un processus d'extraction de termes notamment dans l'objectif d'enrichir l'indexation conceptuelle de documents textuels.

This paper describes FCICU team participation in SemEval 2015 for Semantic Textual Similarity challenge. Our main contribution is to propose a word-sense similarity method using BabelNet relationships. In the English subtask challenge, we submitted three systems (runs) to assess the proposed method. In Run1, we used our proposed method coupled with a string kernel mapping function to calculate the textual similarity. In Run2, we used the method with a tree kernel function. In Run3, we averaged Run1 with a previously proposed surface-based approach as a kind of integration. The three runs are ranked 41 st , 57 th , and 20 th of 73 systems, with mean correlation 0.702, 0.597, and 0.759 respectively. For the interpretable task, we submitted a modified version of Run1 achieving mean F1 0.846, 0.461, 0.722, and 0.44 for alignment, type, score, and score with type respectively.

Text clustering is an important method for organising the increasing volume of digital content, aiding in the structuring and discovery of hidden patterns in uncategorised data. The effectiveness of text clustering largely depends on the selection of textual embeddings and clustering algorithms. This study argues that recent advancements in large language models (LLMs) have the potential to enhance this task. The research investigates how different textual embeddings, particularly those utilised in LLMs, and various clustering algorithms influence the clustering of text datasets. A series of experiments were conducted to evaluate the impact of embeddings on clustering results, the role of dimensionality reduction through summarisation, and the adjustment of model size. The findings indicate that LLM embeddings are superior at capturing subtleties in structured language. OpenAI's GPT-3.5 Turbo model yields better results in three out of five clustering metrics across most tested datasets. Most LLM embeddings show improvements in cluster purity and provide a more informative silhouette score, reflecting a refined structural understanding of text data compared to traditional methods. Among the more lightweight models, BERT demonstrates leading performance. Additionally, it was observed that increasing model dimensionality and employing summarisation techniques do not consistently enhance clustering efficiency, suggesting that these strategies require careful consideration for practical application. These results highlight a complex balance between the need for refined text representation and computational feasibility in text clustering applications. This study extends traditional text clustering frameworks by integrating embeddings from LLMs, offering improved methodologies and suggesting new avenues for future research in various types of textual analysis.

This paper presents a statistical Word Sense Disambiguation with application in Portuguese-Chinese Machine Translation systems.. Due to the limited availability of Portuguese-Chinese resources in the form of digital corpora and annotated Treebank, an unsupervised learning and a non-aligned bflingual corpus are applied. The proposed method first identifies words related to each of the ambiguous words based on their surrounding words and relative distance. A mathematical model is then applied in the identification of the most suitable sense of an ambiguous word in terms of the related words. All the senses discovered are converted into a set of rules and stored in the Sense Knowledge base for later use in disambiguation and translation process. Preliminary experiment results show an improvement of 6%o in asgning correctiy the corresponding translation over the baseline method.

Word sense disambiguation is a state of art solution attempts to determine the sense of a word from contextual features in a running text. Major barriers to building a high-performing word sense disambiguation system include the difficulty of labeling data for this task and of predicting fine-grained sense distinctions. In this paper, we address a different formulation of the wordsense disambiguation task and an analysis is being done about understanding of language. Rather than considering this task on its own. The present research based upon the analysis can be named as 'The law of conservation of thought'. Through this an atomic structure of thought, its conservation and transformation as well as the thought boundary detection process for disambiguation of the words in the text are analyzed.

New text analysis softwares issued from fields of research such as Machine Learning and Natural Languages Processing prove to be relevant tools for the language sciences. Littératron is a new data-processing tool for the automatic extraction of syntactic patterns, designed at LIP6 by Jean-Gabriel Ganascia. Associated with a linear text analyser, it reveals the stylistic peculiarities of a text. We will see that Littératron carries out a linguistic diagnosis of learners if used in language sciences, especially in the field of acquisition of written French as a foreign language. The learner can be from a heterogeneous group (various language levels and various mother tongues) or from a homogeneous group (only one language level and one mother tongue, here, Arabic). The interest of this approach is related to three fields: first, language didactics, on a purely educational basis; next, computational linguistics; finally, computer-assisted learning. De nouveaux logiciels d"analyse textuelle tirent partie des progrès récents effectués en apprentissage symbolique et dans le traitement automatique des langues naturelles. Conçu au LIP6 par Jean-Gabriel Ganascia, le Littératron est l'un d'entre eux ; il extrait automatiquement des motifs syntaxiques i à partir de textes écrits en langage naturel. Plus exactement, le Littératron prend comme entré un arbre d'analyse syntaxique et donne en sortie un certain nombre de motifs syntaxiques récurrents. Associé à un analyseur de textes, qui engendre l'arbre d'analyse syntaxique à partir de textes écrits en langage naturel, il révèle les singularités stylistiques de ces textes. Nous allons voir qu"utilisé en sciences du langage, dans le domaine de l"acquisition du français écrit, le Littératron permet d'effectuer un diagnostic linguistique de l"apprenant, que celui-ci provienne d"une classe de langue hétérogène (différentes langues maternelles) ou homogène (une seule langue maternelle, en l"occurrence ici l"arabe). L"intérêt de cette approche concerne trois domaines : d"une part la didactique des langues, à titre éducatif ; d"autre part, la linguistique computationnelle, et enfin l"enseignement assisté par ordinateur.

Building systems that find and relate relevant information whereas the non relevant information is ignored. Relevant information is determined by a predefined guide of the domain, in which the type of information to be extracted must be as specific as possible From the NLP point of view, IE systems must work on different levels: from word recognition to sentence analysis From sentence-level understanding to the whole text

This paper presents an exhaustive study about the Temporal Expression (TE) influence in the task of Word Sense Disambiguation (WSD). The hypothesis was that previous identification of some words or word groups could improved the efficiency of WSD systems. In this case, the experiments carried out show that the identification of temporal expressions made up of one or more words (i.e. today or the following day) improves around 10% precision of the Word Domain Disambiguation Framework. The improvement of the WSD task is achieved by extracting temporal expressions from the corpus which allows us to limit the spread of a search across the EuroWordNet hierarchy. The corpus used to this research was the Spanish lexical sample task from Senseval-2 1 .

Abstract. Dictionary base methods have the advantage that they can be applied to texts written in different languages if there exist an electronically dictionary for that specific language. As a consequence, these kinds of methods can be used to identify the part of ...

E-law module is the web application which works mainly as the set of information retrieval and extraction tools dedicated for the lawyers. E-law module consists of following tools: (1) document search engine; (2) context oriented search engine plugin; (3) legal phrase oriented machine translation; (4) document metatagger; (5) verdict finder. Machine translation, document meta-tagger and verdict finder tools are available for the general public. Other tools are restricted and are accessible after logging into the module.

Identifying the correct meaning of words in context or discovering new word senses is particularly useful for several tasks such as question answering, information extraction, information retrieval, and text summarization. However, specially in the context of user-generated contents and on-line communication (e.g. Twitter), new meanings are continuously crafted as the result of existing words being used in novel contexts. Consequently, lexical semantics inventories and systems have difficulties to cope with semantic drifting problems. In this work, we propose an approach to induce and disambiguate word senses of some target words in collections of short texts, such as tweets, through the use of fuzzy lexico-semantic patterns that we define as sequences of Morpho-semantic Components (MSC). We learn these patterns, that we call M SC + patterns, from text data automatically. Experimental results show that instances of some M SC + patterns arise in a number of tweets, but sometimes using different words to convey the sense of the respective MSC in some tweets where pattern instances appear. The exploitation of M SC + patterns when they induce semantics on target words enable effective word sense disambiguation mechanisms leading to improvements in the state of the art.