The paper explores the development of Automatic Speech Recognition (ASR) models for Armenian, by using data from two standard dialects (Eastern Armenian and Western Armenian). The goal is to develop a joint bi-variational model. We achieve state-of-the-art results. Results from our ASR experiments demonstrate the impact of dataset selection and data volume on model performance. The study reveals limited transferability between dialects, although integrating datasets from both dialects enhances overall performance. The paper underscores the importance of dataset diversity and volume in ASR model training for under-resourced languages like Armenian.

Transformer language models are now a solid baseline for Named Entity Recognition and can be significantly improved by leveraging complementary resources, either by integrating external knowledge or by annotating additional data. In a preliminary step, this work presents experiments on fine-tuning transformer models. Then, a set of experiments has been conducted with a Wikipedia-based reclassification system. Additionally, we conducted a small annotation campaign on the Farsi language to evaluate the impact of additional data. These two methods with complementary resources showed improvements compared to fine-tuning only.

To develop an influencer detection system, we designed an influence model based on the analysis of conversations in the “Change My View” debate forum. This led us to identify enunciative features (argumentation, emotion expression, view change, ...) related to influence between participants. In this paper, we present the annotation campaign we conducted to build up a reference corpus on these enunciative features. The annotation task was to identify in social media posts the text segments that corresponded to each enunciative feature. The posts to be annotated were extracted from two social media: the “Change My View” debate forum, with discussions on various topics, and Twitter, with posts from users identified as supporters of ISIS (Islamic State of Iraq and Syria). Over a thousand posts have been double or triple annotated throughout five annotation sessions gathering a total of 27 annotators. Some of the sessions involved the same annotators, which allowed us to analyse the evolution of their annotation work. Most of the sessions resulted in a reconciliation phase between the annotators, allowing for discussion and iterative improvement of the guidelines. We measured and analysed inter-annotator agreements over the course of the sessions, which allowed us to validate our iterative approach.

L’influence sociale est un phénomène important dans divers domaines, tels que l’économie et la politique, qui a gagné en résonnance avec la popularité des médias sociaux, notamment les réseaux sociaux et les forums. La majorité des travaux sur ce sujet propose des approches fondées sur des théories en sciences humaines (sociologie, linguistique), et des techniques d’analyse de réseau (mesures de propagation et de centralité) ou de TAL. Dans cet article, nous présentons un modèle d’influence inspiré de travaux en psychologie sociale, sur lequel nous construisons un système combinant un module de TAL pour détecter les messages reflétant les processus d’influence, associé à une analyse par centralité de la transmission de ces messages. Nos expériences sur le forum de débats Change My View montrent que l’approche par hybridation, comparée à la centralité seule, aide à mieux détecter les influenceurs.

The Quechua linguistic family has a limited number of NLP resources, most of them being dedicated to Southern Quechua, whereas the varieties of Central Quechua have, to the best of our knowledge, no specific resources (software, lexicon or corpus). Our work addresses this issue by producing two resources for the Ancash Quechua: a full digital version of a dictionary, and an OCR model adapted to the considered variety. In this paper, we describe the steps towards this goal: we first measure performances of existing models for the task of digitising a Quechua dictionary, then adapt a model for the Ancash variety, and finally create a reliable resource for NLP in XML-TEI format. We hope that this work will be a basis for initiating NLP projects for Central Quechua, and that it will encourage digitisation initiatives for under-resourced languages.

In this article, we try to solve the problem of classification of counterfactual statements and extraction of antecedents/consequences in raw data, by mobilizing on one hand Support vector machine (SVMs) and on the other hand Natural Language Understanding (NLU) infrastructures available on the market for conversational agents. Our experiments allowed us to test different pipelines of two known platforms (Snips NLU and Rasa NLU). The results obtained show that a Rasa NLU pipeline, built with a well-preprocessed dataset and tuned algorithms, allows to model accurately the structure of a counterfactual event, in order to facilitate the identification and the extraction of its components.

Une simple détection d’opinions positives ou négatives ne satisfait plus les chercheurs et les entreprises. Le monde des affaires est à la recherche d’un «aperçu des affaires». Beaucoup de méthodes peuvent être utilisées pour traiter le problème. Cependant, leurs performances, lorsque les classes ne sont pas équilibrées, peuvent être dégradées. Notre travail se concentre sur l’étude des techniques visant à traiter les données déséquilibrées en parfumerie. Cinq méthodes ont été comparées : Smote, Adasyn, Tomek links, Smote-TL et la modification du poids des classe. L’algorithme d’apprentissage choisi est le SVM et l’évaluation est réalisée par le calcul des scores de précision, de rappel et de f-mesure. Selon les résultats expérimentaux, la méthode en ajustant le poids sur des coût d’erreurs avec SVM, nous permet d’obtenir notre meilleure F-mesure.

In many languages such as Bambara or Arabic, tone markers (diacritics) may be written but are actually often omitted. NLP applications are confronted to ambiguities and subsequent difficulties when processing texts. To circumvent this problem, tonalization may be used, as a word sense disambiguation task, relying on context to add diacritics that partially disambiguate words as well as senses. In this paper, we describe our implementation of a Bambara tonalizer that adds tone markers using machine learning (CRFs). To make our tool efficient, we used differential coding, word segmentation and edit operation filtering. We describe our approach that allows tractable machine learning and improves accuracy: our model may be learned within minutes on a 358K-word corpus and reaches 92.3% accuracy.

Dans cet article, nous présentons un processus d’identification automatique de l’origine dialectale pour la langue arabe de textes écrits en caractères arabes ou en écriture latine (arabizi). Nous décrivons le processus d’annotation des ressources construites et du système de translittération adopté. Deux approches d’identification de la langue sont comparées : la première est linguistique et exploite des dictionnaires, la seconde est statistique et repose sur des méthodes traditionnelles d’apprentissage automatique (n-grammes). L’évaluation de ces approches montre que la méthode linguistique donne des résultats satisfaisants, sans être dépendante des corpus d’apprentissage.

This paper gives an overview of the MultiTal project, which aims to create a research infrastructure that ensures long-term distribution of NLP tools descriptions. The goal is to make NLP tools more accessible and usable to end-users of different disciplines. The infrastructure is built on a meta-data scheme modelling and standardising multilingual NLP tools documentation. The model is conceptualised using an OWL ontology. The formal representation of the ontology allows us to automatically generate organised and structured documentation in different languages for each represented tool.

Recognition of real-world entities is crucial for most NLP applications. Since its introduction some twenty years ago, named entity processing has undergone a significant evolution with, among others, the definition of new tasks (e.g. entity linking) and the emergence of new types of data (e.g. speech transcriptions, micro-blogging). These pose certainly new challenges which affect not only methods and algorithms but especially linguistic resources. Where do we stand with respect to named entity resources? This paper aims at providing a systematic overview of named entity resources, accounting for qualities such as multilingualism, dynamicity and interoperability, and to identify shortfalls in order to guide future developments.

Dans cet article, nous présentons une méthodologie pour l’identification de messages suspectés d’être produits par des Community Managers à des fins commerciales déguisées dans des documents du Web 2.0. Le champ d’application est la malbouffe (junkfood) et le corpus est multilingue (anglais, chinois, français). Nous exposons dans un premier temps la stratégie de constitution et d’annotation de nos corpus, en explicitant notamment notre guide d’annotation, puis nous développons la méthode adoptée, basée sur la combinaison d’une analyse textométrique et d’un apprentissage supervisé.

Dans cet article, nous présentons et analysons les résultats du système de reconnaissance d’entités nommées CasEN lors de sa participation à la campagne d’évaluation Ester2. Nous identifions quelles ont été les difficultés pour notre système, essentiellement : les mots hors-vocabulaire, la métonymie, les frontières des entités nommées. Puis nous proposons une approche pour améliorer les performances de systèmes à base de connaissances, en utilisant des techniques exhaustives de fouille de données séquentielles afin d’extraire des motifs qui représentent les structures linguistiques en jeu lors de la reconnaissance d’entités nommées. Enfin, nous décrivons l’expérimentation menée à cet effet, donnons les résultats obtenus à ce jour et en faisons une première analyse.

In this paper, we present a detailed and critical analysis of the behaviour of the CasEN named entity recognition system during the French Ester2 evaluation campaign. In this project, CasEN has been confronted with the task of detecting and categorizing named entities in manual and automatic transcriptions of radio broadcastings. At first, we give a general presentation of the Ester2 campaign. Then, we describe our system, based on transducers. Next, we depict how systems were evaluated during this campaign and we report the main official results. Afterwards, we investigate in details the influence of some annotation biases which have significantly affected the estimation of the performances of systems. At last, we conduct an in-depth analysis of the effective errors of the CasEN system, providing us with some useful indications about phenomena that gave rise to errors (e.g. metonymy, encapsulation, detection of right boundaries) and are as many challenges for named entity recognition systems.