Search Results for author:
Found 61 papers, 3 papers with code
The SPECTRANS System Description for the WMT21 Terminology Task
Experiment 2 uses OpenNMT to fine-tune the model.
How Distributed are Distributed Representations? An Observation on the Locality of Syntactic Information in Verb Agreement Tasks
This work addresses the question of the localization of syntactic information encoded in the transformers representations.
Comparison between NMT and PBSMT Performance for Translating Noisy User-Generated Content
This work compares the performances achieved by Phrase-Based Statistical Machine Translation systems (PB-SMT) and attention-based Neuronal Machine Translation systems (NMT) when translating User Generated Content (UGC), as encountered in social medias, from French to English.
Flux d’informations dans les systèmes encodeur-décodeur. Application à l’explication des biais de genre dans les systèmes de traduction automatique. (Information flow in encoder-decoder systems applied to the explanation of gender bias in machine translation systems)
Ce travail présente deux séries d’expériences visant à identifier les flux d’information dans les systèmes de traduction neuronaux.
Biais de genre dans un système de traduction automatiqueneuronale : une étude préliminaire (Gender Bias in Neural Translation : a preliminary study )
Cet article présente les premiers résultats d’une étude en cours sur les biais de genre dans les corpus d’entraînements et dans les systèmes de traduction neuronale.
Les représentations distribuées sont-elles vraiment distribuées ? Observations sur la localisation de l’information syntaxique dans les tâches d’accord du verbe en français (How Distributed are Distributed Representations ? An Observation on the Locality of Syntactic)
Ce travail aborde la question de la localisation de l’information syntaxique qui est encodée dans les représentations de transformers.
Fine-tuning pre-trained models for Automatic Speech Recognition, experiments on a fieldwork corpus of Japhug (Trans-Himalayan family)
This is a report on results obtained in the development of speech recognition tools intended to support linguistic documentation efforts.
Automatic Speech Recognition
Automatic Speech Recognition (ASR)
+1
Establishing degrees of closeness between audio recordings along different dimensions using large-scale cross-lingual models
We propose a new unsupervised method using ABX tests on audio recordings with carefully curated metadata to shed light on the type of information present in the representations.
Using Artificial French Data to Understand the Emergence of Gender Bias in Transformer Language Models
Numerous studies have demonstrated the ability of neural language models to learn various linguistic properties without direct supervision.
From `Snippet-lects' to Doculects and Dialects: Leveraging Neural Representations of Speech for Placing Audio Signals in a Language Landscape
We use max-pooling to aggregate the neural representations from a "snippet-lect" (the speech in a 5-second audio snippet) to a "doculect" (the speech in a given resource), then to dialects and languages.
ProsAudit, a prosodic benchmark for self-supervised speech models
In the lexical task, the model needs to correctly distinguish between pauses inserted between words and within words.
Assessing the Capacity of Transformer to Abstract Syntactic Representations: A Contrastive Analysis Based on Long-distance Agreement
The long-distance agreement, evidence for syntactic structure, is increasingly used to assess the syntactic generalization of Neural Language Models.
Is the Language Familiarity Effect gradual? A computational modelling approach
According to the Language Familiarity Effect (LFE), people are better at discriminating between speakers of their native language.
Probing phoneme, language and speaker information in unsupervised speech representations
Language information, however, is very salient in the bilingual model only, suggesting CPC models learn to discriminate languages when trained on multiple languages.
Screening Gender Transfer in Neural Machine Translation
This paper aims at identifying the information flow in state-of-the-art machine translation systems, taking as example the transfer of gender when translating from French into English.
Mining Naturally-occurring Corrections and Paraphrases from Wikipedia's Revision History
Naturally-occurring instances of linguistic phenomena are important both for training and for evaluating automatic processes on text.
Noisy UGC Translation at the Character Level: Revisiting Open-Vocabulary Capabilities and Robustness of Char-Based Models
This work explores the capacities of character-based Neural Machine Translation to translate noisy User-Generated Content (UGC) with a strong focus on exploring the limits of such approaches to handle productive UGC phenomena, which almost by definition, cannot be seen at training time.
Understanding the Impact of UGC Specificities on Translation Quality
This work takes a critical look at the evaluation of user-generated content automatic translation, the well-known specificities of which raise many challenges for MT.
Are Transformers a Modern Version of ELIZA? Observations on French Object Verb Agreement
Many recent works have demonstrated that unsupervised sentence representations of neural networks encode syntactic information by observing that neural language models are able to predict the agreement between a verb and its subject.
Are Neural Networks Extracting Linguistic Properties or Memorizing Training Data? An Observation with a Multilingual Probe for Predicting Tense
We evaluate the ability of Bert embeddings to represent tense information, taking French and Chinese as a case study.
User-friendly automatic transcription of low-resource languages: Plugging ESPnet into Elpis
This paper reports on progress integrating the speech recognition toolkit ESPnet into Elpis, a web front-end originally designed to provide access to the Kaldi automatic speech recognition toolkit.
Automatic Speech Recognition
Automatic Speech Recognition (ASR)
+1
Analyse d'erreurs de transcriptions phon\'emiques automatiques d'une langue « rare » : le na (mosuo) (Analyzing errors in automatic phonemic transcriptions of the Na (Mosuo) language (SinoTibetan family) Automatic phonemic transcription tools now reach high levels of accuracy on a single speaker with relatively small amounts of training data: on the order two to three hours of transcribed speech)
Les syst{\`e}mes de reconnaissance automatique de la parole atteignent d{\'e}sormais des degr{\'e}s de pr{\'e}cision {\'e}lev{\'e}s sur la base d{'}un corpus d{'}entra{\^\i}nement limit{\'e} {\`a} deux ou trois heures d{'}enregistrements transcrits (pour un syst{\`e}me mono-locuteur).
Phonemic Transcription of Low-Resource Languages: To What Extent can Preprocessing be Automated?
What is at stake is the accessibility of language archive data for a range of NLP tasks and beyond.
Automatic Speech Recognition
Automatic Speech Recognition (ASR)
+1
Phonetic Normalization for Machine Translation of User Generated Content
We present an approach to correct noisy User Generated Content (UGC) in French aiming to produce a pretreatement pipeline to improve Machine Translation for this kind of non-canonical corpora.
Combien d'exemples de tests sont-ils n\'ecessaires \`a une \'evaluation fiable ? Quelques observations sur l'\'evaluation de l'analyse morphosyntaxique du fran\ccais. (Some observations on the evaluation of PoS taggers)
L{'}objectif de ce travail est de pr{\'e}senter plusieurs observations, sur l{'}{\'e}valuation des analyseurs morphosyntaxique en fran{\c{c}}ais, visant {\`a} remettre en cause le cadre habituel de l{'}apprentissage statistique dans lequel les ensembles de test et d{'}apprentissage sont fix{\'e}s arbitrairement et ind{\'e}pendemment du mod{\`e}le consid{\'e}r{\'e}.
How Bad are PoS Tagger in Cross-Corpora Settings? Evaluating Annotation Divergence in the UD Project.
The performance of Part-of-Speech tagging varies significantly across the treebanks of the Universal Dependencies project.
Quantifying training challenges of dependency parsers
Not all dependencies are equal when training a dependency parser: some are straightforward enough to be learned with only a sample of data, others embed more complexity.
Automated Paraphrase Lattice Creation for HyTER Machine Translation Evaluation
We propose a variant of a well-known machine translation (MT) evaluation metric, HyTER (Dreyer and Marcu, 2012), which exploits reference translations enriched with meaning equivalent expressions.
Exploiting Dynamic Oracles to Train Projective Dependency Parsers on Non-Projective Trees
Because the most common transition systems are projective, training a transition-based dependency parser often implies to either ignore or rewrite the non-projective training examples, which has an adverse impact on accuracy.
Automatically Selecting the Best Dependency Annotation Design with Dynamic Oracles
This work introduces a new strategy to compare the numerous conventions that have been proposed over the years for expressing dependency structures and discover the one for which a parser will achieve the highest parsing performance.
Analyse morpho-syntaxique en pr\'esence d'alternance codique (PoS tagging of Code Switching)
L{'}alternance codique est le ph{\'e}nom{\`e}ne qui consiste {\`a} alterner les langues au cours d{'}une m{\^e}me conversation ou d{'}une m{\^e}me phrase.
Divergences entre annotations dans le projet Universal Dependencies et leur impact sur l'\'evaluation des performance d'\'etiquetage morpho-syntaxique (Evaluating Annotation Divergences in the UD Project)
Ce travail montre que la d{\'e}gradation des performances souvent observ{\'e}e lors de l{'}application d{'}un analyseur morpho-syntaxique {\`a} des donn{\'e}es hors domaine r{\'e}sulte souvent d{'}incoh{\'e}rences entre les annotations des ensembles de test et d{'}apprentissage.
LIMSI@CoNLL'17: UD Shared Task
This paper describes LIMSI{'}s submission to the CoNLL 2017 UD Shared Task, which is focused on small treebanks, and how to improve low-resourced parsing only by ad hoc combination of multiple views and resources.
Adaptation au domaine pour l'analyse morpho-syntaxique (Domain Adaptation for PoS tagging)
Ce travail cherche {\`a} comprendre pourquoi les performances d{'}un analyseur morpho-syntaxiques chutent fortement lorsque celui-ci est utilis{\'e} sur des donn{\'e}es hors domaine.
Don't Stop Me Now! Using Global Dynamic Oracles to Correct Training Biases of Transition-Based Dependency Parsers
This paper formalizes a sound extension of dynamic oracles to global training, in the frame of transition-based dependency parsers.
Zero-resource Dependency Parsing: Boosting Delexicalized Cross-lingual Transfer with Linguistic Knowledge
This paper studies cross-lingual transfer for dependency parsing, focusing on very low-resource settings where delexicalized transfer is the only fully automatic option.
Apprentissage d'analyseur en d\'ependances cross-lingue par projection partielle de d\'ependances (Cross-lingual learning of dependency parsers from partially projected dependencies )
Cet article pr{\'e}sente une m{\'e}thode simple de transfert cross-lingue de d{\'e}pendances.
Ne nous arr\^etons pas en si bon chemin : am\'eliorations de l'apprentissage global d'analyseurs en d\'ependances par transition (Don't Stop Me Now ! Improved Update Strategies for Global Training of Transition-Based)
Dans cet article, nous proposons trois am{\'e}liorations simples pour l{'}apprentissage global d{'}analyseurs en d{\'e}pendances par transition de type A RC E AGER : un oracle non d{\'e}terministe, la reprise sur le m{\^e}me exemple apr{\`e}s une mise {\`a} jour et l{'}entra{\^\i}nement en configurations sous-optimales.
Investigating gender adaptation for speech translation
In this paper we investigate the impact of the integration of context into dialogue translation.
Cross-lingual and Supervised Models for Morphosyntactic Annotation: a Comparison on Romanian
Because of the small size of Romanian corpora, the performance of a PoS tagger or a dependency parser trained with the standard supervised methods fall far short from the performance achieved in most languages.
Oublier ce qu'on sait, pour mieux apprendre ce qu'on ne sait pas : une \'etude sur les contraintes de type dans les mod\`eles CRF
Quand on dispose de connaissances a priori sur les sorties possibles d{'}un probl{\`e}me d{'}{\'e}tiquetage, il semble souhaitable d{'}inclure cette information lors de l{'}apprentissage pour simplifier la t{\^a}che de mod{\'e}lisation et acc{\'e}l{\'e}rer les traitements.
Apprentissage par imitation pour l'\'etiquetage de s\'equences : vers une formalisation des m\'ethodes d'\'etiquetage easy-first
Gr{\^a}ce au lien que nous faisons entre apprentissage structur{\'e} et apprentissage par renforcement, nous sommes en mesure de proposer une m{\'e}thode th{\'e}oriquement bien justifi{\'e}e pour apprendre des m{\'e}thodes d{'}inf{\'e}rence approch{\'e}e. Les exp{\'e}riences que nous r{\'e}alisons sur quatre t{\^a}ches de TAL valident l{'}approche propos{\'e}e.
A Corpus of Machine Translation Errors Extracted from Translation Students Exercises
In this paper, we present a freely available corpus of automatic translations accompanied with post-edited versions, annotated with labels identifying the different kinds of errors made by the MT system.
