3.1.1 Research axis 1

Our first objective is to stay at a state-of-the-art level in key NLP tasks such as shallow processing, part-of-speech tagging and (syntactic) parsing, which are core expertise domains of ALMAnaCH members. This will also require us to improve the generation of semantic representations (semantic parsing), and to begin to explore tasks such as machine translation, which now relies on neural architectures also used for some of the above-mentioned tasks. Given the generalisation of neural models in NLP, we will also be involved in better understanding how such models work and what they learn, something that is directly related to the investigation of language variation (Research axis 2). We will also work on the integration of both linguistic and non-linguistic contextual information to improve automatic linguistic analysis. This is an emerging and promising line of research in NLP. We will have to identify, model and take advantage of each type of contextual information available. Addressing these issues will enable the development of new lines of research related to conversational content. Applications include improved information and knowledge extraction algorithms. We will especially focus on challenging datasets such as domain-specific texts (e.g. financial, legal) as well as historical documents, in the larger context of the development of digital humanities. We currently also explore the even more challenging new direction of a cognitively inspired NLP, in order to tackle the possibility to enrich the architecture of state-of-the-art algorithms, such as RNNGs, based on human neuroimaging-driven data.

3.1.2 Research axis 2

Language variation must be better understood and modelled in all its forms. In this regard, we will put a strong emphasis on four types of language variation and their mutual interaction: sociolinguistic variation in synchrony (including non-canonical spelling and syntax in user-generated content), complexity-based variation in relation to language-related disabilities, and diachronic variation (computational exploration of language change and language history, with a focus on Old to all forms of Modern French, as well as Indo-European languages in general). In addition, the noise introduced by Optical Character Recognition and Handwritten Text Recognition systems, especially in the context of historical documents, bears some similarities to that of non-canonical input in user-generated content (e.g. erroneous characters). This noise constitutes a more transverse kind of variation stemming from the way language is graphically encoded, which we call language-encoding variation. Other types of language variation will also become important research topics for ALMAnaCH in the future. This includes dialectal variation (e.g. work on Arabic varieties, something on which we have already started working, producing the first annotated data set on Maghrebi Arabizi, the Arabic variants used on social media by people from North-African countries, written using a non-fixed Latin-script transcription) as well as the study and exploitation of paraphrases in a broader context than the above-mentioned complexity-based variation.

Both research axes above rely on the availability of language resources (corpora, lexicons), which is the focus of our third, transverse research axis.

3.1.3 Research axis 3

Language resource development (raw and annotated corpora, lexical resources) is not just a necessary preliminary step to create both evaluation datasets for NLP systems and training datasets for NLP systems based on machine learning. When dealing with datasets of interest to researchers from the humanities (e.g. large archives), it is also a goal per se and a preliminary step before making such datasets available and exploitable online. It involves a number of scientific challenges, among which (i) tackling issues related to the digitalisation of non-electronic datasets, (ii) tackling issues related to the fact that many DH-related datasets are domain-specific and/or not written in contemporary languages; (iii) the development of semi-automatic and automatic algorithms to speed up the work (e.g. automatic extraction of lexical information, low-resource learning for the development of pre-annotation algorithms, transfer methods to leverage existing tools and/or resources for other languages, etc.) and (iv) the development of formal models to represent linguistic information in the best possible way, thus requiring expertise at least in NLP and in typological and formal linguistics. Such endeavours are domains of expertise of the ALMAnaCH team, and a large part of our research activities will be dedicated to language resource development. In this regard, we aim to retain our leading role in the representation and management of lexical resource and treebank development and also to develop a complete processing line for the transcription, analysis and processing of complex documents of interest to the humanities, in particular archival documents. This research axis 3 will benefit the whole team and beyond, and will benefit from and feed the work of the other research axes.

3.2.1 Processing of natural language at all levels: morphology, syntax, semantics

Our expertise in NLP is the outcome of more than 10 years in developing new models of analysis and accurate techniques for the full processing of any kind of language input since the early days of the Atoll project-team and the rise of linguistically informed data-driven models as put forward within the Alpage project-team.

Traditionally, a full natural language process (NLP) chain is organised as a pipeline where each stage of analysis represents a traditional linguistic field (in a structuralism view) from morphological analysis to purely semantic representations. The problem is that this architecture is vulnerable to error propagation and very domain sensitive: each of these stage must be compatible at the lexical and structure levels they provide. We arguably built the best performing NLP chain for French 97, 145 and one of the best for robust multilingual parsing as shown by our results in various shared tasks over the years 141, 137, 146, 108. So we pursue our efforts on each of our components we developed: tokenisers (e.g. SxPipe), part-of-speech taggers (e.g. MElt), constituency parsers and dependency parsers (e.g. FRMG, DyALog-SR) as well as our recent neural semantic graph parsers 137.

In particular, we continue to explore the hybridisation of symbolic and statistical approaches, and extend it to neural approaches, as initiated in the context of our participation to the CoNLL 2017 multilingual parsing shared task 2 and to Extrinsic Parsing Evaluation Shared Task 3.

Fundamentally, we want to build tools that are less sensitive to variation, more easily configurable, and self-adapting. Our short-term goal is to explore techniques such as multi-task learning (cf. already  129) to propose a joint model of tokenisation, normalisation, morphological analysis and syntactic analysis. We also explore adversarial learning, considering the drastic variation we face in parsing user-generated content and processing historical texts, both seen as noisy input that needs to be handled at training and decoding time.

3.2.2 Integrating context in NLP systems

While those points are fundamental, therefore necessary, if we want to build the next generation of NLP tools, we need to push the envelop even further by tackling the biggest current challenge in NLP: handling the context within which a speech act is taking place.

There is indeed a strong tendency in NLP to assume that each sentence is independent from its siblings sentences as well as its context of enunciation, with the obvious objective to simplify models and reduce the complexity of predictions. While this practice is already questionable when processing full-length edited documents, it becomes clearly problematic when dealing with short sentences that are noisy, full of ellipses and external references, as commonly found in User-Generated Content (UGC).

A more expressive and context-aware structural representation of a linguistic production is required to accurately model UGC. Let us consider for instance the case for Syntax-based Machine Translation of social media content, as is carried out by the ALMAnaCH-led ANR project Parsiti (PI: DS). A Facebook post may be part of a discussion thread, which may include links to external content. Such information is required for a complete representation of the post's context, and in turn its accurate machine translation. Even for the presumably simpler task of POS tagging of dialogue sequences, the addition of context-based features (namely information about the speaker and dialogue moves) was beneficial 110. In the case of UGC, working across sentence boundaries was explored for instance, with limited success, by 96 for document-wise parsing and by 128 for POS tagging.

Taking the context into account requires new inference methods able to share information between sentences as well as new learning methods capable of finding out which information is to be made available, and where. Integrating contextual information at all steps of an NLP pipeline is among the main research questions addressed in this research strand. In the short term, we focus on morphological and syntactic disambiguation within close-world scenarios, as found in video games and domain-specific UGC. In the long term, we investigate the integration of linguistically motivated semantic information into joint learning models.

From a more general perspective, contexts may take many forms and require imagination to discern them, get useful data sets, and find ways to exploit them. A context may be a question associated with an answer, a rating associated with a comment (as provided by many web services), a thread of discussions (e-mails, social media, digital assistants, chatbots—on which see below–), but also meta data about some situation (such as discussions between gamers in relation with the state of the game) or multiple points of views (pictures and captions, movies and subtitles). Even if the relationship between a language production and its context is imprecise and indirect, it is still a valuable source of information, notwithstanding the need for less supervised machine learning techniques (cf. the use of LSTM neural networks by Google to automatically suggest replies to emails).

3.2.3 Information and knowledge extraction

The use of local contexts as discussed above is a new and promising approach. However, a more traditional notion of global context or world knowledge remains an open question and still raises difficult issues. Indeed, many aspects of language such as ambiguities and ellipsis can only be handled using world knowledge. Linked Open Data (LODs) such as DBpedia, WordNet, BabelNet, or Framebase provide such knowledge and we plan to exploit them.

However, each specialised domain (economy, law, medicine...) exhibits its own set of concepts with associated terms. This is also true of communities (e.g. on social media), and it is even possible to find communities discussing the same topics (e.g. immigration) with very distinct vocabularies. Global LODs weakly related to language may be too general and not sufficient for a specific language variant. Following and extending previous work in ALPAGE, we put an emphasis on information acquisition from corpora, including error mining techniques in parsed corpora (to detect specific usages of a word that are missing in existing resources), terminology extraction, and word clustering.

Word clustering is of specific importance. It relies on the distributional hypothesis initially formulated by Harris, which states that words occurring in similar contexts tend to be semantically close. The latest developments of these ideas (with word2vec or GloVe) have led to the embedding of words (through vectors) in low-dimensional semantic spaces. In particular, words that are typical of several communities (see above) can be embedded in a same semantic space in order to establish mappings between them. It is also possible in such spaces to study static configurations and vector shifts with respect to variables such as time, using topological theories (such as pretopology), for instance to explore shifts in meaning over time (cf. the ANR project Profiterole concerning ancient French texts) or between communities (cf. the ANR project SoSweet). It is also worth mentioning on-going work (in computational semantics) whose goal is to combine word embeddings to embed expressions, sentences, paragraphs or even documents into semantic spaces, e.g. to explore the similarity of documents at various time periods.

Besides general knowledge about a domain, it is important to detect and keep trace of more specific pieces of information when processing a document and maintaining a context, especially about (recurring) Named Entities (persons, organisations, locations...) —something that is the focus of future work in collaboration with Patrice Lopez on named entity detection in scientific texts. Through the co-supervision of a PhD funded by the LabEx EFL (see below), we are also involved in pronominal coreference resolution (finding the referent of pronouns). Finally, we plan to continue working on deeper syntactic representations (as initiated with the Deep Sequoia Treebank), thus paving the way towards deeper semantic representations. Such information is instrumental when looking for more precise and complete information about who does what, to whom, when and where in a document. These lines of research are motivated by the need to extract useful contextual information, but it is also worth noting their strong potential in industrial applications.

3.3.1 Theoretical and empirical synchronic linguistics

Permanent members involved: all

We aim to explore computational models to deal with language variation. It is important to get more insights about language in general and about the way humans apprehend it. We will do so in at least two directions, associating computational linguistics with formal and descriptive linguistics on the one hand (especially at the morphological level) and with cognitive linguistics on the other hand (especially at the syntactic level).

Recent advances in morphology rely on quantitative and computational approaches and, sometimes, on collaboration with descriptive linguists—see for instance the special issue of the Morphology journal on “computational methods for descriptive and theoretical morphology”, edited and introduced by 94. In this regard, ALMAnaCH members have taken part in the design of quantitative approaches to defining and measuring morphological complexity and to assess the internal structure of morphological systems (inflection classes, predictability of inflected forms...). Such studies provide valuable insights on these prominent questions in theoretical morphology. They also improve the linguistic relevance and the development speed of NLP-oriented lexicons, as also demonstrated by ALMAnaCH members. We shall therefore pursue these investigations, and orientate them towards their use in diachronic models (cf. Section 3.3.3).

Regarding cognitive linguistics, we have the perfect opportunity with the starting ANR-NSF project “Neuro-Computational Models of Natural Language” (NCM-NL) to go in this direction, by examining potential correlations between medical imagery applied on patients listening to a reading of “Le Petit Prince” and computation models applied on the novel. A secondary prospective benefit from the project will be information about processing evolution (by the patients) along the novel, possibly due to the use of contextual information by humans.

3.3.2 Sociolinguistic variation

Because language is central in our social interactions, it is legitimate to ask how the rise of digital content and its tight integration in our daily life has become a factor acting on language. This is even more actual as the recent rise of novel digital services opens new areas of expression, which support new linguistic behaviours. In particular, social media such as Twitter provide channels of communication through which speakers/writers use their language in ways that differ from standard written and oral forms. The result is the emergence of new language varieties.

A very similar situation exists with regard to historical texts, especially documentary texts or graffiti but even literary texts, that do not follow standardised orthography, morphology or syntax.

However, NLP tools are designed for standard forms of language and exhibit a drastic loss of accuracy when applied to social media varieties or non-standardised historical sources. To define appropriate tools, descriptions of these varieties are needed. However, to validate such descriptions, tools are also needed. We address this chicken-and-egg problem in an interdisciplinary fashion, by working both on linguistic descriptions and on the development of NLP tools. Recently, socio-demographic variables have been shown to bear a strong impact on NLP processing tools (see for instance 106 and references therein). This is why, in a first step, jointly with researchers involved in the ANR project SoSweet (ENS Lyon and Inria project-team Dante), we will study how these variables can be factored out by our models and, in a second step, how they can be accurately predicted from sources lacking these kinds of featured descriptions.

3.3.3 Diachronic variation

Language change is a type of variation pertaining to the diachronic axis. Yet any language change, whatever its nature (phonetic, syntactic...), results from a particular case of synchronic variation (competing phonetic realisations, competing syntactic constructions...). The articulation of diachronic and synchronic variation is influenced to a large extent by both language-internal factors (i.e. generalisation of context-specific facts) and/or external factors (determined by social class, register, domain, and other types of variation).

Very few computational models of language change have been developed. Simple deterministic finite-state-based phonetic evolution models have been used in different contexts. The PIElexicon project 122 uses such models to automatically generate forms attested in (classical) Indo-European languages but is based on an idiosyncratic and unacceptable reconstruction of the Proto-Indo-European language. Probabilistic finite-state models have also been used for automatic cognate detection and proto-form reconstruction, for example by 95 and 107. Such models rely on a good understanding of the phonetic evolution of the languages at hand.

In ALMAnaCH, our goal is to work on modelling phonetic, morphological and lexical diachronic evolution, with an emphasis on computational etymological research and on the computational modelling of the evolution of morphological systems (morphological grammar and morphological lexicon). These efforts will be in direct interaction with sub-strand 3b (development of lexical resources). We want to go beyond the above-mentioned purely phonetic models of language and lexicon evolution, as they fail to take into account a number of crucial dimensions, among which: (1) spelling, spelling variation and the relationship between spelling and phonetics; (2) synchronic variation (geographical, genre-related, etc.); (3) morphology, especially through intra-paradigmatic and inter-paradigmatic analogical levelling phenomena, (4) lexical creation, including via affixal derivation, back-formation processes and borrowings.

We apply our models to two main tasks. The first task, as developed for example in the context of the ANR project Profiterole, consists in predicting non-attested or non-documented words at a certain date based on attestations of older or newer stages of the same word (e.g., predicting a non-documented Middle French word based on its Vulgar Latin and Old French predecessors and its Modern French successor). Morphological models and lexical diachronic evolution models will provide independent ways to perform the same predictions, thus reinforcing our hypotheses or pointing to new challenges.

The second application task is computational etymology and proto-language reconstruction. Our lexical diachronic evolution models will be paired with semantic resources (wordnets, word embeddings, and other corpus-based statistical information). This will allow us to formally validate or suggest etymological or cognate relations between lexical entries from different languages of a same language family, provided they are all inherited. Such an approach could also be adapted to include the automatic detection of borrowings from one language to another (e.g. for studying the non-inherited layers in the Ancient Greek lexicon). In the longer term, we will investigate the feasibility of the automatic (unsupervised) acquisition of phonetic change models, especially when provided with lexical data for numerous languages from the same language family.

These lines of research will rely on etymological data sets and standards for representing etymological information (see Section 3.4.2).

Diachronic evolution also applies to syntax, and in the context of the ANR project Profiterole, we are beginning to explore more or less automatic ways of detecting these evolutions and suggest modifications, relying on fine-grained syntactic descriptions (as provided by meta-grammars), unsupervised sentence clustering (generalising previous works on error mining, cf. 9), and constraint relaxation (in meta-grammar classes). The underlying idea is that a new syntactic construction evolves from a more ancient one by small, iterative modifications, for instance by changing word order, adding or deleting functional words, etc.

3.3.4 Accessibility-related variation

Language variation does not always pertain to the textual input of NLP tools. It can also be characterised by their intended output. This is the perspective from which we investigate the issue of text simplification (for a recent survey, see for instance  142). Text simplification is an important task for improving the accessibility to information, for instance for people suffering from disabilities and for non-native speakers learning a given language 124. To this end, guidelines have been developed to help writing documents that are easier to read and understand, such as the FALC (“Facile À Lire et à Comprendre”) guidelines for French. 4

Fully automated text simplification is not suitable for producing high-quality simplified texts. Besides, the involvement of disabled people in the production of simplified texts plays an important social role. Therefore, following previous works 105, 135, our goal will be to develop tools for the computer-aided simplification of textual documents, especially administrative documents. Many of the FALC guidelines can only be linguistically expressed using complex, syntactic constraints, and the amount of available “parallel” data (aligned raw and simplified documents) is limited. We will therefore investigate hybrid techniques involving rule-based, statistical and neural approaches based on parsing results (for an example of previous parsing-based work, see  89). Lexical simplification, another aspect of text simplification 112, 125, will also be pursued. In this regard, we have already started a collaboration with Facebook's AI Research in Paris, the UNAPEI (the largest French federation of associations defending and supporting people with intellectual disabilities and their families), and the French Secretariat of State in charge of Disabled Persons.

Accessibility can also be related to the various presentation forms of a document. This is the context in which we have initiated the OPALINE project, funded by the Programme d'Investissement d'Avenir - Fonds pour la Société Numérique. The objective is for us to further develop the GROBID text-extraction suite5 in order to be able to re-publish existing books or dictionaries, available in PDF, in a format that is accessible by visually impaired persons.

3.4.1 Construction, management and automatic annotation of Text Corpora

Corpus creation and management (including automatic annotation) is often a time-consuming and technically challenging task. In many cases, it also raises scientific issues related for instance with linguistic questions (what is the elementary unit in a text?) as well as computer-science challenges (for instance when OCR or HTR are involved). It is therefore necessary to design a work-flow that makes it possible to deal with data collections, even if they are initially available as photos, scans, wikipedia dumps, etc.

These challenges are particularly relevant when dealing with ancient languages or scripts where fonts, OCR techniques, language models may be not extant or of inferior quality, as a result, among others, of the variety of writing systems and the lack of textual data. We will therefore work on improving print OCR for some of these languages, especially by moving towards joint OCR and language models. Of course, contemporary texts can be often gathered in very large volumes, as we already do within the ANR project SoSweet, resulting in different, specific issues.

ALMAnaCH pays a specific attention to the re-usability 6 of all resources produced and maintained within its various projects and research activities. To this end, we will ensure maximum compatibility with available international standards for representing textual sources and their annotations. More precisely we will take the TEI (Text Encoding Initiative) guidelines as well the standards produced by ISO committee TC 37/SC 4 as essential points of reference.

From our ongoing projects in the field of Digital Humanities and emerging initiatives in this field, we observe a real need for complete but easy work-flows for exploiting corpora, starting from a set of raw documents and reaching the level where one can browse the main concepts and entities, explore their relationship, extract specific pieces of information, always with the ability to return to (fragments of) the original documents. The pieces of information extracted from the corpora also need to be represented as knowledge databases (for instance as RDF “linked data”), published and linked with other existing databases (for instance for people and locations).

The process may be seen as progressively enriching the documents with new layers of annotations produced by various NLP modules and possibly validated by users, preferably in a collaborative way. It relies on the use of clearly identified representation formats for the annotations, as advocated within ISO TC 37/SC 4 standards and the TEI guidelines, but also on the existence of well-designed collaborative interfaces for browsing, querying, visualisation, and validation. ALMAnaCH has been or is working on several of the NLP bricks needed for setting such a work-flow, and has a solid expertise in the issues related to standardisation (of documents and annotations). However, putting all these elements in a unified work-flow that is simple to deploy and configure remains to be done. In particular, work-flow and interface should maybe not be dissociated, in the sense that the work-flow should be easily piloted and configured from the interface. An option will be to identify pertinent emerging platforms in DH (such as Transkribus) and to propose collaborations to ensure that NLP modules can be easily integrated.

It should be noted that such work-flows have actually a large potential besides DH, for instance for exploiting internal documentation (for a company) or exploring existing relationships between entities.

3.4.2 Development of Lexical Resources

ALPAGE, the Inria predecessor of ALMAnaCH, has put a strong emphasis in the development of morphological, syntactic and wordnet-like semantic lexical resources for French as well as other languages (see for instance 8, 1). Such resources play a crucial role in all NLP tools, as has been proven among other tasks for POS tagging 131, 133, 146 and parsing, and some of the lexical resource development will be targeted towards the improvement of NLP tools. They will also play a central role for studying diachrony in the lexicon, for example for Ancient to Contemporary French in the context of the Profiterole project. They will also be one of the primary sources of linguistic information for augmenting language models used in OCR systems for ancient scripts, and will allow us to develop automatic annotation tools (e.g. POS taggers) for low-resourced languages (see already 147), especially ancient languages. Finally, semantic lexicons such as wordnets will play a crucial role in assessing lexical similarity and automating etymological research.

Therefore, an important effort towards the development of new morphological lexicons will be initiated, with a focus on ancient languages of interest. Following previous work by ALMAnaCH members, we will try and leverage all existing resources whenever possible such as electronic dictionaries, OCRised dictionaries, both modern and ancient 130, 109, 132, while using and developing (semi)automatic lexical information extraction techniques based on existing corpora 134, 136. A new line of research will be to integrate the diachronic axis by linking lexicons that are in diachronic relation with one another thanks to phonetic and morphological change laws (e.g. XIIth century French with XVth century French and contemporary French). Another novelty will be the integration of etymological information in these lexical resources, which requires the formalisation, the standardisation, and the extraction of etymological information from OCRised dictionaries or other electronic resources, as well as the automatic generation of candidate etymologies. These directions of research are already investigated in ALMAnaCH 109, 132.

An underlying effort for this research will be to further the development of the GROBID-dictionaries software, which provides cascading CRF (Conditional Random Fields) models for the segmentation and analysis of existing print dictionaries. The first results we have obtained have allowed us to set up specific collaborations to improve our performances in the domains of a) recent general purpose dictionaries such as the Petit Larousse (Nénufar project, funded by the DGLFLF in collaboration with the University of Montpellier), b) etymological dictionaries (in collaboration with the Berlin Brandenburg Academy of sciences) and c) patrimonial dictionaries such as the Dictionnaire Universel de Basnage (an ANR project, including a PhD thesis at ALMAnaCH, has recently started on this topic in collaboration with the University of Grenoble-Alpes and the University Sorbonne Nouvelle in Paris).

In the same way as we signalled the importance of standards for the representation of interoperable corpora and their annotations, we will keep making the best use of the existing standardisation background for the representation of our various lexical resources. There again, the TEI guidelines play a central role, and we have recently participated in the “TEI Lex 0” initiative to provide a reference subset for the “Dictionary” chapter of the guidelines. We are also responsible, as project leader, of the edition of the new part 4 of the ISO standard 24613 (LMF, Lexical Markup Framework)  127 dedicated to the definition of the TEI serialisation of the LMF model (defined in ISO 24613 part 1 `Core model', 2 `Machine Readable Dictionaries' and 3 `Etymology'). We consider that contributing to standards allows us to stabilise our knowledge and transfer our competence.

3.4.3 Development of Annotated Corpora

Along with the creation of lexical resources, ALMAnaCH is also involved in the creation of corpora either fully manually annotated (gold standard) or automatically annotated with state-of-the-art pipeline processing chains (silver standard). Annotations will either be only morphosyntactic or will cover more complex linguistic levels (constituency and/or dependency syntax, deep syntax, maybe semantics). Former members of the ALPAGE project have a renowned experience in those aspects (see for instance 140, 126, 139, 114) and will participate to the creation of valuable resources originating from the historical domain genre.

Under the auspices of the ANR Parsiti project, led by ALMAnaCH (PI: DS), we aim to explore the interaction of extra-linguistic context and speech acts. Exploiting extra-linguistics context highlights the benefits of expanding the scope of current NLP tools beyond unit boundaries. Such information can be of spatial and temporal nature, for instance. They have been shown to improve Entity Linking over social media streams 100. In our case, we decided to focus on a closed world scenario in order to study context and speech acts interaction. To do so, we are developing a multimodal data set made of live sessions of a first person shooter video game (Alien vs. Predator) where we transcribed all human players interactions and face expressions streamlined with a log of all in-game events linked to the video recording of the game session, as well as the recording of the human players themselves. The in-games events are ontologically organised and enable the modelling of the extra-linguistics context with different levels of granularity. Recorded over many games sessions, we already transcribed over 2 hours of speech that will serve as a basis for exploratory work, needed for the prototyping of our context-enhanced NLP tools. In the next step of this line of work, we will focus on enriching this data set with linguistic annotations, with an emphasis on co-references resolutions and predicate argument structures. The midterm goal is to use that data set to validate a various range of approaches when facing multimodal data in a close-world environment.

Node infrastructure:

Each of the Jean-Zay nodes10 we use consists of 4 GPU Nvidia Tesla V100 SXM2 32GB, 192GB of RAM, and two Intel Xeon Gold 6248 processors. One Nvidia Tesla V100 card is rated at around 300W,11 while the Xeon Gold 6248 processor is rated at 150W,12. For the DRAM we can use the work of 98 to estimate the total power draw of 192GB of RAM at around 20W. Thus, the total power draw of one Jean-Zay node at peak utilization adds up to around 1520W.

With this information, we use the formula proposed by 144 and compute the total power required for each setting:

Where $c$ and $g$ are the number of CPUs and GPUs respectively, $p_c$ is the average power draw (in W) from all CPU sockets, $p_r$ the average power draw from all DRAM sockets, and $p_g$ the average power draw of a single GPU. We estimate the total power consumption by adding GPU, CPU and DRAM consumption, and then multiplying by the Power Usage Effectiveness (PUE), which accounts for the additional energy required to support the compute infrastructure. We use a PUE coefficient of 1.20, which is the value reported by the IDRIS for the Jean-Zay supercomputer. For the real time $t$ we have to divide the reported time for each Jean-Zay project by 4, as Jean-Zay reports the computing time of each project in GPU-hours and not in per node-hours. In table 1 we report the training times in hours, as well as the total power draw (in kWh) of each Jean-Zay project associated to the ALMAnaCH team during 2021. We use this information to compute the total power consumption (multiplying by the PUE) of each project, also reported in table 1.

We can further estimate the CO2 emissions in kilograms of each single project by multiplying the total power consumption by the average CO2 emissions per kWh in our region which were around 32g/kWh in average for 202113. Thus the total CO2 emissions in kg for one single model can be computed as:

All emissions are also reported in table 1. The total emission estimate for the team adds-up to 664.4kg of CO2. The carbon footprint of a single passenger on a single trip Paris to New York, fighting economy, amounts to around 946kg of CO214.

7.1.1 Enqi

• Author:
  Benoit Sagot
• Contact:
  Benoit Sagot

7.1.2 OSCAR

• Name:
  Open Super-large Crawled ALMAnaCH coRpus
• Keywords:
  Raw corpus, Multilingual corpus
• Functional Description:

  OSCAR is a huge multilingual corpus obtained by language classification and filtering of the Common Crawl corpus using the goclassy architecture.

  OSCAR is currently shuffled at line level and no metadata is provided. Thus it is mainly intended to be used in the training of unsupervised language models for natural language processing.

  Data is distributed by language in both original and deduplicated form. There are currently 166 different languages available.

• Release Contributions:
  Version 21.09 was generated using Ungoliant version v1, a new generation tool, faster and better documented/tested than the previous one, goclassy, used for OSCAR 1.0 (aka OSCAR 2019). As per OSCAR Schema v1.1, each document/record now has associated metadata. New languages with respect to version 2019: Manx, Rusyn, Scots and West Flemish. Their size and quality still has to be assessed. Removed languages with respect to version 2019: Central Bikol and Cantonese. Cantonsese was of a very low quality. Central Bikol corpus is still available on OSCAR 2019.
• URL:
  https://­oscar-corpus.­com/
• Publications:
  hal-02148693, hal-03301590, hal-03536361, hal-03177623
• Contact:
  Pedro Ortiz Suarez
• Participants:
  Pedro Ortiz Suarez, Benoit Sagot, Julien Abadji

7.1.3 ACCESS

• Keyword:
  Text Simplification
• Functional Description:
  Text simplification aims at making a text easier to read and understand by simplifying grammar and structure while keeping the underlying information identical. It is often considered an all-purpose generic task where the same simplification is suitable for all, however multiple audiences can benefit from simplified text in different ways. We adapt a discrete parametrization mechanism that provides explicit control on simplification systems based on Sequence-to-Sequence models. As a result, users can condition the simplifications returned by a model on attributes such as length, amount of paraphrasing, lexical complexity and syntactic complexity. We also show that carefully chosen values of these attributes allow out-of-the-box Sequence-to-Sequence models to outperform their standard counterparts on simplification benchmarks. Our model, which we call ACCESS (as shorthand for AudienCe-CEntric Sentence Simplification), establishes the state of the art at 41.87 SARI on the WikiLarge test set, a +1.42 improvement over the best previously reported score.
• URL:
  https://­github.­com/­facebookresearch/­access
• Publication:
  hal-02445874
• Contact:
  Louis Martin
• Participants:
  Louis Martin, Benoit Sagot, Éric De La Clergerie, Antoine Bordes

7.1.4 ASSET

• Keyword:
  Text Simplification
• Functional Description:
  In order to simplify a sentence, human editors perform multiple rewriting transformations: they split it into several shorter sentences, paraphrase words (i.e. replacing complex words or phrases by simpler synonyms), reorder components, and/or delete information deemed unnecessary. Despite these varied range of possible text alterations, current models for automatic sentence simplification are evaluated using datasets that are focused on a single transformation, such as lexical paraphrasing or splitting. This makes it impossible to understand the ability of simplification models in more realistic settings. To alleviate this limitation, this paper introduces ASSET, a new dataset for assessing sentence simplification in English. ASSET is a crowdsourced multi-reference corpus where each simplification was produced by executing several rewriting transformations. Through quantitative and qualitative experiments, we show that simplifications in ASSET are better at capturing characteristics of simplicity when compared to other standard evaluation datasets for the task. Furthermore, we motivate the need for developing better methods for automatic evaluation using ASSET, since we show that current popular metrics may not be suitable when multiple simplification transformations are performed.
• URL:
  https://­github.­com/­facebookresearch/­asset
• Publication:
  hal-02889823
• Contact:
  Louis Martin
• Participants:
  Louis Martin, Benoit Sagot, Éric De La Clergerie, Antoine Bordes, Fernando Alva-Manchego, Lucia Specia, Carolina Scarton

7.1.5 EASSE

• Keyword:
  Text Simplification
• Functional Description:
  We introduce EASSE, a Python package aiming to facilitate and standardise automatic evaluation and comparison of Sentence Simplification (SS) systems. EASSE provides a single access point to a broad range of evaluation resources: standard automatic metrics for assessing SS outputs (e.g. SARI), word-level accuracy scores for certain simplification transformations, reference-independent quality estimation features (e.g. compression ratio), and standard test data for SS evaluation (e.g. TurkCorpus). Finally, EASSE generates easy-to-visualise reports on the various metrics and features above and on how a particular SS output fares against reference simplifications. Through experiments, we show that these functionalities allow for better comparison and understanding of the performance of SS systems.
• URL:
  https://­github.­com/­feralvam/­easse
• Contact:
  Louis Martin

7.1.6 tseval

• Keyword:
  Text Simplification
• Functional Description:
  The evaluation of text simplification (TS) systems remains an open challenge. As the task has common points with machine translation (MT), TS is often evaluated using MT metrics such as BLEU. However, such metrics require high quality reference data, which is rarely available for TS. TS has the advantage over MT of being a monolingual task, which allows for direct comparisons to be made between the simplified text and its original version. In this paper, we compare multiple approaches to reference-less quality estimation of sentence-level text simplification systems, based on the dataset used for the QATS 2016 shared task. We distinguish three different dimensions: grammaticality, meaning preservation and simplicity. We show that n-gram-based MT metrics such as BLEU and METEOR correlate the most with human judgment of grammaticality and meaning preservation, whereas simplicity is best evaluated by basic length-based metrics.
• URL:
  https://­github.­com/­facebookresearch/­text-simplification-evaluation
• Contact:
  Louis Martin

7.1.7 PAGnol

• Keywords:
  Language model, French, Text generation
• Functional Description:
  PAGnol is a collection of large French language models, geared towards free-form text generation. With 1.5 billion parameters, PAGnol-XL is the largest model available for French. PAGnol is based on the GPT-3 architecture with some GPT-2 specific components, and uses scaling laws predictions for efficient training. Using scaling laws, we efficiently rain PAGnol-XL (1.5B parameters) with the same computational budget as much smaller Bert-based models for French. PAGnol-XL is the largest model trained to date for the French language.
• Contact:
  Djame Seddah
• Partner:
  Lighton

7.1.8 PFSMB

• Name:
  Parallel French Social Media Bank
• Keywords:
  Machine translation, User-generated content, Social medias
• Functional Description:
  The PFSMB is a collection of French-English parallel sentences manually translated from an extension of the French Social Media Bank (Seddah et al., 2012)] which contains texts collected on Facebook, Twitter, as well as from the forums of JeuxVideos.com and Doctissimo.fr. This corpus, consists of 1,554 comments in French annotated with different kind of linguistic information: Part-of-Speech tags, surface syntactic representations, as well as a normalized form whenever necessary. Comments have been translated from French to English by a native French speaker and extremely fluent, near-native, English speaker. Typographic and grammatical error were corrected in the gold translations but the language register was kept. For instance, idiomatic expressions were mapped directly to the corresponding ones in English (e.g. 'mdr' has been translated to 'lol' and letter repetitions were also kept (e.g. 'ouiii' has been translated to yesss').
• Publications:
  hal-02270524, hal-00780895
• Contact:
  Djame Seddah

7.1.9 EtymDB

• Name:
  Etymological DataBase
• Keyword:
  Lexicon
• Functional Description:
  EtymDB is an etymological database automatically extracted from wiktionary, available in several formats (TSV, XML/TEI).
• Release Contributions:
  Extraction from a more recent version of wiktionary, improvement of the extraction process.
• URL:
  https://­files.­inria.­fr/­almanach/­software_and_resources/­default/­EtymDB-en.­html
• Publications:
  hal-02678100, hal-01592061, hal-01584013
• Contact:
  Benoit Sagot
• Participants:
  Benoit Sagot, Clementine Fourrier

7.1.10 KaMI-Lib

• Name:
  KaMI (Kraken Model Inspector) - Python Library
• Keywords:
  HTR, OCR, Python, Handwritten Text Recognition, Image segmentation, Library
• Functional Description:

  KaMI-lib (Kraken as Model Inspector) is a Python library for evaluating transcription models (handwritten text recognition and optical character recognition) trained either with the Kraken engine (http://kraken.re) or without it.

  It provides a single class for comparing strings (e.g. extracted from text files) and for generating scores in order to evaluate the automatic transcription’s performance. The Kraken engine is implemented in KaMI-lib in order to produce a prediction with a pre-trained transcription model and to compare it to a ground truth (in PAGE XML or XML ALTO format) associated with its image.

  KaMI-lib uses different metrics to evaluate a transcription model: the Word Error Rate (WER), the Character Error Rate (CER), and the Word Accuracy (Wacc). In addition, KaMI-lib provides the edit distances and the operations performed on the different strings. It is also possible to weigh the cost of operations in order to adjust scores.

  It is also possible to get different scores with text pre-processing functions applied to the ground truth and the prediction, such as deleting all diacritics, ponctuations, or numbers, ignoring upper case, etc. By doing so, KaMI-lib aims to give a better understanding of text features’ impacts on transcription results. This functionality also aims to make users adapt the creation of training data according to their texts’ specificites, and optimize the training process.

  Documentation is available here: https://gitlab.inria.fr/dh-projects/kami/kami-lib

• URL:
  https://­github.­com/­KaMI-tools-project/­KaMi-lib
• Publications:
  hal-03495762, hal-03008579
• Contact:
  Lucas Terriel
• Participants:
  Alix Chague, Lucas Terriel, Hugo Scheithauer

7.1.11 Ungoliant

• Name:
  Ungoliant
• Keyword:
  Natural language processing
• Functional Description:
  Ungoliant is a high-performance pipeline that provides tools to build corpus generation pipelines from CommonCrawl. It currently is the generation pipeline for OSCAR corpus. Ungoliant is a replacement of the goclassy pipeline.
• URL:
  https://­github.­com/­oscar-project/­ungoliant
• Publications:
  hal-03301590, hal-03536361
• Contact:
  Julien Abadji
• Participants:
  Julien Abadji, Pedro Ortiz Suarez, Benoit Sagot

7.1.12 HTR-United

• Keywords:
  HTR, OCR
• Functional Description:

  HTR-United is a Github organization without any other form of legal personality. It aims at gathering HTR/OCR transcriptions of all periods and styles of writing, mostly but not exclusively in French. It was born from the mere necessity for projects- to possess potentiel ground truth to rapidly train models on smaller corpora.

  Datasets shared or referenced with HTR-United must, at minimum, take the form of: (i) an ensemble of ALTO XML and/or PAGE XML files containing either only information on the segmentation, either the segmentation and the corresponding transcription, (ii) an ensemble of corresponding images. They can be shared in the form of a simple permalink to ressources hosted somewhere else, or can be the contact information necessary to request access to the images. It must be possible to recompose the link between the XML files and the image without any intermediary process, (iii) a documentation on the transcription practices followed for the segmentation and the transcription. In the cases of a Github repository, this documentation must be summarized in the README.

  A corpus can be sub-diveded into smaller ensembles if it seems necessary.

• Release Contributions:
  First version.
• URL:
  https://­htr-united.­github.­io/
• Contact:
  Alix Chague

Image-enhanced MT

In the context of Matthieu Futeral-Peter's PhD thesis and in collaboration with Ivan Laptev and Cordelia Schmid, his co-supervisors from the WILLOW project-team, we have beenn investigating how using image data can improve the processing of text. In 2022 we focused on the incorporation of the image modality in MT, i.e. Multimodal Machine Translation (MMT). One of the major challenges of MT is ambiguity, which can in some cases be resolved by accompanying context such as an image. However, recent work in MMT has shown that obtaining improvements from images is challenging, limited not only by the difficulty of building effective cross-modal representations but also by the lack of specific evaluation and training data. We developed and presented a new MMT approach based on a strong text-only MT model, which uses neural adapters and a novel guided self-attention mechanism and which is jointly trained on both visual masking and MMT 74. We also released CoMMuTE, a Contrastive Multilingual Multimodal Translation Evaluation dataset, composed of ambiguous sentences and their possible translations, accompanied by disambiguating images corresponding to each translation. Our approach obtains competitive results over strong text-only models on standard English-to-French benchmarks and outperforms these baselines and state-of-the-art MMT systems with a large margin on our contrastive test set.

Bridging the gap between text and speech in MT using sentence embeddings

In the context of Paul-Ambroise Duquenne's PhD thesis and in collaboration with Holger Schwenk, his co-superviser at META, we proposed a new modular architecture for text and speech translation, which is based on a common fixed-size multilingual and multimodal internal representation, and encoders and decoders which are independently trained 25. We explored several variants of teacher-student training to learn text and speech encoders for multiple languages, which are compatible with the embedding space of the LASER encoder 90. In contrast to preceding works on multilingual and multimodal representations, we also trained text decoders for multiple languages which are able to generate translations given the joint representation. Finally, we demonstrated that it is possible to train a speech decoder using raw audio only. We showed that these encoders and decoders can be freely combined to achieve very competitive performance in T2T, S2T and (zero-shot) S2S translation.

Modelling raw speech: unsupervised sequence embedding and “word” boundary detection

Finding word boundaries in continuous speech is challenging as there is little or no equivalent of a “space” delimiter between words. Popular Bayesian non-parametric models for text segmentation 104, 103 use a Dirichlet process to jointly segment sentences and build a lexicon of word types. In the context of Robin Algayres's PhD thesis, co-supervised by Emmanuel Dupoux, from the CoML team at Inria Paris, EHESS and ENS, we introduced a new algorithm, named DP-Parse, which uses similar principles but only relies on an instance lexicon of word tokens, avoiding the clustering errors that arise with a lexicon of word types 14. On the Zero Resource Speech Benchmark 2017 speech segmentation task, our model sets a new state of the art in 5 languages. The algorithm monotonically improves with better input representations, achieving even higher scores when fed with weakly supervised inputs. Despite lacking a type lexicon, DP-Parse can be pipelined to a language model and learn semantic and syntactic representations as assessed by a new spoken word embedding benchmark.

We also introduced a simple neural encoder architecture that can be trained using an unsupervised contrastive learning objective and which gets its positive samples from data-augmented $k$-Nearest Neighbors search 21. We show that when built on top of recent self-supervised audio representations (MFCCs, CPC, HuBERT and Wav2Vec 2.0, both Base and Large), this method can be applied iteratively and yield competitive performance as evaluated on two tasks: query-by-example of random sequences of speech, and spoken term discovery—a task that can be key to finding “word” boundaries, including in DPParse. On both tasks our method pushes the state of the art by a significant margin across 5 different languages. Finally, we establish a benchmark on a query-by-example task on the LibriSpeech dataset to monitor future improvements in the field.

Questioning further how speech sequences can be represented, we investigated the impact of relying first on transforming the audio into a sequence of discrete units (or pseudo-text) and then training a language model directly on such pseudo-text, as is commonly done. Note that this can be achieved either in a supervised way or unsupervisedly, as in our work cited above. In the context of Tu Anh Nguyen's PhD thesis, also co-supervised by Emmanuel Dupoux (this time as a META fellow), our scientific question was the following: is such a discrete bottleneck necessary, potentially introducing irreversible errors in the encoding of the speech signal, or could we learn a language model without discrete units at all? We studied the role of discrete versus continuous representations in spoken language modelling and showed that discretisation is indeed essential 17. We showed that it removes linguistically irrelevant information from the continuous features, helping to improve language modeling performances. On the basis of this study, we trained a language model on the discrete units of the HuBERT features, reaching new state-of-the-art results in the lexical, syntactic and semantic metrics of the Zero Resource Speech Challenge 2021 (Track 1-Speech Only).

Dialogue-level speech language modelling

In collaboration with a number of colleagues from META and also in the context of Tu Anh Nguyen's PhD thesis, we introduced dGSLM, the first “textless” model able to generate audio samples of naturalistic spoken dialogues 78. It uses recent work on unsupervised spoken unit discovery (HuBERT + $k$-NN clustering) coupled with a dual-tower transformer architecture with cross-attention trained on 2000 hours of two-channel raw conversational audio (Fisher dataset) without any text or labels. We showed that our model is able to generate speech, laughter and other paralinguistic signals in the two channels simultaneously and reproduces more naturalistic and fluid turn-taking compared to a text-based cascaded model.

Zero-shot hate-speech detection in a cross-lingual scenario

Given that many of our target languages are considered low-resource, we devoted a large amount of time to experiments in various scenarios to decide which auxiliary tasks could help alleviate the lack of target domain data. Our experiments showed that in such scenarios, Named entity recognition and sentiment analysis were tasks likely to boost the classifiers as they seem to provide interesting features that boost classification performance. Interestingly, the morpho-syntactic tasks helped in the case of related languages (training data in Spanish, target language in Italian for example). Because of these results, we expanded our north-African Arabic dialect data set with a named-entity annotation layer and a sentiment analysis layer.

We conducted an extensive set of experiments in order to validate what to do in the worst case scenario (not enough data to train a classifier for a target language). Using hate-speech data sets for Italian, English and Spanish, we were able to (i) find the optimal set of auxiliary tasks in this scenario (ii) find the right architecture and (iii) optimise our models in this case. We published our findings in the prestigious AACL conference (Asian Chapter of the Association for computational linguistics) as well as in the French NLP conference, TALN, see 35, 44

Class unbalance and multimodal classification of hateful memes

One of the issues we are currently facing is the class unbalance issue that arises when labels are not evenly distributed in a dataset. As shown in our report on the radicalisation dataset, this is an important issue. Given that this was also a problem we faced when working on the related task of hate speech, we investigated and developed methods to cope with it in a multimodal setting. The architecture we proposed was based on a joint encoding of text and images using different encoders (visualBERT 111, CLIP 123 and OFA 149), which we used to extract information from hateful memes and their associated texts. Our promising results showed that we were able to provide accurate classification of multimodal content. This work was published in the Constraints Workshop at ACL 2022 36.

10.1.1 H2020 projects

10.1.2 Collaborations in European Programs, except FP7 and H2020

10.2.1 ANR

10.2.2 Other National Initiatives

Framework agreement with Inria AP-TAL

• Duration:
  1 Apr 2020–present.
• PIs:
  Laurent Romary.
• Coordinator for ALMAnaCH:
  Laurent Romary.
• Partner:
  • APHP
• Summary:
  Within the AP-TAL and HopiTAL projects, ALMAnaCH is involved in collaborative work with APHP and other Inria teams whose goal is to help dealing with the COVID-19 pandemics. ALMAnaCH's contributions are related to the deployment of NLP techniques on COVID-19-related non-structured text data.

Participants: Laurent Romary, Tanti Kristanti.

11.1.1 Scientific events: organisation

11.1.2 Scientific events: selection

11.1.3 Journal

11.1.4 Invited talks

• Rachel Bawden:
  • Advanced Language Processing Winter School (ALPS) 2022 (19 Jan 2022): Social session on “Data resources in NLP: how to construct them and how to evaluate resource papers””.
  • LIFT TAL 2022: Journées Jointes des Groupements de Recherche “Linguistique Informatique, Formelle et de Terrain” et “Traitement Automatique des Langues” (15 Mar 2022): “Traduction automatique dans des scénarios à faibles ressources : application à des questions linguistiques”.
  • Multispeech, Loria, Nancy (3 Mar 2022): “Low-resource MT: Few-shot learning and historical language normalisation”.
  • Cognitive Science Research Group, Queen Mary University of London, UK (9 Mar 2022): “Low-resource MT: Few-shot learning and historical language normalisation”.
  • Round table on “Réduction des biais et des inégalités : L’IA peut-elle avoir un rôle ?”, organised by 3IA Côte d’Azur in collaboration with the 3IA (ANITI, MIAI, PR[AI]RIE) (22 Mar 2022): “Traitement automatique des langues pour tous : l’importance de s’adapter au langage varié des réseaux sociaux”.
  • Pôle Modélisation du laboratoire MoDyCo, Université Paris Nanterre (25 Oct 2022): “Traduction Automatique et Linguistique : une relation en crise ?”
• Benoît Sagot:
  • Société de Linguistique de Paris (19 Feb 2022): “Le traitement automatique des langues au service de la linguistique : quelques travaux à l’interface entre linguistique et informatique”.
  • Language Technologies & Digital Humanities Conference 2022, Ljubljana, Slovenia (16 Sep 2022): “Large-scale language models: challenges and perspectives”.
  • Czech-French AI Workshop on Artificial Intelligence, Prague, Czech Republic (12 Sep 2022): “Large-scale Language Models and Their Training Corpora”.
• Hugo Scheithauer:
  • Atelier Culture-Inria, Archives Nationales, (22 Mar 2022): “DataCatalogue : présentation du projet”. Joint with Laurent Romary, Federico Nurra and Frédérique Duyrat.
  • Transkribus / eScriptorium : Transcrire, annoter et éditer numériquement des documents d’archives, Bibliothèque nationale de France (9 May 2022): “LectAuRep : Données d’archives en français des XIXe et XXe siècles”.
  • DHNord 2022, online (21 Jun 2022): “Enrichir le patrimoine écrit archivistique grâce aux technologies numériques : ingénierie du projet LectAuRep”. Joint with Aurélia Rostaing.
  • Conseil Scientifique de la Bibliothèque nationale de France (28 Jun 2022): “DataCatalogue : ingénierie de projet”. Joint with Frédérique Duyrat.
  • Segmenter et annoter les images : déconstruire pour reconstruire, Institut national d'histoire de l'art, Paris (15 Nov 2022): “LectAuRep (2018-2021) : Projet de lecture automatique de répertoires de notaires”. Joint with Aurélia Rostaing.
  • La reconnaissance des écritures manuscrites et ses usages dans les archives (29 Nov 2022): “LectAuRep : un projet de recherche et développement pour la transcription automatique des répertoires de notaires”. Joint with Aurélia Rostaing.
• Lucas Terriel:
  • Atelier Culture-Inria, Archives Nationales, (22 Mar 2022): “NER4Archives (named entity recognition for archives) : Conception et réalisation d’un outil de détection, de classification et de résolution des entités nommées dans les instruments de recherche archivistiques encodés en XML/EAD”.
• Alix Chagué:
  • Documents anciens et reconnaissance automatique des écritures manuscrites, École nationale des chartes, Paris (24 Jun 2022): “Sharing HTR datasets with standardized metadata: the HTR‑United initiative”. Joint with Thibault Clérice.
  • Chairing a Round Table at DHNord2022, Lille (22 Jun 2022): “Enjeux et débats autour de l’organisation du travail dans les projets en humanités numériques”.
• Niyati Bafna:
  • Linguistic Mondays, Charles University (24 Oct 2022): “Empirical Models for an Indic Dialect Continuum”.
• Djamé Seddah:
  • ENS Lyon (Online) (21 Feb 2022): “CamemBert must die! (jk,lol)”.
  • Journée d'étude ATALA “Robustesse des Systèmes de TAL”, Maison de la Recherche, Sorbonne Univ. (25 Nov 2022): “Beyond Sesame street-based naming schemes: Camembert must Die (jk, Lol)”.

11.1.5 Scientific expertise

• Lydia Nishimwe:
  • Expert for Consultation by the Cour des Comptes (May 2022).
• Rachel Bawden:
  • Expert for Consultation by the Cour des Comptes (May 2022).
• Benoît Sagot:
  • Expert for Inria focus group on the evolution of the administrative research assistant position.

11.1.6 Research administration

• Éric de La Clergerie:
  • Member of the scientific board for the RECOLNAT infrastructure (Website).
• Rachel Bawden:
  • Member of the scientific board for Société de Linguistique de Paris (Administratrice).
• Laurent Romary:
  • President of the scientific board for ABES (Website).
  • Member of the scientific board for the ELEXIS Interoperability and Sustainability Committee (ISC) (ELEXIS is the European Lexicographic Infrastructure.).
  • Member of the scientific board for the Schloss Dagstuhl Scientific Advisory Board (Website).
  • Member of the international advisory board for the Research Infrastructure project LINDAT/CLARIAH-CZ.
• Benoît Sagot:
  • Member of the scientific board for Inria Paris's Comité des Projets (member of the Scientific Board of the Inria Paris research centre (Bureau du Comité des Projets)).
  • Member of the scientific board for Société de Linguistique de Paris (Administrateur).

11.2.1 Teaching

• Alix Chagué:
  • Master's course (M2) as part of the Master “Documentation et Humanités Numériques”. Méthodologie de la recherche et préprofessionalisation (18 hours), coorganised with Françoise Dalex. École du Louvre, France.
  • Master's course (M2) as part of the Master “Documentation et Humanités Numériques”. Typologie, formats et outils d'exploitation des documents numériques : introduction à XML TEI (6 hours). École du Louvre, France.
  • Tutorial for the workshop “HTR@BnF”, Bibliothèque nationale de France, Paris. eScriptorium (4 hours), coorganised with Daniel Stoekl Ben Ezra;Peter Stokes. Bibliothèque nationale de France, Paris.10 May 2022.
  • Tutorial for the eScriptorium workshop at the DH2022 conference, University of Tokyo. Hands-on Introduction to eScriptorium, an Open-Source Platform for HTR (3.5 hours), coorganised with Daniel Stoekl Ben Ezra;Peter Stokes. University of Tokyo.26 Jul 2022.
• Hugo Scheithauer:
  • Master's course (M2) as part of the Master “Documentation et Humanités Numériques”. Collecte et publication de jeux de données avec Python (3.3 hours). École du Louvre, France.
  • Master's course (M2) as part of the Master “Documentation et Humanités Numériques”. Introduction à Python et à l'algorithmie (9 hours). École du Louvre, France.
  • Master's course (M2) as part of the Master “Documentation et Humanités Numériques”. Collecte et publication de jeux de données avec Python (4.3 hours). école du Louvre, France.
  • Master's course (M2) as part of the Master “Documentation et Humanités Numériques”. Méthodologie de la recherche et préprofessionalisation (18 hours), coorganised with Françoise Dalex. École du Louvre, France.
• Syrielle Montariol:
  • Master's course (M2) as part of the Master in statistics. Text Mining (24 hours), coorganised with Guillaume Gravier. ENSAI.
• Benoît Sagot:
  • Master's course (M2) as part of the Master “Mathématiques, Vision Apprentissage”. Speech and Language Processing (20 hours), coorganised with Emmanuel Dupoux and Paul Michel. ENS Paris-Saclay, France.
  • Certificate Data AI Product Owner Société Générale: on site session, PariSanté Campus. Introduction to NLP (7 hours). PariSanté Campus.5 Jul 2022.
  • Certificate Data AI Product Owner Société des Ingénieurs de l'Automobile, PariSanté Campus. Introduction to NLP (14 hours). PariSanté Campus.11 Jul 2022–13 Jul 2022.
  • Certificate Data AI Product Owner Société Générale: online session, Online. Introduction to NLP, part 1 (3.5 hours). Online.13 Oct 2022.
  • Certificate Data AI Product Owner Société Générale: online session, Online. Introduction to NLP, part 2 (3.5 hours). Online.15 Nov 2022.
• Floriane Chiffoleau:
  • 2-day training, URFIST de Bretagne et des Pays de la Loire, France. Introduction à la TEI (12 hours). URFIST de Bretagne et des Pays de la Loire, France.1 Nov 2022–1 Dec 2022.
• Djamé Seddah:
  • Tutorial for Advanced Language Processing winter school , Online. “CamemBert must die! (jk,lol. Beyond Sesame street-based naming schemes: Camembert vs CharacterBert, a study on the performance robustness of large monolingual language models and their character-based counterparts” (1 hour). Online.20 Jan 2022.

11.2.2 Supervision

11.2.3 Juries

11.3.1 Articles and contents

11.3.2 Education

• Tu Anh Nguyen was a speaker at MASSP - Math and Science Summer Program (Summer School for High school students on Data Science), “Introductory courses on “Unsupervised Learning” & “Deep Learning””. Online, 7 Aug 2022 (6 hours).
• Rachel Bawden was a speaker at the Rendez-Vous des Jeunes Mathématiciennes et Informaticiennes Inria 2022 (Presentation on NLP to high school girls), “Traitement Automatique des Langues (TAL)”. Inria Paris, 1 Oct 2022 (1 hour).

11.3.3 Interventions

• Éric de La Clergerie, with IRCAM, Sorbonne Université, gave a talk at Deep Voice Paris (Professional workshop), “Le biais est dans le texte?”. Paris, 17 Jun 2022 (1.5 hours).
• Djamé Seddah, with IRCAM, Sorbonne Université, gave a talk at Deep Voice Paris (Professional workshop), “Le biais est dans le texte?”. Paris, 17 Jun 2022 (1.5 hours).
• Éric de La Clergerie gave a talk at Healthcare NLP meetup (Professional workshop), “Addressing the processing of medical textual documents”. Paris, 26 Oct 2022 (1 hour).
• Alix Chagué, with AI4LAM, gave a talk at AI4LAM Community Call (Professional workshop), “HTR-United: Commons for Automatic Text Recognition?”. Online, 15 Nov 2022 (1 hour).

International journals

• 14 articleR.Robin Algayres, T.Tristan Ricoul, J.Julien Karadayi, H.Hugo Laurençon, S.Salah Zaiem, A.Abdelrahman Mohamed, B.Benoît Sagot and E.Emmanuel Dupoux. DP-Parse: Finding Word Boundaries from Raw Speech with an Instance Lexicon.Transactions of the Association for Computational Linguistics10September 2022, 1051–1065
  HALDOIback to text
• 15 articleB.Barry Haddow, R.Rachel Bawden, A. V.Antonio Valerio Miceli Barone, J.Jindřich Helcl and A.Alexandra Birch. Survey of Low-Resource Machine Translation.Computational Linguistics4832022, 673--732
  HAL
• 16 articleJ.Julia Kreutzer, I.Isaac Caswell, L.Lisa Wang, A.Ahsan Wahab, D.Daan van Esch, N.Nasanbayar Ulzii-Orshikh, A.Allahsera Tapo, N.Nishant Subramani, A.Artem Sokolov, C.Claytone Sikasote, M.Monang Setyawan, S.Supheakmungkol Sarin, S.Sokhar Samb, B.Benoît Sagot, C.Clara Rivera, A.Annette Rios, I.Isabel Papadimitriou, S.Salomey Osei, P.Pedro Ortiz Suarez, I.Iroro Orife, K.Kelechi Ogueji, R. A.Rubungo Andre Niyongabo, T. Q.Toan Q. Nguyen, M.Mathias Müller, A.André Müller, S. H.Shamsuddeen Hassan Muhammad, N.Nanda Muhammad, A.Ayanda Mnyakeni, J.Jamshidbek Mirzakhalov, T.Tapiwanashe Matangira, C.Colin Leong, N.Nze Lawson, S.Sneha Kudugunta, Y.Yacine Jernite, M.Mathias Jenny, O.Orhan Firat, B. F.Bonaventure F. P. Dossou, S.Sakhile Dlamini, N.Nisansa de Silva, S. Ç.Sakine Çabuk Balli, S.Stella Biderman, A.Alessia Battisti, A.Ahmed Baruwa, A.Ankur Bapna, P.Pallavi Baljekar, I. A.Israel Abebe Azime, A.Ayodele Awokoya, D.Duygu Ataman, O.Orevaoghene Ahia, O.Oghenefego Ahia, S.Sweta Agrawal and M.Mofetoluwa Adeyemi. Quality at a Glance: An Audit of Web-Crawled Multilingual Datasets.Transactions of the Association for Computational Linguistics10January 2022, 50-72
  HALDOIback to text
• 17 article T. A.Tu Anh Nguyen, B.Benoît Sagot and E.Emmanuel Dupoux. Are discrete units necessary for Spoken Language Modeling? IEEE Journal of Selected Topics in Signal Processing August 2022
  HAL back to text

National journals

• 18 articleA.Alix Chagué and L.Laurent Romary. Artificial Intelligence, openiing a field of possibilities.Arabesques107September 2022, 4-5
  HALDOI

International peer-reviewed conferences

• 19 inproceedingsJ.Julien Abadji, P.Pedro Ortiz Suarez, L.Laurent Romary and B.Benoît Sagot. Towards a Cleaner Document-Oriented Multilingual Crawled Corpus.Thirteenth Language Resources and Evaluation Conference - LREC 2022Proceedings of the Thirteenth Language Resources and Evaluation ConferenceMarseille, FranceJune 2022
  HALback to textback to text
• 20 inproceedings J. O.Jesujoba O Alabi, L.Lydia Nishimwe, B.Benjamin Muller, C.Camille Rey, B.Benoît Sagot and R.Rachel Bawden. Inria-ALMAnaCH at the WMT 2022 shared task: Does Transcription Help Cross-Script Machine Translation? Proceedings of the Seventh Conference on Machine Translation EMNLP 2022 - Seventh Conference on Machine Translation (WMT22 - Workshop on Statistical Machine Translation) Abu Dhabi, United Arab Emirates 2022
  HAL back to text
• 21 inproceedingsR.Robin Algayres, A.Adel Nabli, B.Benoît Sagot and E.Emmanuel Dupoux. Speech Sequence Embeddings using Nearest Neighbors Contrastive Learning.Interspeech 2022 - 23rd INTERSPEECH ConferenceIncheon, South KoreaSeptember 2022
  HALback to text
• 22 inproceedingsR.Rachel Bawden, J.Jonathan Poinhos, E.Eleni Kogkitsidou, P.Philippe Gambette, B.Benoît Sagot and S.Simon Gabay. Automatic Normalisation of Early Modern French.LREC 2022 - 13th Language Resources and Evaluation ConferenceMarseille, FranceJune 2022, 3354-3366
  HALDOIback to text
• 23 inproceedingsT.Thibault Charmet, I.Inès Cherichi, M.Matthieu Allain, U.Urszula Czerwinska, A.Amaury Fouret, B.Benoît Sagot and R.Rachel Bawden. Complex Labelling and Similarity Prediction in Legal Texts: Automatic Analysis of France's Court of Cassation Rulings.LREC 2022 - 13th Language Resources and Evaluation ConferenceMarseille, FranceJune 2022
  HALback to text
• 24 inproceedingsA.Anna Chepaikina, R.Robert Bossy, C.Catherine Roussey and S.Stephan Bernard. Thesaurus Enrichment via Coordination Extraction.16th International Conference on Metadata and Semantics Research (MTSR 2022)London, United KingdomNovember 2022
  HAL
• 25 inproceedingsP.-A.Paul-Ambroise Duquenne, H.Hongyu Gong, B.Benoît Sagot and H.Holger Schwenk. T-Modules: Translation Modules for Zero-Shot Cross-Modal Machine Translation.EMNLP 2022 - 2022 Conference on Empirical Methods in Natural Language ProcessingAbu Dhabi, United Arab EmiratesDecember 2022
  HALback to text
• 26 inproceedingsG.Ghazi Felhi, J.Joseph Le Roux and D.Djamé Seddah. Exploiting Inductive Bias in Transformers for Unsupervised Disentanglement of Syntax and Semantics with VAEs.Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language TechnologiesNAACL 2022 - Conference of the North American Chapter of the Association for Computational Linguistics: Human Language TechnologiesSeattle, United StatesAssociation for Computational LinguisticsJuly 2022, 5763-5776
  HALDOI
• 27 inproceedingsC.Clémentine Fourrier and S.Syrielle Montariol. Caveats of Measuring Semantic Change of Cognates and Borrowings using Multilingual Word Embeddings.LChange'22 - 3rd International Workshop on Computational Approaches to Historical Language Change 2022Dublin, IrelandMay 2022
  HAL
• 28 inproceedingsC.Clémentine Fourrier and B.Benoît Sagot. Probing Multilingual Cognate Prediction Models.ACL 2022 - Findings of the Association for Computational LinguisticsDublin, IrelandMay 2022
  HALback to text
• 29 inproceedingsS.Simon Gabay, R.Rachel Bawden, P.Philippe Gambette, J.Jonathan Poinhos, E.Eleni Kogkitsidou and B.Benoît Sagot. Le changement linguistique au XVIIe s. : nouvelles approches scriptométriques.Actes du 8e Congrès Mondial de Linguistique FrançaiseCMLF 2022 - 8e Congrès Mondial de Linguistique Française138SHS Web of conferencesOrléans, FranceEDP Sciences2022, 02006.1-14
  HALDOIback to text
• 30 inproceedingsS.Simon Gabay, P.Pedro Ortiz Suarez, A.Alexandre Bartz, A.Alix Chagué, R.Rachel Bawden, P.Philippe Gambette and B.Benoît Sagot. From FreEM to D'AlemBERT: a Large Corpus and a Language Model for Early Modern French.13th Language Resources and Evaluation Conference - LREC 2022Proceedings of the 13th Language Resources and Evaluation ConferenceMarseille, FranceJune 2022, 3367-3374
  HALback to textback to text
• 31 inproceedingsN.Nathan Godey, R.Roman Castagné, E.Eric Villemonte de La Clergerie and B.Benoît Sagot. MANTa: Efficient Gradient-Based Tokenization for Robust End-to-End Language Modeling.EMNLP 2022 - The 2022 Conference on Empirical Methods in Natural Language ProcessingAbu Dhabi, United Arab EmiratesDecember 2022
  HALback to text
• 32 inproceedingsO.Omer Goldman, F.Francesco Tinner, H.Hila Gonen, B.Benjamin Muller, V.Victoria Basmov, S.Shadrack Kirimi, L.Lydia Nishimwe, B.Benoît Sagot, D.Djamé Seddah, R.Reut Tsarfaty and D.Duygu Ataman. The MRL 2022 Shared Task on Multilingual Clause-level Morphology.1st Shared Task on Multilingual Clause-level MorphologyAbu Dhabi, United Arab EmiratesDecember 2022
  HAL
• 33 inproceedingsT.Tom Kocmi, R.Rachel Bawden, O.Ondřej Bojar, A.Anton Dvorkovich, C.Christian Federmann, M.Mark Fishel, T.Thamme Gowda, Y.Yvette Graham, R.Roman Grundkiewicz, B.Barry Haddow, R.Rebecca Knowles, P.Philipp Koehn, C.Christof Monz, M.Makoto Morishita, M.Masaaki Nagata, T.Toshiaki Nakazawa, M.Michal Novák, M.Martin Popel, M.Maja Popović and M.Mariya Shmatova. Findings of the 2022 Conference on Machine Translation (WMT22).Proceedings of the Seventh Conference on Machine Translation (WMT)WMT 2022 - Seventh Conference on Machine TranslationAbu Dhabi, United Arab Emirates2022
  HALback to text
• 34 inproceedingsL.Louis Martin, A.Angela Fan, É.Éric Villemonte de la Clergerie, A.Antoine Bordes and B.Benoît Sagot. MUSS: Multilingual Unsupervised Sentence Simplification by Mining Paraphrases.LREC 2022 - 13th Language Resources and Evaluation ConferenceMarseille, FranceJune 2022
  HAL
• 35 inproceedingsS.Syrielle Montariol, A.Arij Riabi and D.Djamé Seddah. Multilingual Auxiliary Tasks Training: Bridging the Gap between Languages for Zero-Shot Transfer of Hate Speech Detection Models.Findings of AACL 2022AACL-IJCNLP 2022 - 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language ProcessingOnline, FranceNovember 2022
  HALback to text
• 36 inproceedingsS.Syrielle Montariol, É.Étienne Simon, A.Arij Riabi and D.Djamé Seddah. Fine-tuning and Sampling Strategies for Multimodal Role Labeling of Entities under Class Imbalance.Proceedings of the Workshop on Combating Online Hostile Posts in Regional Languages during Emergency SituationsCONSTRAINT 2022 - Workshop on Combating Online Hostile Posts in Regional Languages during Emergency SituationsDublin, IrelandAssociation for Computational LinguisticsMay 2022, 55-65
  HALDOIback to text
• 37 inproceedingsM.Mariana Neves, A.Antonio Jimeno Yepes, A.Amy Siu, R.Roland Roller, P.Philippe Thomas, M.Maika Vicente Navarro, L.Lana Yeganova, D.Dina Wiemann, G. M.Giorgio Maria Di Nunzio, F.Federica Vezzani, C.Christel Gérardin, R.Rachel Bawden, D.Darryl Johan Estrada, S.Salvador Lima-López, E.Eulàlia Farré-Maduell, M.Martin Krallinger, C.Cristian Grozea and A.Aurélie Névéol. Findings of the WMT 2022 Biomedical Translation Shared Task: Monolingual Clinical Case Reports.Proceedings of the Seventh Conference on Machine TranslationWMT22 - Seventh Conference on Machine TranslationAbu Dhabi, United Arab Emirates2022, 694-723
  HALback to text
• 38 inproceedingsT. A.Tu Anh Nguyen, E.Eugene Kharitonov, J.Jade Copet, Y.Yossi Adi, W.-N.Wei-Ning Hsu, A.Ali Elkahky, P.Paden Tomasello, R.Robin Algayres, B.Benoît Sagot, A.Abdelrahman Mohamed and E.Emmanuel Dupoux. Generative Spoken Dialogue Language Modeling.SLT-2022 - IEEE Spoken Language Technology WorkshopDoha-Qatar, QatarFebruary 2023
  HAL
• 39 inproceedings A.Arij Riabi, B.Benoît Sagot and D.Djamé Seddah. Can Character-based Language Models Improve Downstream Task Performance in Low-Resource and Noisy Language Scenarios? Seventh Workshop on Noisy User-generated Text (W-NUT 2021, colocated with EMNLP 2021) Proceedings of the Seventh Workshop on Noisy User-generated Text (W-NUT 2021) Punta Cana, Dominican Republic January 2022
  HAL back to text
• 40 inproceedingsV.Victor Sanh, A.Albert Webson, C.Colin Raffel, S. H.Stephen H. Bach, L.Lintang Sutawika, Z.Zaid Alyafeai, A.Antoine Chaffin, A.Arnaud Stiegler, T.Teven Le Scao, A.Arun Raja, M.Manan Dey, M. S.M Saiful Bari, C.Canwen Xu, U.Urmish Thakker, S.Shanya Sharma, E.Eliza Szczechla, T.Taewoon Kim, G.Gunjan Chhablani, N.Nihal V. Nayak, D.Debajyoti Datta, J.Jonathan Chang, M.-J. T.Mike Tian-Jian Jiang, H.Han Wang, M.Matteo Manica, S.Sheng Shen, Z.-X.Zheng-Xin Yong, H.Harshit Pandey, M.Michael Mckenna, R.Rachel Bawden, T.Thomas Wang, T.Trishala Neeraj, J.Jos Rozen, A.Abheesht Sharma, A.Andrea Santilli, T.Thibault Fevry, J. A.Jason Alan Fries, R.Ryan Teehan, T.Tali Bers, S.Stella Biderman, L.Leo Gao, T.Thomas Wolf and A. M.Alexander M. Rush. Multitask Prompted Training Enables Zero-Shot Task Generalization.ICLR 2022 - Tenth International Conference on Learning RepresentationsOnline, Unknown Region2022
  HALback to text
• 41 inproceedingsF.Francesco de Toni, C.Christopher Akiki, J.Javier de La Rosa, C.Clémentine Fourrier, E.Enrique Manjavacas, S.Stefan Schweter and D.Daniel van Strien. Entities, Dates, and Languages: Zero-Shot on Historical Texts with T0.BigScience 2022 - International Workshop on Challenges & Perspectives in Creating Large Language Models 2022Dublin, IrelandMay 2022
  HALback to text

National peer-reviewed Conferences

• 42 inproceedingsS.Simon Gabay, P.Pedro Ortiz Suarez, R.Rachel Bawden, A.Alexandre Bartz, P.Philippe Gambette and B.Benoît Sagot. Le projet FREEM : ressources, outils et enjeux pour l’étude du français d’Ancien Régime.Actes de la 29e Conférence sur le Traitement Automatique des Langues Naturelles. Volume 1 : conférence principaleTALN 2022 - Traitement Automatique des Langues NaturellesAvignon, FranceATALA2022, 154-165
  HALback to text
• 43 inproceedingsB.Benjamin Muller, A.Antonios Anastasopoulos, B.Benoît Sagot and D.Djamé Seddah. When Being Unseen from mBERT is just the Beginning : Handling New Languages With Multilingual Language Models.Actes de la 29e Conférence sur le Traitement Automatique des Langues Naturelles. Volume 1 : conférence principaleTALN 2022 - 29° conférence sur le Traitement Automatique des Langues NaturellesAvignon, FranceATALA2022, 450-451
  HALback to text
• 44 inproceedingsA.Arij Riabi, S.Syrielle Montariol and D.Djamé Seddah. Tâches Auxiliaires Multilingues pour le Transfert de Modèles de Détection de Discours Haineux.Actes de la 29e Conférence sur le Traitement Automatique des Langues Naturelles. Volume 1 : conférence principaleTraitement Automatique des Langues NaturellesAvignon, FranceATALA2022, 413-423
  HALback to text

Conferences without proceedings

• 45 inproceedingsA.Alix Chagué and T.Thibault Clérice. Sharing HTR datasets with standardized metadata: the HTR-United initiative.Documents anciens et reconnaissance automatique des écritures manuscritesParis, FranceJune 2022
  HAL
• 46 inproceedingsA.Alix Chagué, H.Hugo Scheithauer, L.Lucas Terriel, F.Floriane Chiffoleau and Y.Yves Tadjo-Takianpi. Take a sip of TEI and relax: a proposition for an end-to-end workflow to enrich and publish data created with automatic text recognition.Digital Humanities 2022 : Responding to Asian DiversityTokyo, JapanJuly 2022
  HALback to text
• 47 inproceedingsF.Floriane Chiffoleau and H.Hugo Scheithauer. From a collection of documents to a published edition : how to use an end-to-end publication pipeline.TEI 2022 - Text Encoding Initiative 2022 ConferenceNewcastle, United KingdomSeptember 2022
  HALback to text
• 48 inproceedingsF.Florence Clavaud, L.Laurent Romary, P.Pauline Charbonnier, L.Lucas Terriel, G.Gaetano Piraino and V.Vincent Verdese. NER4Archives (named entity recognition for archives) : Conception et réalisation d’un outil de détection, de classification et de résolution des entités nommées dans les instruments de recherche archivistiques encodés en XML/EAD..Atelier Culture-INRIAPierrefitte sur Seine, FranceMarch 2022
  HALback to text
• 49 inproceedingsG.Ghazi Felhi, J.Joseph Le Roux and D.Djamé Seddah. Towards Unsupervised Content Disentanglement in Sentence Representations via Syntactic Roles.CtrlGen: Controllable Generative Modeling in Language and Visionvirtual, FranceJanuary 2022
  HAL
• 50 inproceedingsS.Simon Gabay, R.Rachel Bawden, B.Benoît Sagot and P.Philippe Gambette. Vers l’étude linguistique sur données artificielles: Le cas des systèmes graphiques en diachronie longue.Variation(s) en françaisNancy, FranceNovember 2022
  HALback to text
• 51 inproceedingsL.Loïc Grobol, M.Mathilde Regnault, P.Pedro Ortiz Suarez, B.Benoît Sagot, L.Laurent Romary and B.Benoît Crabbé. BERTrade: Using Contextual Embeddings to Parse Old French.Proceedings of the 13th Language Resources and Evaluation Conference13th Language Resources and Evaluation ConferenceMarseille, FranceJune 2022
  HAL
• 52 inproceedingsE. H.El Haff Karim, W.Wissam Antoun, F.Florence Le Ber and V.Véronique Pitchon. Reconnaissance des entités nommées pour l'analyse des pharmacopées médiévales.EGC 2023 - Extraction et Gestion des ConnaissancesLyon, FranceJanuary 2023
  HAL
• 53 inproceedingsA.Ariane Pinche, K.Kelly Christensen and S.Simon Gabay. Between automatic and manual encoding: Towards a generic TEI model for historical prints and manuscripts.TEI 2022 conference : Text as dataNewcastle, United KingdomSeptember 2022
  HALDOIback to text
• 54 inproceedingsL.Laurent Romary and H.Hugo Scheithauer. DataCatalogue : enjeux et réalisations.Un outil numérique pour interroger les catalogues de vente : le projet DataCatalogueParis, FranceOctober 2022
  HALback to text
• 55 inproceedingsA.Aurélia Rostaing and H.Hugo Scheithauer. Enrichir le patrimoine écrit archivistique grâce aux technologies numériques : Ingénierie du projet LectAuRep (Lecture automatique de répertoires).DHNord 2022 - Travailler en Humanités Numériques : collaborations, complémentarités et tensionsOnline, FranceJune 2022
  HALback to text
• 56 inproceedingsA.Aurélia Rostaing and H.Hugo Scheithauer. LectAuRep (2018-2021) :Projet de lecture automatique de répertoires de notaires.Segmenter et annoter les images : déconstruire pour reconstruireParis, FranceNovember 2022
  HALback to text
• 57 inproceedingsA.Aurélia Rostaing and H.Hugo Scheithauer. LectAuRep : Un projet de recherche et développement pour la transcription automatique de répertoires de notaires.La reconnaissance des écritures manuscrites et ses usages dans les archivesPierrefitte-sur-Seine, FranceNovember 2022
  HALback to text
• 58 inproceedingsY.Yves Rychener, X.Xavier Renard, D.Djamé Seddah, P.Pascal Frossard and M.Marcin Detyniecki. On the Granularity of Explanations in Model Agnostic NLP Interpretability.XKDD 2022 - ECML PKDD 2022 International Workshop on eXplainable Knowledge Discovery in Data MiningGrenoble, FranceSeptember 2022
  HAL
• 59 inproceedingsB.Benoît Sagot, L.Laurent Romary, R.Rachel Bawden, P. J.Pedro Javier Ortiz Suárez, K.Kelly Christensen, S.Simon Gabay, A.Ariane Pinche and J.-B.Jean-Baptiste Camps. Gallic(orpor)a : Extraction, annotation et diffusion de l’information textuelle et visuelle en diachronie longue: Restitution des travaux.DataLab de la BnF : Restitution des travaux 2022Paris, FranceDecember 2022
  HALback to text
• 60 inproceedingsH.Hugo Scheithauer. LectAuRep : Données d’archives en français des XIXe et XXe siècles.Transkribus / eScriptorium : Transcrire, annoter et éditer numériquement des documents d’archivesParis, FranceMay 2022
  HALback to text
• 61 inproceedingsH.Hugo Scheithauer, L.Laurent Romary, F.Frédérique Duyrat and F.Federico Nurra. Datacatalogue: project presentation.Atelier Culture-InriaPierrefitte-sur-Seine, FranceMarch 2022
  HALback to text
• 62 inproceedingsY.You Zuo, Y.Yixuan Li, A.Alma Parias García and K.Kim Gerdes. Technological taxonomies for hypernym and hyponym retrieval in patent texts.ToTh 2022 - Terminology & Ontology: Theories and applicationsChambéry, France2022
  HAL
• 63 inproceedingsY.You Zuo, H.Houda Mouzoun, S.Samir Ghamri Doudane, K.Kim Gerdes and B.Benoît Sagot. Patent Classification using Extreme Multi-label Learning: A Case Study of French Patents.SIGIR 2022 - PatentSemTech workshop - 3rd Workshop on Patent Text Mining and Semantic TechnologiesMadrid, Spain2022
  HALback to text

Scientific book chapters

• 64 inbookJ.Jack Bowers. Pathways and patterns of metaphor and metonymy in Mixtepec-Mixtec body-part terms.The Grammar of Body-Part Expressions: A view from the AmericasRoberto Zariquiey; Pilar M. Valenzuela2022, 91-135
  HALDOI
• 65 inbook A.Alix Chagué, V.Victoria Le Fourner, M.Manuela Martini and E.Eric Villemonte de La Clergerie. Deux siècles de sources disparates sur l’industrie textile en France : comment automatiser les traitements d’un corpus non-uniforme ? La fabrique numérique des corpus en sciences humaines et sociales Presses Universitaires du Septentrion; https://www.septentrion.com/livre/?GCOI=27574100990460 December 2022
  HAL back to text
• 66 inbookV.Victoria Le Fourner, A.Alix Chagué, M.Manuela Martini and A.Anaïs Albert. Structurer automatiquement un corpus homogène issu de la reconnaissance d’écriture manuscrite : les jugements du Conseil des prud’hommes des tissus parisiens.La fabrique numérique des corpus en sciences humaines et socialesPresses Universitaires du SeptentrionDecember 2022, https://www.septentrion.com/livre/?GCOI=27574100990460
  HALback to text

Doctoral dissertations and habilitation theses

• 67 thesisC.Clémentine Fourrier. Neural Approaches to Historical Word Reconstruction.Université PSL (Paris Sciences & Lettres)September 2022
  HALback to text
• 68 thesis B.Benjamin Muller. How Can We Make Language Models Better at Handling the Diversity and Variability of Natural Languages ? Sorbonne Université November 2022
  HAL back to text
• 69 thesisP.Pedro Ortiz Suarez. A Data-driven Approach to Natural Language Processing for Contemporary and Historical French.Sorbonne UniversitéJune 2022
  HALback to text

Reports & preprints

• 70 miscW.Wissam Antoun, B.Benoît Sagot and D.Djamé Seddah. Data-Efficient French Language Modeling with CamemBERTa.January 2023
  HAL
• 71 miscR.Rachel Bawden and F.François Yvon. Investigating the Translation Performance of a Large Multilingual Language Model: the Case of BLOOM.March 2023
  HALDOI
• 72 miscF.Floriane Chiffoleau and A.Anne Baillot. Le projet DAHN : une pipeline pour l'édition numérique de documents d'archives.April 2022
  HALback to text
• 73 miscT.Thibault Clérice, M.Malamatenia Vlachou-Efstathiou and A.Alix Chagué. CREMMA Medii Aevi: Literary manuscript text recognition in Latin.January 2023
  HAL
• 74 miscM.Matthieu Futeral, C.Cordelia Schmid, I.Ivan Laptev, B.Benoît Sagot and R.Rachel Bawden. Tackling Ambiguity with Images: Improved Multimodal Machine Translation and Contrastive Evaluation.December 2022
  HALback to text
• 75 miscY. L.Yu Lu Liu, R.Rachel Bawden, T.Thomas Scialom, B.Benoît Sagot and J. C.Jackie Chi Kit Cheung. MaskEval: Weighted MLM-Based Evaluation for Text Summarization and Simplification.October 2022
  HALback to text
• 76 miscA.Angelina Mcmillan-Major, Z.Zaid Alyafeai, S.Stella Biderman, K.Kimbo Chen, F.Francesco de Toni, G.Gérard Dupont, H.Hady Elsahar, C.Chris Emezue, A. F.Alham Fikri Aji, S.Suzana Ilić, N.Nurulaqilla Khamis, C.Colin Leong, M.Maraim Masoud, A.Aitor Soroa, P.Pedro Ortiz Suarez, Z.Zeerak Talat, D.Daniel van Strien and Y.Yacine Jernite. Documenting Geographically and Contextually Diverse Data Sources: The BigScience Catalogue of Language Data and Resources.February 2022
  HALback to text
• 77 miscS. J.Sabrina J. Mielke, Z.Zaid Alyafeai, E.Elizabeth Salesky, C.Colin Raffel, M.Manan Dey, M.Matthias Gallé, A.Arun Raja, C.Chenglei Si, W. Y.Wilson Y. Lee, B.Benoît Sagot and S.Samson Tan. Between words and characters: A Brief History of Open-Vocabulary Modeling and Tokenization in NLP.January 2022
  HALback to text
• 78 miscT. A.Tu Anh Nguyen, E.Eugene Kharitonov, J.Jade Copet, Y.Yossi Adi, W.-N.Wei-Ning Hsu, A.Ali Elkahky, P.Paden Tomasello, R.Robin Algayres, B.Benoît Sagot, A.Abdelrahman Mohamed and E.Emmanuel Dupoux. Generative Spoken Dialogue Language Modeling: preprint version.October 2022
  HALback to text
• 79 misc T. A.Tu Anh Nguyen, M. D.Maureen De Seyssel, R.Robin Algayres, P.Patricia Rozé, E.Ewan Dunbar and E.Emmanuel Dupoux. Are word boundaries useful for unsupervised language learning? 2022
  HAL DOI
• 80 miscT. L.Teven Le Scao, A.Angela Fan, C.Christopher Akiki, E.Ellie Pavlick, S.Suzana Ilić, D.Daniel Hesslow, R.Roman Castagné, A. S.Alexandra Sasha Luccioni, F.François Yvon, M.Matthias Gallé, J.Jonathan Tow, A. M.Alexander M. Rush, S.Stella Biderman, A.Albert Webson, P. S.Pawan Sasanka Ammanamanchi, T.Thomas Wang, B.Benoît Sagot, N.Niklas Muennighoff, A. V.Albert Villanova del Moral, O.Olatunji Ruwase, R.Rachel Bawden, S.Stas Bekman, A.Angelina Mcmillan-Major, I.Iz Beltagy, H.Huu Nguyen, L.Lucile Saulnier, S.Samson Tan, P.Pedro Ortiz Suarez, V.Victor Sanh, H.Hugo Laurençon, Y.Yacine Jernite, J.Julien Launay, M.Margaret Mitchell, C.Colin Raffel, A.Aaron Gokaslan, A.Adi Simhi, A.Aitor Soroa, A. F.Alham Fikri Aji, A.Amit Alfassy, A.Anna Rogers, A. K.Ariel Kreisberg Nitzav, C.Canwen Xu, C.Chenghao Mou, C.Chris Emezue, C.Christopher Klamm, C.Colin Leong, D.Daniel van Strien, D. I.David Ifeoluwa Adelani, D.Dragomir Radev, E. G.Eduardo González Ponferrada, E.Efrat Levkovizh, E.Ethan Kim, E. B.Eyal Bar Natan, F.Francesco de Toni, G.Gérard Dupont, G.Germán Kruszewski, G.Giada Pistilli, H.Hady Elsahar, H.Hamza Benyamina, H.Hieu Tran, I.Ian Yu, I.Idris Abdulmumin, I.Isaac Johnson, I.Itziar Gonzalez-Dios, J.Javier de la Rosa, J.Jenny Chim, J.Jesse Dodge, J.Jian Zhu, J.Jonathan Chang, J.Jörg Frohberg, J.Joseph Tobing, J.Joydeep Bhattacharjee, K.Khalid Almubarak, K.Kimbo Chen, K.Kyle Lo, L.Leandro von Werra, L.Leon Weber, L.Long Phan, L. B.Loubna Ben Allal, L.Ludovic Tanguy, M.Manan Dey, M. R.Manuel Romero Muñoz, M.Maraim Masoud, M.María Grandury, M.Mario Šaško, M.Max Huang, M.Maximin Coavoux, M.Mayank Singh, M.-J. T.Mike Tian-Jian Jiang, M. C.Minh Chien Vu, M. A.Mohammad A. Jauhar, M.Mustafa Ghaleb, N.Nishant Subramani, N.Nora Kassner, N.Nurulaqilla Khamis, O.Olivier Nguyen, O.Omar Espejel, O.Ona de Gibert, P.Paulo Villegas, P.Peter Henderson, P.Pierre Colombo, P.Priscilla Amuok, Q.Quentin Lhoest, R.Rheza Harliman, R.Rishi Bommasani, R. L.Roberto Luis López, R.Rui Ribeiro, S.Salomey Osei, S.Sampo Pyysalo, S.Sebastian Nagel, S.Shamik Bose, S. H.Shamsuddeen Hassan Muhammad, S.Shanya Sharma, S.Shayne Longpre, S.Somaieh Nikpoor, S.Stanislav Silberberg, S.Suhas Pai, S.Sydney Zink, T. T.Tiago Timponi Torrent, T.Timo Schick, T.Tristan Thrush, V.Valentin Danchev, V.Vassilina Nikoulina, V.Veronika Laippala, V.Violette Lepercq, V.Vrinda Prabhu, Z.Zaid Alyafeai, Z.Zeerak Talat, A.Arun Raja, B.Benjamin Heinzerling, C.Chenglei Si, E.Elizabeth Salesky, S. J.Sabrina J. Mielke, W. Y.Wilson Y. Lee, A.Abheesht Sharma, A.Andrea Santilli, A.Antoine Chaffin, A.Arnaud Stiegler, D.Debajyoti Datta, E.Eliza Szczechla, G.Gunjan Chhablani, H.Han Wang, H.Harshit Pandey, H.Hendrik Strobelt, J. A.Jason Alan Fries, J.Jos Rozen, L.Leo Gao, L.Lintang Sutawika, M. S.M Saiful Bari, M. S.Maged S. Al-Shaibani, M.Matteo Manica, N.Nihal Nayak, R.Ryan Teehan, S.Samuel Albanie, S.Sheng Shen, S.Srulik Ben-David, S. H.Stephen H. Bach, T.Taewoon Kim, T.Tali Bers, T.Thibault Fevry, T.Trishala Neeraj, U.Urmish Thakker, V.Vikas Raunak, X.Xiangru Tang, Z.-X.Zheng-Xin Yong, Z.Zhiqing Sun, S.Shaked Brody, Y.Yallow Uri, H.Hadar Tojarieh, A.Adam Roberts, H. W.Hyung Won Chung, J.Jaesung Tae, J.Jason Phang, O.Ofir Press, C.Conglong Li, D.Deepak Narayanan, H.Hatim Bourfoune, J.Jared Casper, J.Jeff Rasley, M.Max Ryabinin, M.Mayank Mishra, M.Minjia Zhang, M.Mohammad Shoeybi, M.Myriam Peyrounette, N.Nicolas Patry, N.Nouamane Tazi, O.Omar Sanseviero, P.Patrick von Platen, P.Pierre Cornette, P. F.Pierre François Lavallée, R.Rémi Lacroix, S.Samyam Rajbhandari, S.Sanchit Gandhi, S.Shaden Smith, S.Stéphane Requena, S.Suraj Patil, T.Tim Dettmers, A.Ahmed Baruwa, A.Amanpreet Singh, A.Anastasia Cheveleva, A.-L.Anne-Laure Ligozat, A.Arjun Subramonian, A.Aurélie Névéol, C.Charles Lovering, D.Dan Garrette, D.Deepak Tunuguntla, E.Ehud Reiter, E.Ekaterina Taktasheva, E.Ekaterina Voloshina, E.Eli Bogdanov, G. I.Genta Indra Winata, H.Hailey Schoelkopf, J.-C.Jan-Christoph Kalo, J.Jekaterina Novikova, J. Z.Jessica Zosa Forde, J.Jordan Clive, J.Jungo Kasai, K.Ken Kawamura, L.Liam Hazan, M.Marine Carpuat, M.Miruna Clinciu, N.Najoung Kim, N.Newton Cheng, O.Oleg Serikov, O.Omer Antverg, O.Oskar van der Wal, R.Rui Zhang, R.Ruochen Zhang, S.Sebastian Gehrmann, S.Shani Pais, T.Tatiana Shavrina, T.Thomas Scialom, T.Tian Yun, T.Tomasz Limisiewicz, V.Verena Rieser, V.Vitaly Protasov, V.Vladislav Mikhailov, Y.Yada Pruksachatkun, Y.Yonatan Belinkov, Z.Zachary Bamberger, Z.Zdeněk Kasner, A.Alice Rueda, A.Amanda Pestana, A.Amir Feizpour, A.Ammar Khan, A.Amy Faranak, A.Ana Santos, A.Anthony Hevia, A.Antigona Unldreaj, A.Arash Aghagol, A.Arezoo Abdollahi, A.Aycha Tammour, A.Azadeh Hajihosseini, B.Bahareh Behroozi, B.Benjamin Ajibade, B.Bharat Saxena, C. M.Carlos Muñoz Ferrandis, D.Danish Contractor, D.David Lansky, D.Davis David, D.Douwe Kiela, D. A.Duong A. Nguyen, E.Edward Tan, E.Emi Baylor, E.Ezinwanne Ozoani, F.Fatima Mirza, F.Frankline Ononiwu, H.Habib Rezanejad, H.Hessie Jones, I.Indrani Bhattacharya, I.Irene Solaiman, I.Irina Sedenko, I.Isar Nejadgholi, J.Jesse Passmore, J.Josh Seltzer, J. B.Julio Bonis Sanz, L.Livia Dutra, M.Mairon Samagaio, M.Maraim Elbadri, M.Margot Mieskes, M.Marissa Gerchick, M.Martha Akinlolu, M.Michael Mckenna, M.Mike Qiu, M.Muhammed Ghauri, M.Mykola Burynok, N.Nafis Abrar, N.Nazneen Rajani, N.Nour Elkott, N.Nour Fahmy, O.Olanrewaju Samuel, R.Ran An, R.Rasmus Kromann, R.Ryan Hao, S.Samira Alizadeh, S.Sarmad Shubber, S.Silas Wang, S.Sourav Roy, S.Sylvain Viguier, T.Thanh Le, T.Tobi Oyebade, T.Trieu Le, Y.Yoyo Yang, Z.Zach Nguyen, A. R.Abhinav Ramesh Kashyap, A.Alfredo Palasciano, A.Alison Callahan, A.Anima Shukla, A.Antonio Miranda-Escalada, A.Ayush Singh, B.Benjamin Beilharz, B.Bo Wang, C.Caio Brito, C.Chenxi Zhou, C.Chirag Jain, C.Chuxin Xu, C.Clémentine Fourrier, D. L.Daniel León Periñán, D.Daniel Molano, D.Dian Yu, E.Enrique Manjavacas, F.Fabio Barth, F.Florian Fuhrimann, G.Gabriel Altay, G.Giyaseddin Bayrak, G.Gully Burns, H. U.Helena U. Vrabec, I.Imane Bello, I.Ishani Dash, J.Jihyun Kang, J.John Giorgi, J.Jonas Golde, J. D.Jose David Posada, K. R.Karthik Rangasai Sivaraman, L.Lokesh Bulchandani, L.Lu Liu, L.Luisa Shinzato, M. H.Madeleine Hahn de Bykhovetz, M.Maiko Takeuchi, M.Marc Pàmies, M. A.Maria A Castillo, M.Marianna Nezhurina, M.Mario Sänger, M.Matthias Samwald, M.Michael Cullan, M.Michael Weinberg, M.Michiel de Wolf, M.Mina Mihaljcic, M.Minna Liu, M.Moritz Freidank, M.Myungsun Kang, N.Natasha Seelam, N.Nathan Dahlberg, N. M.Nicholas Michio Broad, N.Nikolaus Muellner, P.Pascale Fung, P.Patrick Haller, R.Ramya Chandrasekhar, R.Renata Eisenberg, R.Robert Martin, R.Rodrigo Canalli, R.Rosaline Su, R.Ruisi Su, S.Samuel Cahyawijaya, S.Samuele Garda, S. S.Shlok S Deshmukh, S.Shubhanshu Mishra, S.Sid Kiblawi, S.Simon Ott, S.Sinee Sang-Aroonsiri, S.Srishti Kumar, S.Stefan Schweter, S.Sushil Bharati, T.Tanmay Laud, T.Théo Gigant, T.Tomoya Kainuma, W.Wojciech Kusa, Y.Yanis Labrak, Y. S.Yash Shailesh Bajaj, Y.Yash Venkatraman, Y.Yifan Xu, Y.Yingxin Xu, Y.Yu Xu, Z.Zhe Tan, Z.Zhongli Xie, Z.Zifan Ye, M.Mathilde Bras, Y.Younes Belkada and T.Thomas Wolf. BLOOM: A 176B-Parameter Open-Access Multilingual Language Model.November 2022
  HALback to text
• 81 miscH.Hugo Scheithauer, A.Alix Chagué and L.Laurent Romary. Which TEI representation for the output of automatic transcriptions and their metadata? An illustrated proposition.April 2022
  HAL

Other scientific publications

• 82 inproceedingsA.Alix Chagué. Conditions de la mutualisation : les principes FAIR et HTR-United.Humanistica 2022Montréal, CanadaMay 2022
  HAL
• 83 miscA.Alix Chagué. Intelligence Artificielle et intelligence collective : des nouveaux eldorados pour rendre les textes patrimoniaux plus accessibles ?May 2022, 7 p.
  HALback to text
• 84 inproceedingsC. L.Cassandra L. Jacobs, A.Aniello De Santo and L.Loïc Grobol. Online and offline processing in zeugma constructions is insensitive to argument order.Human Sentence Processing 2023Pittsburg, United StatesMarch 2023
  HAL
• 85 miscB.Benoît Sagot, L.Laurent Romary, R.Rachel Bawden, P.Pedro Ortiz Suarez, K.Kelly Christensen, S.Simon Gabay, A.Ariane Pinche and J.-B.Jean-Baptiste Camps. Gallic(orpor)a: Extraction, annotation et diffusion de l'information textuelle et visuelle en diachronie longue: Restitution des travaux.December 2022
  HAL

Scientific popularization

• 86 inproceedingsA.Alix Chagué. Corpus, méthodes et ressources pour la transcription automatique des documents manuscrits patrimoniaux francophones contemporains.89e Congrès de l'Acfas, Section 310 - Le numérique dans les sciences humaines : édition et visualisationMontréal, CanadaMay 2022
  HAL
• 87 articleA.Alix Chagué. eScriptorium: a free application to automatically transcribe manuscripts.Arabesques107September 2022, 25
  HALDOI