The main objective of the CLD2025 project is to facilitate the urgent task of documenting endangered languages by leveraging the potential of computational methods. A breakthrough is now possible: machine learning tools (such as artificial neural networks and Bayesian models) have improved to a point where they can effectively help to perform linguistic annotation tasks such as automatic transcription of audio recordings, automatic glossing of texts, and automatic word discovery. Thorough documentation of the world’s dwindling linguistic diversity is much more feasible with these tools than under a manual workflow. For instance, manual transcription of 50 hours of speech (a sizeable fieldwork corpus) can take hundreds of hours’ work, creating a bottleneck in the language documentation workflow. Another key task, referred to in linguistics as interlinear glossing (in a nutshell: word-by-word translation/annotation), is even more time-consuming, and is moreover difficult to perform manually with the required level of consistency. Models created through machine learning have the potential to aid in these time-consuming and difficult tasks. But Natural Language Processing (NLP) remains little-used in language documentation for a variety of reasons such as that the technology is still new (and evolving rapidly), user-friendly interfaces are still under development, and there are few case studies demonstrating practical usefulness in a low-resource setting. Field linguists typically rely on manual methods throughout the documentation process. The objective of the CLD2025 project is therefore to enable the implementation of these techniques in the mid term (by 2025) by developing a co-construction of models and tools by field linguists and computational linguists, and the development of interfaces and systems that allow real use by field linguists. We are building on the achievements of the BULB project in terms of corpora and modes of acquisition, as well as the development of models for transcription and segmentation. We are not developing corpora here, but rather focusing on how to exploit existing corpora. We address automatic processing problems (phoneme and tone transcription, unit discovery, automatic glossing), some of which are original (tonal transcription, automatic glossing), by validating them on endangered languages of very varied natures: Bantu Mboshi C25, Mande Kakabe, a Sino-Tibetan language, Yongning Na (Mosuo), and 3 Nakh-Daghestanian languages, Khinalug, Kryz (Kryts), Budugh. We will perform work to leverage the results of the improved automatic processing to the linguistic work level: the automatic speech and language processing mechanisms and results will be used to explore phonetic-phonological issues on segmental, supra-segmental and tonal levels of the languages addressed in the project, Finally, from the beginning of the project, the focus will be on the usability of the tools and models developed. This point highlights the fundamentally interdisciplinary aspect of the work carried out here by computational scientists and field linguists. To do so, a recognized field linguist will work full-time on the project, and will participate, both through her experience and expertise in the definition, development and evaluation of the different systems developed in the project.

The proposal aims at developing tools for diagnostic, localization, and measurements of automatic transcription errors. This proposal is based on a consortium of academic actors of very first plan in this field. The objective is to study in detail (at the perceptive, acoustico-phonetics, lexical, and syntactic levels) the errors in order to bring a precise diagnosis of possible lacks of the current classical models on certain classes of linguistic phenomena. At the application level, the proposal is justified by an observation: a high number of applications in the field of content access from multimedia data are made possible by the use of automatic transcriptions of speech: subtitling of video emissions, search for precise extracts in audio-visual archives, automated reports of meetings, extraction of information and structuring of information (Speech Analytics) in contents multimedia (Web, call centers, ...). However their deployment on a large scale is often slowed down by the fact that transcription from automatic speech recognition systems contains too many errors. Research and development in speech recognition related, successfully until now, to the improvement of methods and models implemented in the process of transcription, measured thanks to the word error rate; however, last a certain performance level, the marginal cost induced to reduce the residual errors increases then exponentially. Transcription errors thus persist, which are more or less awkward according to the applications. Information retrieval is tolerant with errors (up to 30%), but systematic errors on certain named entities can be prohibitive. On the contrary, subtitling or meeting transcription have a very weak tolerance with the errors, and even very low word error rates compared to the state of the art (lower than 5%) are too high for the end-users. Error processing is not limited to increase the acceptability of the applications based on the automatic transcription of the word. Error classification, impact measurement by perceptive tests, error diagnosis state-of-the-art systems, are the first crucial stage in order to identify the lacks of the current models and to prepare the future Automatic Speech Recognition system generations. The proposal aims, by a close cooperation between complementary partners who excel in their field, to set up an infrastructure of detection, diagnosis, and qualitative measurement which makes it possible to create a virtuous circle of improvement of large and very large vocabulary continuous speech recognition systems.

Standard French /R/ is - like most rhotics - characterized by a great deal of variability. According to the literature, its differents realizations are mainly due to proprieties of contiguous phonemes and to its position in the word. These accounts leave unexplained a great deal of realizations labeled as "free variants". Phonetically, it could be realized most frequently as a fricative, a trill, an approximant, voiced, unvoiced or sometimes partially unvoiced. These troubles in understanding variations of French /R/ imply a number of consequences : learners of French have great difficulties with the acquisition of this sound, Automatic Speech Recognition systems - even though they manage to modelize this sound as a function of its contiguous sounds - use pronunciation dictionaries that are overly simplified for words containing /R/. As for phonological theories, they usually consider it as specific because of, among other reasons, its propensity to fluctuate between obstruents and sonorants. In the project REPER, we hypothesize that articulatory variation is structured according to a continuum going from voiced approximant to voiceless fricative. Realizations outside this continuum can be explained by a series of prosodic and semantic factors mostly. These different realizations and great sensibility to reduction compared to other phonemes suggests a possible on-going evolution of French /R/. An acoustic target (rather than articulatory) seems to be the most important point of stability across these realizations. This acoustic target is found in formant trajectories of contiguous vowels : a rise in first and third formants. We will investigate the status of French /R/ in three working packages that complete each other and allow a complete comprehension of the phenomenon : - A first working package dealing with acoustic analyses of large corpora of approximately a hunded hours of continuous speech (also comparing two styles: journalistic and spontaneous speech, and comparing journalistic french to a corpus of journalistic german in order to evaluate the variation of /R/ in both languages as these two phonemes have some similtudes). All corpora are already available and have been tagged for analysis by the coordinator. Large corpora of continuous speech - and them alone - allow to quantify the different realizations of /R/, the amplitude of its variation along with the number of strategies and their usage. They will also enable to clarify the sources of variations only available in large continuous speech studies such as the grammatical, semantic, pragmatic, syntacic and phonotactic context of /R/. - The second working package is a multi-sensor physiological analysis (using a controlled corpus) so as to analyze the articulatory gestures of friction (through ultrasound device and piezoelectric accelerometer) and voicing (through the non invasive External PhottoGlottoGraph recently patented by the Laboratory of Phonetics and Phonology) whether they go along that hypothesized continuum going from voiced approximant to voiceless fricative. A comparison between the acoustic target of F1 and F3 raising and the articulatory variability will be discussed This device ready to use in our laboratory will help determine the intended articulatory gestures and their timing in an articulatory phonology framework. Compared to French liquid /l/, the fricative /Z/ and the plosive /g/, we will compared their voicing and devoicing pattern, as well their approximant realization in specific contexts. - The third working package deals with perception and will explore the representation of the French /R/ in the brain. Firstly, we will test on French naive listeners the identification of graphical /R/ despite its numerous realization variants. In a second step, we will try and show their ability to discriminate between allophones of /R/ with classical behavioural perception tests and with an Electro-Encephalo-Graphic experiment.

VoxCrim deals with speaker authentication and more precisely forensic voice comparison. VoxCrim targets national security and legal/justice application contexts. The project's aims correspond to "ANR Défi 9". VoxCrim proposes a validated scientific objective framework available for all kinds of forensic voice comparison methods (automatic or phonetic). The goal is to develop certified standards to delimit the specific areas where voice comparison methods are applicable. VoxCrim gathers computer scientists (LIA) with phoneticians (LPL, LPP), experts in standardization (LNE) and members of the forensic department of Police Nationale (SDPTS). Members of the forensics department of Gendarmerie Nationale will take part in the project as a center of expertise. VoxCrim's aims are organised along two time scales: (1) analyses of well-controlled recordings in the near future, and (2) an enlarged area of application later on. (1) VoxCrim proposes to develop methods and databases based on the results of ANR Fabiole in order to extend the innovative concept of "box rule" (i.e. a set of conditions where voice comparison is perfectly applicable and where certification is possible). (2) Then, the project aims to add dimensions to the standards of voice comparison, such as the influence of the socio-cultural and linguistic environments. It is necessary to add these characteristics of the voice because they are closely linked to the real-life context police services are faced with. For this purpose, a database (PTVVox, with two recordings conditions i.e micro and GSM conditions) recorded by the SDPTS has been created and will be exploited. Two types of complementary analyses will be conducted: a) an acoustic analysis where relevant cues will be extracted for voice characterization and b) perceptual experiments where the ability of listeners to discriminate voices will be tested. Acoustic analyses will contribute to delimit inter and intra speaker variability in order to test the robustness of these cues. Perceptual experiments will verify if the acoustic cues previously highlighted are actually used by listeners to identify speakers. These experiments will be conducted particularly in GSM condition (corresponding to the most common case in forensic cases). This project will allow the SDPTS labs to extend their ISO 17025 validation approaches to voice comparison. In order to disseminate the knowledge and the questions about voice comparison, seminars will be organized with the speech community and members of the judicial system. The project will also help to gather – and provide training for – a pool of young researchers specializing in voice comparison and the forensic applications thereof in order to compensate for a well-known lack of specialists in France.

ALMAS (the Arabic word for ‘diamond’), is an international and interdisciplinary consortium composed of excellent scientific partners based in different countries of Europe and America and aiming at renewing the study of the living and extinct languages of South Arabia (Oman, southwestern Saudi Arabia, Yemen). All these languages belong to the Semitic family and fall into three ‘fields’, which have up to now corresponded largely to separate academic traditions with few interconnections: a set of four ‘Ancient South Arabian’ languages (ASA), now extinct (Sabaic, Qatabanic, Minaic, Hadramitic); a group of six living ‘Modern South Arabian’ (MSA) languages with no written tradition, all endangered (Mehri, Harsusi, Bathari, Hobyot, Jibbali, Soqotri); a rich array of highly diversified and archaic Arabic vernaculars spoken throughout the region (‘South Arabian Arabic’: SAA). In order to disentangle the complex linguistic landscape of South Arabia, ALMAS has been designed around seven scientific tasks: three field-specific tasks concern the investigation of the individual languages (ASA, MSA, SAA); four transversal tasks are organized according to different lines of action: contact linguistics, phonetics and phonology, morphology and Semitic comparison. ALMAS will apply a threefold scientific methodology relying on (1) the already available original sources, (2) the collection of new data and (3) the IT treatment of both these groups of sources. First, a systematic re-examination of the literature and of the text critical editions will be undertaken, which is an essential pre-requisite to any research. Secondly, ALMAS will conduct regular fieldwork seasons in two of the modern countries of South Arabia, in order to collect new epigraphic and oral data. Finally, ALMAS will treat all relevant data through the most modern IT tools for digitization and exploitation of linguistic data. The novelty of ALMAS lies in the interdisciplinary synergy it creates in three ways: by cross-fertilizing synchronic and diachronic approaches to the abovementioned languages; by stimulating contacts between researchers of the three ‘fields’ of the project; by developing complementarity between linguists from different schools and approaches. ALMAS will set a landmark in the domain, by documenting the languages through fieldwork and creating a digitally-enhanced open-access database, thus contributing to the study and the protection of these languages; analyzing the data in order to reach an adequate understanding of the languages’ structures; reevaluating the relationships between the languages (phylogenetic relatedness and/or language contact). Moreover, as ALMAS focuses on endangered or understudied languages, it will have a vast socio-cultural impact In particular, ALMAS will help preserve linguistic heritage; foster cultural awareness among speakers of minority languages; influence educational policies and practices. In order to support the efforts of the consortium and all its manifold activities, the project includes dedicated structures for the scientific and administrative management of the network. A technical task intends to develop a web-based system to manage, analyze and share the study material and the scientific results of the project. Finally, ALMAS results will be disseminated through a diversified dissemination and exploitation plan addressed to different settings in the Western world and in the Arab countries (scientific milieu, governments, education, broad public).