Tutelle du CNRSCMLD


  • Lattice
  • Transfers

Organismes tutelles

  • CNRS
  • ENS
  • PSL

Computational Methods for Endangered Language Documentation and Description

February 1st-2nd, 2018

Ecole Normale Supérieure, 45 rue d’Ulm, 75005 Paris

Organizers: Thierry Poibeau, Michael Rießler, Niko Partanen, Benjamin Fagard


There is a significant gap between digital methods applied in corpus building and corpus exploration for the numerous small and often endangered, low-resource languages compared to the high-resource majority languages. Corpora for endangered minority languages are typically built out of spoken data, which have first to be recorded and transcribed and are therefore relatively small. Majority language corpora, on the other hand are considerably bigger and include predominantly language data from diverse digital (or digitalized) written sources.

Whereas majority language corpus linguists develop and apply Natural Language Processing tools and attempt to automatize the annotation process, usually with the help of manually checked gold corpus, field linguists most typically rely on manual (or occasionally semi-manual) methods during the entire process. In many cases of fieldwork-based endangered language documentation projects, manual methods are in fact the most convenient choice, rather than to start developing computational linguistic resources from scratch. This is especially true if the linguistic structures of the languages in question are yet unknown, there is no established writing system, and the available corpus data are finite and small in quantity.

However, there are also many small or medium-size endangered languages for which the basic grammatical structures have already been described and which have established writing systems. This situation is common in Northern Eurasia, where basically all minority languages are also written today. Still, most of these languages have not been in the focus of computational and corpus linguistic research so far. This is true despite the fact that there are written corpus data of significant size available for several of these languages.

The workshop aims at examining the application of specific methods from Natural Language Processing in order to analyze data from endangered and low-resource languages from Northern Eurasia and other parts of the world. The workshop defines language technologies in a very broad sense and therefore includes also computational methods for signal processing in general, as such technologies can be applied effectively to the work with text corpora linked to multimedia data.

Invited speakers

Invited speakers (partial list, to be completed) include Joakim Nivre, Francis Tyers, Trond Tosterud, Svetlana Toldova, Olga Majewska and Jargal Badagarov.


This work has received support of TransferS (laboratoire d’excellence, program “Investissements d’avenir” ANR-10-IDEX-0001-02 PSL* and ANR-10-LABX-0099).

This workshoop is held in coordination with the CNRS-funded programme PRC n° 1052 : Analyse systématique du groupe nominal dans les langues finno-ougriennes : continuité référentielle et codage de la structure informationnelle de l’énoncé.

Haut de page