Once Upon a Time, Linguistics, AI and human needs - Maud Pironneau Cheffe de projet linguistique Druide informatique
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Once Upon a Time, Linguistics, AI and human needs 27 years of experience in Natural Language Processing Maud Pironneau Che e de projet linguistique Druide informatique www.bestppt.com ff
Overview Who I am ? Who is Druide ? What is Antidote ? How Antidote was made ? French and English : same approach ? What is the linguist’s place in this context ? www.bestppt.com
Druide informatique Fondation WebElixir Mobile App English Antidote + Service 1993 2006 2008 2015 2020 Developing and Website quality Edition tailored A rich set of Simple and marketing writing assurance specifically for the dictionaries and consolidated way of assistance software, reduced size and guides as well as a using its writing in French touchscreen interface powerful corrector assistance software of Apple’s iPhone on any platform Founded in 1993, Druide informatique is a Quebec-based company that specializes in developing and marketing writing assistance software. Its subsidiary, Éditions Druide, publishes French-language works of fiction, non-fiction, children’s literature and reference. The group's permanent team of more than 80 employees is complemented by a large network of contributors and partners. 3
Antidote Multilingual English and French Corrector With smart filters Everyone who uses French or English All over the world Dictionaries With powerful search tools Education role Understand what’s going on Guides Clear and detailed Role of referenc You can count on us 4 e
Context Domain : AI in Language Correction : Meeting point between Linguistics and Grammar A product born in 1993 A product made in Québec : a specific linguistic area Many devoted and attentive clients Human needs www.bestppt.com
our lexicon, our dictionaries, our guide How to use it ? s
What’s next Combinations dictionary
Easy to correct ?
Our Expérience With UdeM : Scorali project (2010) Classification : a correction is good or is not. Linguist Work Annotation Work of NLP scientist Have good Mesure F Did it work ? The problem was not the numbers. It was how to explain choices.
A New World : English. Language models and Classification Algorithms No Tables Supervised learning No semantic and grammatical features specificity added Correction by hypothesis : linguists All functionalities American English, British English and Canadian English And translations www.bestppt.com
French and English https://rocketchat.druide/ le-upload/wknLht2KwGFNcMntc/ Accord%20du%20participe%20passe%CC%81.png Two ways of thinking : The rule and the usage. https://rocketchat.druide/ le-upload/wknLh Ex. Code civil/Common law Accord%20du%20participe%20passe% fi fi
Language and Politics in French Création de Liberté de la Langue Création de l’OQL Charte de la langue Rectifications de l’Académie français Françoise, dans sa 1961 français 1990 1635 pureté 1977 1651 e F e
Linguist’s place : creating corpus and annotating it? Corpus : you are looking at the past. How to keep your language alive ? « It is in the word that the germ of all changes is found », Ferdinand de Saussure. What are we looking for? Respect the variations. • Ability • Gender (pronoun agreement) • Register • Regional language variety • Kinds of domains, etc. Models : how to keep our users trust ? How teach and explain ?
Linguist needed
Antidote Merci !
You can also read