Attention, les informations que vous consultez actuellement ne sont pas celles de l'année universitaire en cours. Consulter l'année universitaire 2023-2024.
UE232 - Algorithmes pour le traitement automatique de la parole et du langage
Lieu et planning
-
Autre lieu Paris
ENS, 29 rue d’Ulm 75005 Paris
1er semestre / hebdomadaire, vendredi 16:00-19:00
du 21 janvier 2022 au 1er avril 2022
Nombre de séances : 11
Description
Dernière modification : 29 septembre 2021 07:03
- Type d'UE
- Enseignements fondamentaux de master
- Disciplines
- Méthodes et techniques des sciences sociales
- Page web
- https://github.com/edupoux/MVA_2021_SL
- Langues
- anglais
- Mots-clés
- Intelligence artificielle Linguistique
- Aires culturelles
- -
Intervenant·e·s
- Emmanuel Dupoux [référent·e] directeur d'études, EHESS / Laboratoire de sciences cognitives et psycholinguistiques (LSCP)
Les étudiant·es de l'EHESS qui souhaitent valider cet enseignement devront y assister en présentiel.
Speech and natural language processing is a subfield of artificial intelligence used in an increasing number of technological applications but also scientific applications in in digital humanities, medical science and behavioral science. Yet, while some aspects are on par with human performances, others are lagging behind. This course will present the full stack of speech and language technology, from automatic speech recognition to parsing and semantic processing. The course will present, at each level, the key principles, algorithms and mathematical principles behind the state of the art, and confront them with what is know about human speech and language processing. Students will acquire detailed knowledge of the scientific issues and computational techniques in automatic speech and language processing and will have hands on experience in implementing and evaluating the important algorithms.
Topics:
- speech features & signal processing
- hidden markov & finite state modeling
- probabilistic parsing
- continuous embeddings
- deep learning for language-related tasks (DNNs, RNNs)
- linguistics and psycholinguistics
- comparing human and machine performance
Program:
- Introduction
- ASR1: Features and Acoustic Models
- ASR2: Language Models + presentation TD#1
- ASR3:
- NLP1: Language processing in the wild
- NLP2: Formal languages
- NLP3: Parsing + presentation TD#2
- Automatic Translation
- Chatbots and open issue
Validation:
The validation is in two parts:
On-line QUIZZ (40% of the total grade). This is the only part of the course where you are absolutely required to be connected on-line. You'll be given a link of a google form which will be activated exactly at 11:00am and closed down at 11:30am. Any forms submitted after the deadline will be automatically rejected, and graded as zero. The QUIZZES will contain comprehension questions and the best 5 grades out of the 6 quizzes will be used for the average. Between 11:30 and 12:00 there will be a Q&A period where you'll be able to ask questions about the course and QUIZZ using an on-line connection.
Project (60% of the total grade). You'll work in small groups of 2-4 around a recent paper in speech or language processing which has already some existing code. Your task will be 1. to replicate the main result of the paper 2. run a experiment testing a new question not tested in the paper. You'll present your plan in a one page document in week #3, and your results in a 4 pages documend and 10 minutes oral presentation + 5 minutes questions in week #10.
The list of possible projects is here: https://docs.google.com/spreadsheets/d/115ZIe9V0Y-bbaf40KHEjRobgTuqPk1-VNEPC88fleKg/edit#gid=0
ATTENTION: since there is no "exam", there is no possibility of "rattrapage" (ie, of compensating a bad mark by taking another exam). So, if the overall grade obtained in this course is less than 10/20, this course will not be considered validated by the MVA Master.
______________________________
Details for the QUIZZ:
The QUIZZ will be composed of comprehension questions regarding the course you've just watched. You will have 30 minutes to complete the Google form which will be activated at 11:00am and closed at 11:30am each week with a quizz session.
Details for the PROJECT:
You will be given a list of papers to choose from.
Your first task (week #1-#3) is to select a paper of interest, and make up a group of 2-4 people to work on this paper.
Your second task (week #3) will be to decide on a plan for the experiment you'd like to run and write a 1p document describing what you want to do and who will do what. Attention! any delay in submitting your paper will cost you points (1/24th of a point for each hour of delay after sunday midnight before the monday of week #4).
Your third task will be to conduct the work and prepare (1) a written document (4 p max) describing what you've done and the main results. You may differ from the 1p, but will have to explain how and why. The 4 p should also contain a statement of contribution (who did what), (2) and oral 10 minutes presentation. (there will be 5 minutes of question aferwards).
Master
-
Méthodologie
– Sciences cognitives
– M2/S3
Suivi et validation – semestriel hebdomadaire = 6 ECTS
MCC – contrôle continu, travaux pratiques
Renseignements
- Contacts additionnels
- -
- Informations pratiques
le cours a lieu à l'ENS et/ou en ligne. Ce cours est mutualisé avec le master MVA de l'ENS Cachan ; peuvent le valider également les étudiants du master MASH (Dauphine) et du master Data Science (Polytechnique).
- Direction de travaux des étudiants
- -
- Réception des candidats
- -
- Pré-requis
Basic linear algebra, calculus, probability theory.
Compte rendu
Le traitement de la parole et du langage naturel est un sous-domaine de l'intelligence artificielle utilisé dans un nombre croissant d'applications ; Pourtant, si dans certaines tâches, les systèmes artificiels sont à la hauteur des performances humaines, pour d'autres, c’est encore loin d’être le cas. Ce cours présente l'ensemble des technologies de la parole et du langage, de la reconnaissance automatique de la parole jusqu’au traitement sémantique. Le cours présente, à chaque niveau linguistique, les principes clés, les algorithmes et les principes mathématiques à l'origine de l'état de l'art, et les confronte à ce que l'on sait sur le traitement de la parole et du langage humain. Les étudiants acquièrent une connaissance détaillée des problèmes scientifiques et des techniques de calcul dans le traitement automatique de la parole et du langage et ont une expérience pratique de la mise en œuvre et de l'évaluation des algorithmes dans un mini-projet.
Les sujets :
- Représentations de la parole et traitement du signal,
- Chaînes de markov caché et modélisation par automates à états finis,
- Modèles probabiliste et apprentissage profond pour la parole et le langage (réseaux convolutionnels, récurrents, transformers),
- Représentation des mots,
- Reconnaissance automatique de la parole, traduction automatique, systèmes de dialogue.
Dernière modification : 29 septembre 2021 07:03
- Type d'UE
- Enseignements fondamentaux de master
- Disciplines
- Méthodes et techniques des sciences sociales
- Page web
- https://github.com/edupoux/MVA_2021_SL
- Langues
- anglais
- Mots-clés
- Intelligence artificielle Linguistique
- Aires culturelles
- -
Intervenant·e·s
- Emmanuel Dupoux [référent·e] directeur d'études, EHESS / Laboratoire de sciences cognitives et psycholinguistiques (LSCP)
Les étudiant·es de l'EHESS qui souhaitent valider cet enseignement devront y assister en présentiel.
Speech and natural language processing is a subfield of artificial intelligence used in an increasing number of technological applications but also scientific applications in in digital humanities, medical science and behavioral science. Yet, while some aspects are on par with human performances, others are lagging behind. This course will present the full stack of speech and language technology, from automatic speech recognition to parsing and semantic processing. The course will present, at each level, the key principles, algorithms and mathematical principles behind the state of the art, and confront them with what is know about human speech and language processing. Students will acquire detailed knowledge of the scientific issues and computational techniques in automatic speech and language processing and will have hands on experience in implementing and evaluating the important algorithms.
Topics:
- speech features & signal processing
- hidden markov & finite state modeling
- probabilistic parsing
- continuous embeddings
- deep learning for language-related tasks (DNNs, RNNs)
- linguistics and psycholinguistics
- comparing human and machine performance
Program:
- Introduction
- ASR1: Features and Acoustic Models
- ASR2: Language Models + presentation TD#1
- ASR3:
- NLP1: Language processing in the wild
- NLP2: Formal languages
- NLP3: Parsing + presentation TD#2
- Automatic Translation
- Chatbots and open issue
Validation:
The validation is in two parts:
On-line QUIZZ (40% of the total grade). This is the only part of the course where you are absolutely required to be connected on-line. You'll be given a link of a google form which will be activated exactly at 11:00am and closed down at 11:30am. Any forms submitted after the deadline will be automatically rejected, and graded as zero. The QUIZZES will contain comprehension questions and the best 5 grades out of the 6 quizzes will be used for the average. Between 11:30 and 12:00 there will be a Q&A period where you'll be able to ask questions about the course and QUIZZ using an on-line connection.
Project (60% of the total grade). You'll work in small groups of 2-4 around a recent paper in speech or language processing which has already some existing code. Your task will be 1. to replicate the main result of the paper 2. run a experiment testing a new question not tested in the paper. You'll present your plan in a one page document in week #3, and your results in a 4 pages documend and 10 minutes oral presentation + 5 minutes questions in week #10.
The list of possible projects is here: https://docs.google.com/spreadsheets/d/115ZIe9V0Y-bbaf40KHEjRobgTuqPk1-VNEPC88fleKg/edit#gid=0
ATTENTION: since there is no "exam", there is no possibility of "rattrapage" (ie, of compensating a bad mark by taking another exam). So, if the overall grade obtained in this course is less than 10/20, this course will not be considered validated by the MVA Master.
______________________________
Details for the QUIZZ:
The QUIZZ will be composed of comprehension questions regarding the course you've just watched. You will have 30 minutes to complete the Google form which will be activated at 11:00am and closed at 11:30am each week with a quizz session.
Details for the PROJECT:
You will be given a list of papers to choose from.
Your first task (week #1-#3) is to select a paper of interest, and make up a group of 2-4 people to work on this paper.
Your second task (week #3) will be to decide on a plan for the experiment you'd like to run and write a 1p document describing what you want to do and who will do what. Attention! any delay in submitting your paper will cost you points (1/24th of a point for each hour of delay after sunday midnight before the monday of week #4).
Your third task will be to conduct the work and prepare (1) a written document (4 p max) describing what you've done and the main results. You may differ from the 1p, but will have to explain how and why. The 4 p should also contain a statement of contribution (who did what), (2) and oral 10 minutes presentation. (there will be 5 minutes of question aferwards).
-
Méthodologie
– Sciences cognitives
– M2/S3
Suivi et validation – semestriel hebdomadaire = 6 ECTS
MCC – contrôle continu, travaux pratiques
- Contacts additionnels
- -
- Informations pratiques
le cours a lieu à l'ENS et/ou en ligne. Ce cours est mutualisé avec le master MVA de l'ENS Cachan ; peuvent le valider également les étudiants du master MASH (Dauphine) et du master Data Science (Polytechnique).
- Direction de travaux des étudiants
- -
- Réception des candidats
- -
- Pré-requis
Basic linear algebra, calculus, probability theory.
-
Autre lieu Paris
ENS, 29 rue d’Ulm 75005 Paris
1er semestre / hebdomadaire, vendredi 16:00-19:00
du 21 janvier 2022 au 1er avril 2022
Nombre de séances : 11
Le traitement de la parole et du langage naturel est un sous-domaine de l'intelligence artificielle utilisé dans un nombre croissant d'applications ; Pourtant, si dans certaines tâches, les systèmes artificiels sont à la hauteur des performances humaines, pour d'autres, c’est encore loin d’être le cas. Ce cours présente l'ensemble des technologies de la parole et du langage, de la reconnaissance automatique de la parole jusqu’au traitement sémantique. Le cours présente, à chaque niveau linguistique, les principes clés, les algorithmes et les principes mathématiques à l'origine de l'état de l'art, et les confronte à ce que l'on sait sur le traitement de la parole et du langage humain. Les étudiants acquièrent une connaissance détaillée des problèmes scientifiques et des techniques de calcul dans le traitement automatique de la parole et du langage et ont une expérience pratique de la mise en œuvre et de l'évaluation des algorithmes dans un mini-projet.
Les sujets :
- Représentations de la parole et traitement du signal,
- Chaînes de markov caché et modélisation par automates à états finis,
- Modèles probabiliste et apprentissage profond pour la parole et le langage (réseaux convolutionnels, récurrents, transformers),
- Représentation des mots,
- Reconnaissance automatique de la parole, traduction automatique, systèmes de dialogue.