Natural Language processing

The Idiap NLP group studies how semantic and discourse processing of texts and dialogues can improve statistical machine translation and information indexing, specifically multimedia retrieval and recommendation. In statistical machine translation (SMT), we have shown that automatic labeling of discourse-level information (such as the senses of discourse connectives or the tense/aspects of verbs) can be combined with phrase-based SMT systems to improve their accuracy. In multimedia retrieval, we have developed a system for just-in-time recommendation of documents in conversations, and showed that topic-aware keyword extraction and semantic search improve its accuracy. Moreover, we have proposed methods for multimedia recommendation, applied to lectures, TV shows, and their segments, and showed that leveraging user comments can improve collaborative filtering. Moreover, searching over networked data was improved by using linguistic information and information from the network itself.

Video Presentation

Current Group Members

POPESCU-BELIS, Andrei
(Senior Researcher)
- website


PAPPAS, Nikolaos
(Postdoctoral Researcher)
- website


HONNET, Pierre-Edouard
(Postdoctoral Researcher)
- website


PU, Xiao
(Research Assistant)
- website


MICULICICH, Lesly (Sadiht)
(Research Assistant)
- website


MRINI, Khalil
(Trainee)
- website


Alumni

Current Projects

Recent Projects

Group News

Idiap has a new opening for an Internship position in Natural Language Processing on DNN-based coreference models
education — Dec 22, 2016

Co-reference is the relation between words or phrases in a text that refer to the same entity. DNNs have been successfully applied to a variety of NLP tasks, but for coreference resolution such approaches have shown limited improvement and a suitable model remains to be found. The goal of this internship is to go beyond classifiers that decide whether a pair of phrases is co-referent or not, and instead learn to represent, in the output layer, each of the entities of a text. In collaboration with a PhD student and a postdoc, the model will be applied to document-level machine translation. The work will thus relate to the EU SUMMA and SNSF MODERN projects.

Best Multimodal Paper Award to Nikolaos Pappas and co-authors at ICMR 2016
research — Jun 10, 2016

The paper "Multilingual Visual Sentiment Concept Matching" by Nikolaos Pappas, Mercan Topkara, Miriam Redi, Brendan Jou, Tao Chen, Hongyi Liu, Shih-Fu Chang has received the Best Multimodal Paper Award at the Annual ACM International Conference on Multimedia Retrieval (ICMR), held on June 6-9 in New York.

Idiap submission to MediaEval 2013 ranked first for the video hyperlinking task
research — Nov 26, 2013

The Idiap NLP group has participated in the search and hyperlinking task at the MediaEval 2013 evaluation campaign. The NLP group was ranked first out of eleven participants on the hyperlinking task (finding video segments related to a given one) and third out of seven on the keyword video search task.

Idiap submission to MediaEval 2013 ranked first for the video hyperlinking task
research — Nov 26, 2013

The Idiap NLP group has participated in the search and hyperlinking task at the MediaEval 2013 evaluation campaign. The NLP group was ranked first out of eleven participants on the hyperlinking task (finding video segments related to a given one) and third out of seven on the keyword video search task.