A5071505335- Natural Language Processing Techniques
- Topic Modeling
- Speech and dialogue systems
- Speech Recognition and Synthesis
- Advanced Text Analysis Techniques
- Multimodal Machine Learning Applications
- Video Analysis and Summarization
- Speech and Audio Processing
- Handwritten Text Recognition Techniques
- Sentiment Analysis and Opinion Mining
- Music and Audio Processing
- Text Readability and Simplification
- AI in Service Interactions
- Linguistics and Discourse Analysis
- Face recognition and analysis
- Image Retrieval and Classification Techniques
- Advanced Image and Video Retrieval Techniques
- Language, Metaphor, and Cognition
- Authorship Attribution and Profiling
- Text and Document Classification Technologies
- French Language Learning Methods
- Language Development and Disorders
- Semantic Web and Ontologies
- Language, Discourse, Communication Strategies
- Face and Expression Recognition
Aix-Marseille Université
2014-2024
Centre National de la Recherche Scientifique
2012-2024
Laboratoire d’Informatique Fondamentale de Marseille
2011-2024
Laboratoire d’Informatique et Systèmes
2018-2024
Université de Toulon
2019-2024
University of Amsterdam
2022
New York University
2017
Hasso Plattner Institute
2017
University of Potsdam
2017
University of California, San Diego
2017
We present an Integer Linear Program for exact inference under a maximum coverage model automatic summarization. compare our model, which operates at the sub-sentence or "concept-level, to sentence-level previously solved with ILP. Our scales more efficiently larger problems because it does not require quadratic number of variables address redundancy in pairs selected sentences. also show how include sentence compression ILP formulation, has desirable property performing and selection...
This paper presents an unsupervised, graph based approach for extractive summarization of meetings. Graph methods such as TextRank have been used sentence extraction from news articles. These model text a with sentences nodes and edges on word overlap. A node is then ranked according to its similarity other nodes. The spontaneous speech in meetings leads incomplete, informed high redundancy calls additional measures extract relevant sentences. We propose extension the algorithm that clusters...
We introduce a model for extractive meeting summarization based on the hypothesis that utterances convey bits of information, or concepts. Using keyphrases as concepts weighted by frequency, and an integer linear program to determine best set utterances, is, covering many possible while satisfying length constraint, we achieve ROUGE scores at least good ROUGE-based oracle derived from human summaries. This brings us critical discussion future summarization.
The CALO Meeting Assistant (MA) provides for distributed meeting capture, annotation, automatic transcription and semantic analysis of multiparty meetings, is part the larger personal assistant system. This paper presents CALO-MA architecture its speech recognition understanding components, which include real-time offline transcription, dialog act segmentation tagging, topic identification segmentation, question-answer pair identification, action item recognition, decision extraction, summarization.
Ability-based design is a useful framework that centralizes the abilities (all users can do) of people with disabilities in approaching assistive technologies. However, although this aspires to support designing all kinds disabilities, it mainly effective supporting those whose be clearly defined and measured, particular, physical sensory attributes ability. As result, ability-based only provides limited guidance intellectual disability, cognitive, physical, sensory, practical vary along...
George Giannakopoulos, Jeff Kubina, John Conroy, Josef Steinberger, Benoit Favre, Mijail Kabadjov, Udo Kruschwitz, Massimo Poesio. Proceedings of the 16th Annual Meeting Special Interest Group on Discourse and Dialogue. 2015.
This paper describes the system developed at LIF for SemEval-2016 evaluation campaign.The goal of Task 4.A was to identify sentiment polarity in tweets.The extends Convolutional Neural Networks (CNN) state art approach.We initialize input representations with embeddings trained on different units: lexical, partof-speech, and embeddings.Neural networks each space are separately, then extracted from their hidden layers concatenated as a fusion neural network.The ranked 2nd obtained an average F1 63.0%.
Early linguistic interaction plays a key role in children's social and cognitive development. However, there is lack of quantitative studies that offer comprehensive insight into the communicative landscape characterizes child-caregiver dialogues.In this study, we apply advanced Natural Language Processing (NLP) techniques to analyze multiple corpora, encompassing data from N = 609 individual children (aged 20 32 months old), over 2,500 conversations around 700k pairs interactive turns. Our...
Abstract Objectives: To describe the epidemiology of nosocomial coagulase-negative staphylococci (CoNS) bacteremia and to evaluate clinical significance a single blood culture positive for CoNS. Design: A 3-year retrospective cohort study based on data prospectively collected through hospital-wide surveillance. Bacteremia was defined according CDC criteria, except that growing CoNS not systematically considered as contaminant. All clinically significant cultures were analysis. Setting: large...
In recent years, joint text-image embeddings have significantly improved thanks to the development of transformer-based Vision-Language models. Despite these advances, we still need better understand representations produced by those this paper, compare pre-trained and fine-tuned at a vision, language multimodal level. To that end, use set probing tasks evaluate performance state-of-the-art models introduce new datasets specifically for probing. These are carefully designed address range...
Progress in both speech and language processing has spurred efforts to support applications that rely on spoken rather than written input. A key challenge moving from text-based documents such is lacks explicit punctuation formatting, which can be crucial for good performance. This article describes different levels of segmentation, approaches automatically recovering segment boundary locations, experimental results demonstrating impact several tasks. The also show a need optimizing...
Despite considerable work in automatic meeting summarization over the last few years, comparing results remains difficult due to varied task conditions and evaluations. To address this issue, we present a method for determining best possible extractive summary given an evaluation metric like ROUGE. Our oracle system is based on knapsack-packing framework, though NP-Hard, can be solved nearly optimally by genetic algorithm. frame new research meaningful context, suggest presenting our...
Speech contains additional information than text that can be valuable for automatic speech summarization. In this paper, we evaluate how to effectively use acoustic/prosodic features extractive meeting summarization, and integrate prosodic with lexical structural further improvement. To properly represent features, propose different normalization methods based on speaker, topic, or local context information. Our experimental results show using only the achieve better performance non-prosodic...
The CALO Meeting Assistant provides for distributed meeting capture, annotation, automatic transcription and semantic analysis of multiparty meetings, is part the larger personal assistant system. This paper summarizes CALO-MA architecture its speech recognition understanding components, which include real-time offline transcription, dialog act segmentation tagging, question-answer pair identification, action item recognition, decision extraction, summarization.
We propose an alternative evaluation metric to Word Error Rate (WER) for the decision audit task of meeting recordings, which exemplifies how evaluate speech recognition within a legitimate application context. Using machine learning on initial seed human-subject experimental data, our handily outperforms WER, correlates very poorly with human subjects’ success in finding decisions given ASR transcripts range WERs.
In concept-based summarization, sentence selection is modelled as a budgeted maximum coverage problem.As this problem NP-hard, pruning low-weight concepts required for the solver to find optimal solutions efficiently.This work shows that reducing number of in model leads lower ROUGE scores, and more importantly presence multiple solutions.We address these issues by extending provide single solution, eliminate need concept using an approximation algorithm achieves comparable performance exact...
Traditional approaches to Information Extraction (IE) from speech input simply consist in applying text based methods the output of an Automatic Speech Recognition (ASR) system. If it gives satisfaction with low Word Error Rate (WER) transcripts, we believe that a tighter integration IE and ASR modules can increase performance more difficult conditions. More specifically this paper focuses on robust extraction Named Entities where temporal mismatch between training test corpora occurs. We...