Papers by Pascale Sebillot
Proceedings of the Third Edition Workshop on Speech, Language & Audio in Multimedia - SLAM '15, 2015
We investigate video hyperlinking based on speech transcripts, leveraging a hierarchical topical ... more We investigate video hyperlinking based on speech transcripts, leveraging a hierarchical topical structure to address two essential aspects of hyperlinking, namely, serendipity control and link justification. We propose and compare different approaches exploiting a hierarchy of topic models as an intermediate representation to compare the transcripts of video segments. These hierarchical representations offer a basis to characterize the hyperlinks, thanks to the knowledge of the topics who contributed to the creation of the links, and to control serendipity by choosing to give more weights to either general or specific topics. Experiments are performed on BBC videos from the Search and Hyperlinking task at MediaEval. Link precisions similar to those of direct text comparison are achieved however exhibiting different targets along with a potential control of serendipity.
One important class of online videos is that of news broadcasts. Most news organisations provide ... more One important class of online videos is that of news broadcasts. Most news organisations provide near-immediate access to topical news broadcasts over the Internet, through RSS streams or podcasts. Until lately, technology has not made it possible for a user to automatically go to the smaller parts, within a longer broadcast, that might interest them. Recent advances in both speech recognition systems and natural language processing have led to a number of robust tools that allow us to provide users with quicker, more focussed access to relevant segments of one or more news broadcast videos. Here we present our new interface for browsing or searching news broadcasts (video/audio) that exploits these new language processing tools to (i) provide immediate access to topical passages within news broadcasts, (ii) browse news broadcasts by events as well as by people, places and organisations, (iii) perform cross lingual search of news broadcasts, (iv) search for news through a map interf...
Proceedings of the 3rd ACM conference on International conference on multimedia retrieval - ICMR '13, 2013
Searching for relevant webpages and following hyperlinks to related content is a widely accepted ... more Searching for relevant webpages and following hyperlinks to related content is a widely accepted and effective approach to information seeking on the textual web. Existing work on multimedia information retrieval has focused on search for individual relevant items or on content linking without specific attention to search results. We describe our research exploring integrated multimodal search and hyperlinking for multimedia data. Our investigation is based on the Medi-aEval 2012 Search and Hyperlinking task. This includes a known-item search task using the Blip10000 internet video collection, where automatically created hyperlinks link each relevant item to related items within the collection. The search test queries and link assessment for this task was generated using the Amazon Mechanical Turk crowdsourcing platform. Our investigation examines a range of alternative methods which seek to address the challenges of search and hyperlinking using multimodal approaches. The results of our experiments are used to propose a research agenda for developing effective techniques for search and hyperlinking of multimedia content.
Computer Assisted Language Learning, 1997
In this paper, we focus on student modeling within an Intelligent Tutoring System (ITS). Elaborat... more In this paper, we focus on student modeling within an Intelligent Tutoring System (ITS). Elaborating a student modeling system implies to determine the formalism in which the student knowledge will be represented, and the processes that will dynamically acquire and synthesize this knowledge. We describe three domain-independent properties that this formalism and these processes must possess to build sound and accurate student models. First, because the student's knowledge evolves in time, the modeling system must be able to represent knowledge that issues from defeasible reasoning. Second, since students may have contradictions in mind, it also must deal with paraconsistent reasoning. Third, because the diagnosis of the student's cognitive state is not certain, the results of the cognitive diagnosis must be considered as hypotheses rather than certain facts. These hypotheses may have to be withdrawn if contradictory information is subsequently acquired. Therefore, the system must also be able to follow hypothetical reasoning. We show how an implementation based upon probabilistic logic can take into account both student's defeasible and paraconsistent reasonings. We also point out where, when and how hypothetical reasoning mechanisms must intervene. We exemplify our results within the framework of Compounds, an ITS about English compounding processes.
Computer Assisted Language Learning, 1993
... For Page 16. 12 Paul Boucher, Fr ed eric Danna et Pascale S ebillot instance, as mentioned ab... more ... For Page 16. 12 Paul Boucher, Fr ed eric Danna et Pascale S ebillot instance, as mentioned above,truck-driver will be analysed syntactically in two di erent ways: truck-drive]V erandtruck-driver]N thereby permitting two distinct semantic analyses. ...
Annual Conference of the International Speech Communication Association, 2009
This paper presents an unsupervised topic-based language model adaptation method which specialize... more This paper presents an unsupervised topic-based language model adaptation method which specializes the standard minimum information discrimination approach by identifying and combining topic-specific features. By acquiring a topic terminology from a thematically coherent corpus, language model adaptation is restrained to the sole probability re-estimation of n-grams ending with some topic-specific words, keeping other probabilities untouched. Experiments are carried out on a large set of spoken documents about various topics. Results show significant perplexity and recognition improvements which outperform results of classical adaptation techniques.
This paper demonstrates the feasibility of automatic acquisition of generative lexicons from corp... more This paper demonstrates the feasibility of automatic acquisition of generative lexicons from corpora through the report of four experiments in machine learning in which various levels of word tagging (categorial and semantic) are handled. The lexical information that is learnt consists of lists of noun-verb couples related by one of the roles of the qualia structure. They provide linguistic knowledge
Language Resources and Evaluation, 2008
Texts generated by automatic speech recognition (ASR) systems have some specificities, related to... more Texts generated by automatic speech recognition (ASR) systems have some specificities, related to the idiosyncrasies of oral productions or the principles of ASR systems, that make them more difficult to exploit than more conventional natural language written texts. This paper aims at studying the interest of morphosyntactic information as a useful resource for ASR. We show the ability of automatic
Language Resources and Evaluation, 2008
Language models used in current automatic speech recognition systems are trained on general-purpo... more Language models used in current automatic speech recognition systems are trained on general-purpose corpora and are therefore not relevant to transcribe spoken documents dealing with successive precise topics, such as long multimedia streams, frequently tackling reports and debates. To overcome this problem, this paper shows that Web resources and natural language processing techniques can be effective to automatically collect a topic specific corpora from the Internet in order to adapt the baseline language model of an automatic speech recognition system. We detail how to characterize the topic of a segment and how to collect Web pages from which a topicspecific language model can be trained. We finally present experiments where an adapted language model is obtained by combining the topic-specific language model with the general purpose one to obtain new transcriptions. The results show that our topic adaptation technique leads to significant transcription quality gains.
Annual Conference of the International Speech Communication Association, 2010
The increasing quantity of video material requires methods to help users navigate such data, amon... more The increasing quantity of video material requires methods to help users navigate such data, among which topic segmentation techniques. The goal of this article is to improve ASRbased topic segmentation methods to deal with peculiarities of professional-video transcripts (transcription errors and lack of repetitions) while remaining generic enough. To this end, we introduce confidence measures and semantic relations in a segmentation method based on lexical cohesion. We show significant improvements of the F1-measure when integrating confidence measures and semantic relations respectively. Such improvement demonstrates that simple clues can conteract errors in automatic transcripts and lack of repetitions.
Lecture Notes in Computer Science, 1992
In this paper, we present a logic-programming modelling of feature filtering mechanisms. Feature ... more In this paper, we present a logic-programming modelling of feature filtering mechanisms. Feature filtering is an important topic in natural language processing. We give a logical-based specification, valid for different theories of features (including their evolution from the lexicon, expressed in terms of axioms, and including control tools). We define the logical foundations of these specifications, which have been shown useful for naturallanguage grammar programmers. The logical-based language and our formalization are very general and independent from any parsing strategy. They permit to represent feature and control systems proposed by different linguistic theories.
Lecture Notes in Computer Science, 2015
Expressive speech processing is an important scientific problem as expressivity introduces a lot ... more Expressive speech processing is an important scientific problem as expressivity introduces a lot of variability into speech. This variability leads to a degradation of speech application performances. Variations are reflected in the linguistic, phonological and acoustic sides of speech. However our main interest is on phonology, more precisely the study of pronunciation and of disfluencies. Both of these fields have huge impacts on speech. This report is a bibliographical review of the state of the art in expressivity and phonology modelling. Although the main focus will be on speech synthesis, we will discuss works about automatic speech recognition as well because expressivity modelling in phonology is a cross-domain problem.
Lecture Notes in Computer Science, 2006
The aim of our paper is to study the interest of part of speech (POS) tagging to improve speech r... more The aim of our paper is to study the interest of part of speech (POS) tagging to improve speech recognition. We first evaluate the part of misrecognized words that can be corrected using POS information; the analysis of a short extract of French radio broadcast news shows that an absolute decrease of the word error rate by 1.1% can be expected. We also demonstrate quantitatively that traditional POS taggers are reliable when applied to spoken corpus, including automatic transcriptions. This new result enables us to effectively use POS tag knowledge to improve, in a postprocessing stage, the quality of transcriptions, especially correcting agreement errors.
Proceedings of the 20th international conference on Computational Linguistics - COLING '04, 2004
Keywords: knowledge representation in ITS, expert model, English compounds
Le calcul de distances entre représentations textuelles est au coeur de nombreuses applications d... more Le calcul de distances entre représentations textuelles est au coeur de nombreuses applications du Traitement Automatique des Langues. Les approches standard initiallement développées pour la recherche d'information sont alors le plus souvent utilisées. Dans la plupart des cas, il est donc adopté une description sac-de-mots (ou sac-d'attributs) avec des pondérations de type TF-IDF ou des variantes, une représentation vectorielle et des fonctions classiques de similarité comme le cosinus. Dans ce papier, nous nous intéressons à l'une de ces tâches, à savoir le clustering sémantique d'entités extraites d'un corpus. Nous défendons l'idée que pour ce type de tâches, il est possible d'utiliser des représentations et des mesures de similarités plus adaptées que celles usuellement employées. Plus précisément, nous explorons l'utilisation de représentations alternatives des entités appelées sacs-de-vecteurs ou sacs-de-sacs-de-mots. Dans ce modèle, chaque entité est définie non pas par un unique vecteur, mais par un ensemble de vecteurs, chacun de ces vecteurs étant construit à partir d'une occurrence de l'entité. Pour utiliser cette représentation, nous utilisons et définissons des extensions des mesures classiques du modèle vectoriel (cosinus, Jaccard, produit scalaire...). Ces différents constituants sont testés sur notre tâche de clustering, et nous montrons que cette représentation en sac-de-vecteurs améliore significativement les résultats par rapport à une approche standard en sac- de-mots. 1 ABSTRACT. Computing distances between textual representation is at the heart of many Natural Language Processing tasks. The standard approaches initially developed for Information Retrieval are then used; most often they rely on a bag-of-words (or bag-of-feature) description with a TF-IDF (or variants) weighting, a vectorial representation and classical similarity functions like cosine. In this paper, we are interested in such a task, namely the semantic clustering of entities extracted from a text. We argue that for this kind of tasks, more suited representations 1. Ces travaux ont été (en partie) réalisés dans le cadre du programme QUAERO, financé par OSEO, agence française pour l'innovation. CORIA 2012, pp. 229-244, Bordeaux, 21-23 mars 2012 2. This work was achieved as part of the Quaero Programme, funded by OSEO, French State agency for innovation.
In this paper, we present a modelling of tree filtering mechanisms in logical-based grammars. Thi... more In this paper, we present a modelling of tree filtering mechanisms in logical-based grammars. This filtering permits restriction of the excessive generative capacity of syn-tactic parsing grammars. In order to preserve the declarative aspect of natural language grammars, we consider syntactic parsing as a deduction process; tree filtering consists in the reduction of the allowed forms of the proof tree corresponding to the derivation process of a sentence. We also evoke a computational implementation in the framework of the Government and Binding theory.
IRISA Campus universitaire de Beaulieu F. 35042 Rennes cedex RÉSUMÉ. Cet article décrit un systèm... more IRISA Campus universitaire de Beaulieu F. 35042 Rennes cedex RÉSUMÉ. Cet article décrit un système de caractérisation et détection de thèmes dans un corpus textuel non spécialisé reposant sur la notion de mots-clés, c'est-à-dire de mots dont l'apparition dans un segment de texte est symptomatique de la présence d'un thème particulier. Le système présenté extrait de manière totalement automatique, sans connaissance a priori sur la nature ou le nombre des thèmes majeurs abordés dans le corpus, une collection de classes de tels mots-clés, représentatives chacune d'un de ces thèmes. Il a pour objectif de répondre aux besoins d'applications nécessitant la connaissance de thèmes, mais sans recours à un expert humain et sans usage de données auxiliaires, sémantiques ou autres. La méthode mise en oeuvre se compose d'une série de traitements essentiellement statistiques exploitant la répartition des mots du corpus sur ses paragraphes. Les listes de mots extraites perm...
In this paper, we present a domain-independent model for the au-tomatic interpretation of nominal... more In this paper, we present a domain-independent model for the au-tomatic interpretation of nominal compounds in English and in French. This model is meant to account for productive rules of interpretation, that are inferred from the morpho-syntactic and semantic characteristics of the nominal constituents. We then propose the corpus-based approach that has been used to adapt this general model to the interpretation of compounds of a specific domain.
Uploads
Papers by Pascale Sebillot