In order to achieve interoperability of information in the context of the Semantic Web, it is necessary to find effective ways to align different ontologies. As the number of ontologies grows for a given domain, and as overlap between ontologies grows proportionally, it is becoming more and more crucial to develop accurate and reliable techniques to perform this task automatically. While traditional approaches to address this challenge are based on string metrics and structure analysis, in this paper we present a methodology to align ontologies automatically using machine learning techniques. Specifically, we use convolutional neural networks to perform string matching between class labels using character embeddings. We also rely on the set of superclasses to perform the best alignment. Our results show that we obtain state-of-the-art performance on ontologies from the Ontology Alignment Evaluation Initiative (OAEI). Our model also maintains good performance when tested on a different domain, which could lead to potential cross-domain applications.

We present an algorithm intended to visually represent the sense of verb related to an object described in a text sequence, as a movement in 3D space. We describe a specific semantic analyzer, based on a standard verbal ontology, dedicated to the interpretation of action verbs as spatial actions. Using this analyzer, our system build a generic 3D graphical path for verbal concepts allowing space representation, listed as SelfMotion concepts in the FrameNet ontology project. The object movement is build by first extracting the words and enriching them with the semantic analyzer. Then, weight tables, necessary to obtain characteristics values (orientation, shape, trajectory...) for the verb are used in order to get a 3D path, as realist as possible. The weight tables were created to make parallel between features defined for SelfMotion verbal concept (some provided by FrameNet, other determined during the project) and values used in the final algorithm used to create 3D moving representations from input text. We evaluate our analyzer on a corpus of short sentences and presents our results.

In this paper, we present an algorithm for improving named entity resolution and entity linking by using surface form generation and rewriting. Surface forms consist of a word or a group of words that matches lexical units like Paris or New York City. Used as matching sequences to select candidate entries in a knowledge base, they contribute to the disambiguation of those candidates through similarity measures. In this context, misspelled textual sequences (entities) can be impossible to identify due to the lack of available matching surface forms. To address this problem, we propose an algorithm for surface form refinement based on Wikipedia resources. The approach extends the surface form coverage of our entity linking system, and rewrites or reformulates misspelled mentions (entities) prior to starting the annotation process. The algorithm is evaluated on the corpus associated with the monolingual English entity linking task of NIST KBP 2013. We show that the algorithm improves the entity linking system performance.

The Semantic Annotation (SA) task consists in establishing the relation between a textual entity (word or group of words designating a named entity of the real world or a concept) and its corresponding entity in an ontology. The main difficulty of this task is that a textual entity might be highly polysemic and potentially related to many different ontological representations. To solve this specific problem, various Information Retrieval techniques can be used. Most of those involves contextual words to estimate wich exact textual entity have to be recognized. In this paper, we present a resource of contextual words that can be used by IR algorithms to establish a link between a named entity (NE) in a text and an entry point to its semantic description in the LinkedData Network.

Les encyclopédies numériques contiennent aujourd’hui de vastes inventaires de formes d’écritures pour des noms de personnes, de lieux, de produits ou d’organisation. Nous présentons un système hybride de détection d’entités nommées qui combine un classifieur à base de Champs Conditionnel Aléatoires avec un ensemble de motifs de détection extraits automatiquement d’un contenu encyclopédique. Nous proposons d’extraire depuis des éditions en plusieurs langues de l’encyclopédie Wikipédia de grandes quantités de formes d’écriture que nous utilisons en tant que motifs de détection des entités nommées. Nous décrivons une méthode qui nous assure de ne conserver dans cette ressources que des formes non ambiguës susceptibles de venir renforcer un système de détection d’entités nommées automatique. Nous procédons à un ensemble d’expériences qui nous permettent de comparer un système d’étiquetage à base de CRF avec un système utilisant exclusivement des motifs de détection. Puis nous fusionnons les résultats des deux systèmes et montrons qu’un gain de performances est obtenu grâce à cette proposition.

L’étiquetage sémantique consiste à associer un ensemble de propriétés à une séquence de mots contenue dans un texte. Bien que proche de la tâche d’étiquetage par entités nommées, qui revient à attribuer une classe de sens à un mot, la tâche d’étiquetage ou d’annotation sémantique cherche à établir la relation entre l’entité dans son texte et sa représentation ontologique. Nous présentons un étiqueteur sémantique qui s’appuie sur un étiqueteur d’entités nommées pour mettre en relation un mot ou un groupe de mots avec sa représentation ontologique. Son originalité est d’utiliser une ontologie intermédiaire de nature statistique pour établir ce lien.

This paper presents a word sense disambiguation (WSD) approach based on syntactic and logical representations. The objective here is to run a number of experiments to compare standard contexts (word windows, sentence windows) with contexts provided by a dependency parser (syntactic context) and a logical analyzer (logico-semantic context). The approach presented here relies on a dependency grammar for the syntactic representations. We also use a pattern knowledge base over the syntactic dependencies to extract flat predicative logical representations. These representations (syntactic and logical) are then used to build context vectors that are exploited in the WSD process. Various state-of-the-art algorithms including Simplified Lesk, Banerjee and Pedersen and frequency of co-occurrences are tested with these syntactic and logical contexts. Preliminary results show that defining context vectors based on these features may improve WSD by comparison with classical word and sentence context windows. However, future experiments are needed to provide more evidence over these issues.

Dans cet article, nous proposons une méthode pour identifier, dans un texte en français, l’ensemble des expressions adverbiales de localisation temporelle, ainsi que tous les verbes, noms et adjectifs dénotant une éventualité (événement ou état). Cette méthode, en plus d’identifier ces expressions, extrait certaines informations sémantiques : la valeur de la localisation temporelle selon la norme TimeML et le type des éventualités. Pour les expressions adverbiales de localisation temporelle, nous utilisons une cascade d’automates, alors que pour l’identification des événements et états nous avons recours à une analyse complète de la phrase. Nos résultats sont proches de travaux comparables sur l’anglais, en l’absence d’évaluation quantitative similaire sur le français.