Section: New Results

Natural Language Processing

Participants : Elena Cabrio, Julien Cojan, Fabien Gandon, Maxime Lefrançois, Serena Villata.

We have proposed a combined framework of natural language processing and argumentation theory to support the users in their interactions within online debate platforms. The framework combines a natural language processing module which exploits the textual entailment approach and detects the arguments in natural language debates and the relationships among them, and an argumentation module which represents the debates as graphs and detects the accepted arguments. The results of this research have been published in one of the major conferences in the field of Artificial Intelligence (ECAI [30] ), and in the areas of natural language processing (ACL [28] ) and argumentation theory (COMMA [29] ).

To enhance users interactions with the Web of Data, query interfaces providing an extensible mapping between natural language expressions, and concepts and relations in structured knowledge bases are becoming particularly relevant. As a result of the first year of her postdoc, Elena Cabrio (together with Julien Cojan) designed QAKiS (Question Answering wiKiframework-based System), that allows end users to submit a query in English to an RDF triple store and obtain the answer in the same language, hiding the complexity of the non intuitive formal query languages involved in the resolution process. At the same time, the expressiveness of these standards is exploited to scale to the huge amounts of available semantic data.

In its current implementation, QAKiS addresses the task of Question Answering (QA) over structured knowledge bases (e.g. DBpedia) where relevant information is expressed also in unstructured form (e.g. Wikipedia pages). Its major novelty is to implement a relation-based match for question interpretation, to convert the user question into a query language (e.g. SPARQL). A demo of the system is available online (http://dbpedia.inria.fr/qakis/ ). The results of this research have been published as a demo paper in the main conference of Semantic Web, ISWC [57] , and in the special issue of the journal Intelligenza Artificiale [14] . QAKiS has been evaluated with respect to state-of-the-art systems taking part into the QALD-2 (Question Answering over Linked Data) challenge at ESWC, obtaining satisfactory results [59] .

In the PhD Thesis of Maxime Lefrançois, we are interested in bridging the world of natural language and the world of the Semantic Web in particular to support multilingual access to the Web of Data, and management of interlingual knowledge bases. In 2011 we introduced the ULiS project that aimed at designing a pivot-based NLP technique called Universal Linguistic System, 100% using the Semantic Web formalisms, and being compliant with the Meaning-Text theory [102] .

We showed that neither Description Logics nor Conceptual Graphs suit our needs, so this Ph.D. now focuses on the formalization of the Unit Graphs mathematical framework that is conceived to fill the gap between the highly linguistically precise Explanatory Combinatorial Dictionaries of the Meaning-Text Theory and the Dependency Grammars, and the highly mathematically grounded model of the Conceptual Graphs.

Maxime finally joined the Multilingual-Web-LT W3C Working Group, and left it when the charter got revised.