Search

International audience; Correcting errors in a data set is a critical issue. This task can be either handmade by experts, or by crowdsourcing methods or automatically done using algorithms. Although even if the rate of errors present in a given lexical network is rather low, it is important to reduce it. We present here automatic methods for detecting potential secondary errors that would result from automatic inference mechanisms when they rely on an initial error manually detected. Encouraging results also invite us to consider strategies that would automatically detect "erroneous" initial relations, which could lead to the automatic detection of the majority of errors in a lexical-semantic network.

Human brains can be seen as knowledge processors in a distributed system. Each of them can achieve, conscious or not, a small part of a treatment too important to be done by one. These are also "hunter / gatherers" of knowledge. Provided that the number of contributors is large enough, the results are usually better quality than if they were the result of the activity of a single person, even if it is a domain expert. This type of activity is done via online games.

The JDM lexical network has been built thanks to on-line games the main of which, JeuxDeMots (JDM), was launched in 2007. It is currently a large lexical network, in constant evolution, containing more than 310,000 terms connected by more than 6.5 million relations. The riddle game Totaki (Tip Of the Tongue with Automated Knowledge Inferences), the initial version of which was elaborated with Michael Zock, was launched in a first version in 2010. The initial aim of this project is to cross validate the JDM lexical network. Totaki uses this lexical network to make proposals from user given clues, and in case of failure players can supply new information, hence enriching the network. Endogenous processes of inference, by deduction, induction, abduction, also allow to find new information not directly available in the network and hence lead to a densification of the network. The assumption about the validation is that if Totaki is able to guess proper terms from user clues, then the lexical network contains appropriate relations between words. Currently, Totaki achieves a 75 % success rate, to be compared to less than 50 % if the guessing is done by human users. One serious application of Totaki is to be viewed as a tool for lexical access and a possible remedy for the tip of the tongue problem. The Wikipedia encyclopaedia, built in a collaborative way, represents a very important volume of knowledge (about 1.5 million articles in its French version). The idea developed in this chapter consists in benefiting from Wikipedia to enrich the JDM network and evaluate the impact on Totaki performance. Instead of relying only on the JDM network, Totaki also makes use of information extracted from Wikipedia. The overall process is then both endogenous and exogenous. In a first part, we shall remind the reader the basic principles of a lexical network, then the aims and the underlying principles of the Totaki game. We shall see on examples Totaki may be used as a game to evaluate and enrich the JDM network, but also it may be considered as a tool for the Tip Of the Tongue problem; partial syntactic or morphologic information may be added to semantic information to help the user. In a second part, we shall show the results of the evaluation of the JDM network, results we obtained playing Totaki. We shall clarify the process allowing the introduction in the Totaki game of data extracted from Wikipedia as a complement in the information from the JDM network, and we shall briefly present the results provided by the first experiments.

Joubert Alain. L'autre côté de l'eau : la Seine barrière mentale ?. In: Études Normandes, 43e année, n°4, 1994. Seine Normande, Seine Européenne. pp. 45-66.

Joubert Alain. La Seine normande : un patrimoine multiforme. In: Études Normandes, 58e année, n°3, 2009. La Seine Normande. pp. 21-38.

Thanks to the participation of a large number of persons via web-based games, a large-sized evolutionary lexical network is available for French. With this resource, we approached the question of the determination of the word usages of a term, and then we introduced the notion of similarity between these various word usages. So, we were able to build for a term its word usage tree: the root groups together all possible usages of this term and a search in the tree corresponds to a refinement of these word usages. The labelling of the various nodes of the word usage tree of a term is made during a width-first search: the root is labelled by the term itself and each node of the tree is labelled by a term stemming from the clique or quasi-clique this node represents. We show on a precise example that it is possible that some nodes of the tree, often leaves, cannot be labelled without ambiguity. This paper ends with an evaluation about word usages detected in our lexical network.

Joubert A., Cordier P., van Staelen V. La Protection de la Nature. Les réserves naturelles et les conséquences de la déforestation. In: Revue de botanique appliquée et d'agriculture coloniale, 17ᵉ année, bulletin n°185, janvier 1937. pp. 4-8.

International audience; Lexical-semantic network construction and validation is a major issue in the NLP. No matter the construction strategies used, automatically inferring new relations from already existing ones is a way to improve the global quality of the resource by densifying the network. In this context, an inference engine has for purpose to formulate new conclusions (i.e. relations between terms) from already existing premises (also relations) on the network. In this paper we devise an inference engine for the JeuxDeMots lexical network which contains terms and typed relations between terms. In the JeuxDeMots project, the lexical network is constructed with the help of a game with a purpose and thousands of players. Polysemous terms may be refined in several senses (bank may be a bank>financial institution or a bank>river) but as the network is indefinitely under construction (in the context of a Never Ending Learning approach) some senses may be missing. The approach we propose is based on the triangulation method implementing semantic transitivity with a blocking mechanism for avoiding proposing dubious new relations. Inferred relations are proposed to contributors to be validated. In case of invalidation, a reconciliation strategy is undertaken to identify the cause of the wrong inference : an exception, an error in the premises or a transitivity confusion due to polysemy with the identification of the proper words senses at stake.