International audienceNatural language generation is based on messages that represent meanings , and goals that are the usual starting points for communicate. How to help people to provide this conceptual input or, in other words, how to communicate thoughts to the computer? In order to express something, one needs to have something to express as an idea, a thought or a concept. The question is how to represent this. In 2009, Michael Zock, Paul Sabatier and Line Jakubiec-Jamet suggested the building of a resource composed of a linguistically motivated ontology, a dictionary and a graph generator. The ontology guides the user to choose among a set of concepts (or words) to build the message from; the dictionary provides knowledge of how to link the chosen elements to yield a message (compositional rules); the graph generator displays the output in visual form (message graph representing the user's input). While the goal of the ontology is to generate (or analyse) sentences and to guide message composition (what to say), the graph's function is to show at an intermediate level the result of the encoding process. The Illico system already proposes a way to help a user for generating (or analyzing) sentences and guiding their composition. Another system, the Drill Tutor, is an exercise generator whose goal is to help people to become fluent in a foreign language. It helps people (users have to make choices from the interface in order to build their messages) to produce a sentence expressing a message from an idea (or a concept) to its linguistic realization (or a correct sentence given in a foreign language). These two systems led us to consider the representation of the conceptual information into a symbolic language; this representation is encoded in a logic system in order to automatically check conceptual well-formedness of messages. This logic system is the Coq system used here only for its high level language. Coq is based on a typed λ-calculus. It is used for analysing conceptual input interpreted as types and also for specifying general definitions representing messages. These definitions are typed and they will be instanciated for type-checking the conceptual well-formedness of messages. 2 Line Jakubiec-Jame
Natural language generation is typically based on messages and goals. We present here our views on how to help people to provide this kind of input, i.e. how to communicate thoughts to the computer, so that it could produce the corresponding surface-forms (sentences). The resource we are building is composed of a linguistically motivated ontology, a dictionary and a graph generator, whose respective functions are (a) guiding the user to make his choices concerning the concepts/words to build the message from, (b) to provide knowledge of how to link the chosen elements to yield a message (compositional rules), and (c) the visual display of the output, i.e. message graph representing the user's input. Our starting point is Illico, a system developed for French. Yet, being designed for sentence completion rather than message construction, it tends to drown the user by providing too many options, a shortcoming that we try to overcome via the mentionned ontology combined with a tool checking conceptual well formedness. In order to make our goal feasable (allow users to express freely ideas of any sort), we will start by limiting ourselves to two small domains: soccer and textbooks designed for learning French.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.