An ontology for human-like interaction systems

Resumen

This report proposes and describes the development of a Ph.D. Thesis aimed at building an ontological knowledge model supporting Human-Like Interaction systems. The main function of such knowledge model in a human-like interaction system is to unify the representation of each concept, relating it to the appropriate terms, as well as to other concepts with which it shares semantic relations. When developing human-like interactive systems, the inclusion of an ontological module can be valuable for both supporting interaction between participants and enabling accurate cooperation of the diverse components of such an interaction system. During human communication, the relation between cognition and messages relies in formalization of concepts, linked to terms (or words) in a language that will enable its utterance (at the expressive layer). Moreover, each participant has a unique conceptualization (ontology), different from other individual’s. Through interaction, is the intersection of both part’s conceptualization what enables communication. Therefore, for human-like interaction is crucial to have a strong conceptualization, backed by a vast net of terms linked to its concepts, and the ability of mapping it with any interlocutor’s ontology to support denotation. Besides, humans usually handle a certain range of similar concepts they can use when building messages. The subject of similarity has been and continues to be widely studied in the fields and literature of computer science, psychology and sociolinguistics. Good similarity measures are necessary for several techniques from these fields such as information retrieval, clustering, data-mining, sense disambiguation, ontology translation and automatic schema matching. Furthermore, the ontological component should also be able to perform certain inferential processes, such as the calculation of semantic similarity between concepts. The principal benefit gained from this procedure is the ability to substitute one concept for another based on a calculation of the similarity of the two, given specific circumstances. Besides, semantic similarity also enables the system building explanations about message meaning. Providing them to the user should help to clarify a given misunderstood concept based on similar concepts, thereby enhancing communicative effectiveness. First strategy can be based on the use of reformulated utterances and synonyms (if any). But many cases would reveal the lack of the focused concept (or synonyms are not appropriate). Then, the system can build explanations, which may include (but is not restricted to) hyperonyms, hyponyms, cohyponyms, holonyms, meronyms, antonyms, foreign terms, etc. But what is really challenging is focusing this feature from the opposite point of view: the system should be able of understanding such explanations regarding previously-unknown semantic relations between known concepts, formalize such knowledge and hoard it for further use. What usually hinders the development of such ontologies is their maintenance, because the related mechanisms are usually too costly. Specifically, feeding these ontologies, a task that is usually performed manually by experts, involves unbearable costs. The fore proposed challenge, to develop advanced mechanisms for automatic knowledge acquisition, appears to be a proper solution: knowledge is obtained through thousands (or more) of human-like interactions with human subjects. Actually, this procedure is the way humans incorporate new concepts and terms to their knowledge (through both oral and written communication). Consequently, the goal is to imitate human behavior to attain a sustainable solution. The referred human behavior involves some supporting concepts and process that have to be taken into account, such as reputation and reliability. The initial reliability on the knowledge will depend upon the trust on the source (which may vary during time) and its reputation on a given subject (facts from the same source could be taking different initial reliability if are referred to different domains). Anyhow, of course, any knowledge’s reliability is subject to review due to feedbacks produced through its use. Summarizing, through the system lifetime, the knowledge bases would be enriched by interacting with the users, thus learning new concepts, terms and relationships, and by continuous refinement of their knowledge. Despite this ontology proposal requires vast amount of knowledge which acquisition would entail unbearable costs, the knowledge crowdsourcing focus may overcome the obstacle, turning the approach into a realistic solution.