Senso Comune


Da Senso Comune.

Dott. Alessandro Oltramari


Alessandro Oltramari (Vicenza, 1976) ha conseguito la laurea in Filosofia presso l'Università di Padova (2000) e il dottorato di ricerca in Scienze Cognitive presso l'Università di Trento (2006). Dal 2000 è collaboratore nell'attività di ricerca del Laboratorio di Ontologia Applicata (LOA) dell'Istituto di Scienze e Tecnologie della Cognizione (ISTC) del CNR, dove si occupa principalmente di ontologie e lessici computazionali, con interesse precipuo nello studio e realizzazione di modelli ibridi che integrino le due diverse tipologie di risorse. In questo contesto è stato co-organizzatore del workshop "OntoLex" nel 2004 (Lisbona), 2005 (Isola Jeju, Corea del Sud), 2006 (Genova), divenuto nel corso di questi anni principale luogo d'incontro per ricercatori e specialisti del settore. E' stato "Visiting Research Associate" presso il Dipartimento di Psicologia dell'Università di Princeton (USA) nei periodi marzo-maggio 2005 e agosto-ottobre 2006, ospite del Laboratorio di Scienze Cognitive in cui fin dagli anni '80 viene sviluppato WordNet. E' autore di una ventina di articoli scientifici nei seguenti campi: ontologia formale e applicata, linguistica computazionale, filosofia della mente, semantica cognitiva. Tra i suoi interessi scientifici vi sono gli aspetti fondazionali di intelligenza artificiale, analisi della metafora, affective computing, interazione uomo-macchina. A partire dal gennaio 2007 collabora con Cogito S.r.l. per lo sviluppo di tecnologie linguistiche.

Tesi di dottorato

Hybridism in Cognitive Science and Technology: Foundational and Implementational Issues

La tesi definisce un approccio allo studio della cognizione in cui "rappresentazioni concettuali" e "presentazioni della percezione" si configurano come dimensioni fondamentali della “mente incarnata”. Viene inoltre analizzato il ruolo di tali strutture mentali nel contesto delle tecnologie semantiche (ontologie e database lessicali) e dei negoziati di significato tra agenti (umani e artificiali). E’ infine presentato un modello computazionale degli stati mentali (razionali ed emozionali).

Leggi l'introduzione

Knowledge can be seen as the main outcome of the process of understanding: by means of interacting with the environment, intelligent agents are able to interpret and represent world situations, suitably acting and reacting to preserve their existence and pursue specific goals. If knowledge needs to be represented in order to become a publicly-accessible resource, as a feature of the mind it is constitutively private and unaccessible from ‘outside’; in principle there are no direct entryways from the environment to the mental realm of a single agent: adequate links are needed to enable communication between these two levels, something we could preliminary refer to as media of expression.

Natural language is the most powerful medium human beings adopt in everyday life to express knowledge: specific statements are formulated to encode the results of understanding and to bargain information with other humans. One of the most relevant problems in this context is semantic ambiguity: it is a common experience, for instance, to miss the sense of a word or even of an entire sentence in a discussion; or to intend something different when pointing to the same term adopted by another agent. In this respect, our linguistic practice fails to detect or convey adequate ‘concepts’ (roughly, the meanings of the words we utilize in expressing knowledge): this is the main reason why 1) knowledge needs to be carefully negotiated and 2) mutual comprehension is anything but a plain enterprise in a community of intelligent agents.

But this is almost the end of the story; in the current work, instead, we propose to go back to the starting point and originally reconsider the above-mentioned scenario from both a theoretical and applicative perspective. Representing knowledge is a necessary step for communication; but knowledge can be represented in so far that world phenomena are previously presented to humans, namely structured and perceived as unique experiences of the subject. For instance, one can hear the sound of a boiling teapot and express a sentence about that; trivially, the linguistic rendering of the experience is something different from the personal hearing event: in the former humans exploit natural language to transmit the conceptual content of experience; in the latter, a concrete sense-driven process supplies perceptual information to the subject. What is not trivial, instead, is the analysis of the relations holding between knowledge representations and perceptual presentations, namely what we refer to in the thesis as ‘analysis of static and dynamic dimensions of human cognition’: chapter 1 aims at investigating this topic, focusing on the fact that if it’s worthwhile to consider meaning as an abstract content that can be linguistically communicated, the intrinsic nature and source of meaning is far from being disembodied. Knowledge contents are neither simple products of abstraction nor bare perception-driven units: they genuinely correspond to an interwoven combination of the two. The ‘embodiment of meaning’ is mainly reflected by the frequent use of metaphors, which shows how the physical features we experience in the environment are used to structure cognition and language. We will refer to the overall perspective underlying these considerations as the Cognitive Matrix.

These arguments also apply to human-computer interaction, with the substantial difference that here conceptual contents have to be translated in a machine- readable format: to ignore the embodied nature of meaning when switching from natural to computational languages would be to discard useful information to face semantic ambiguity in knowledge technologies. If static and dynamic dimensions of cognition constitute meaning, as we show in the current work, then they need to be also encoded in those information systems dealing with world knowledge description. On this basis, after illustrating the difficulties behind this desideratum, chapter 2 outlines the components and the characteristics of the Knowledge Matrix. The Knowledge Matrix is a technological approximated counterpart of the Cognitive Matrix: it corresponds to the applicative framework by means of which information resources can be implemented coherently to the manifold structures of cognition. Accomplishing this research direction, chapter 3 presents the archi- tecture of a prototype knowledge base for the mental domain, specifically focused on emotional concepts (and metaphors). Quoting an expression by Nicholas Negroponte, the “quality of the (digital) world we live by” depends on the quality of information we access to, that needs to be selected, filtered, and organized. Enabling cognitive adequacy of knowl- edge technologies allows humans and machines to improve the mutual access to information: in this work, our study of the process of understanding reveals the intrinsic limits of ‘Knowledge Representation’ as detached from dynamic struc- tures of embodied cognition. In this perspective we propose an hybrid approach to cognitive science and technology to overcome the delineated problems and im- prove information systems accordingly. The target of ‘making computer understand and feel like humans’ is a far-away horizon in AI and Computer Science; the more limited objective of providing a genuine cognitive framework to knowledge technologies is a nearer-by landmark we try to approach in the current work.

Scarica la tesi completa