How do people know as much as they do with as little information as they get? The problem takes many forms; learning vocabulary from text is an especially dramatic and convenient case for research. A new general theory of acquired similarity and knowledge representation, latent semantic analysis (LSA), is presented and used to successfully simulate such learning and several other psycholinguistic phenomena. By inducing global knowledge indirectly from local co-occurrence data in a large body of representative text, LSA acquired knowledge about the full vocabulary of English at a comparable rate to schoolchildren. LSA uses no prior linguistic or perceptual similarity knowledge; it is based solely on a general mathematical learning method that achieves powerful inductive effects by extracting the right number of dimensions (e.g., 300) to represent objects and contexts. Relations to other theories, phenomena, and problems are sketched. Prologue "How much do we know at any time? Much more, or so I believe, than we know we know!" —Agatha Christie, The Moving Finger A typical American seventh grader knows the meaning of

Abstract not found

We examine the question of which aspects of language are uniquely human and uniquely linguistic in light of recent arguments by Hauser, Chomsky, and Fitch that the only such aspect is syntactic recursion, the rest of language being either specific to humans but not to language (e.g., words and concepts) or not specific to humans (e.g., speech perception). We find this argument problematic. It ignores the many aspects of grammar that are not recursive, such as phonology, morphology, case, and agreement. It is inconsistent with the anatomy and neural control of the human vocal tract. And it is weakened by experiments showing that speech perception cannot be reduced to primate audition, that word learning cannot be reduced to fact learning, and that at least one gene involved in speech and language was evolutionarily selected in the human lineage but is not specific to recursion. The recursion-only claim, we suggest, is motivated by Chomsky’s recent approach to syntax, the Minimalist Program, which de-emphasizes the same aspects of language. The approach, however, is sufficiently problematic that it cannot be used to support claims about evolution. We contest other arguments from Chomsky that language is not an adaptation, namely that it is “perfect, ” nonredundant, unusable in any partial form, and badly designed for communication. The hypothesis that language is a complex adaptation for communication which evolved piecemeal avoids all these problems.

Abstract. Geographical information exists in multiple forms such as cartographical maps, images, and texts. Effective retrieval systems for geographical information are currently studied by both geospatial information scientists and library/information scientists. The two groups take quite different approaches which have rarely been explicitly compared and connected. This paper discusses the advantages and limitations of current geographical information retrieval (GIR) and textual information retrieval (IR) systems in dealing with multimedia geographical information, and proposes a new retrieval model, called GeoVSM (Geographical + Vector Space Model), which integrates coordinate-based geographical indexing with keyword-based vector space model in representing information space. Document ranking by relevance is supported by document-query similarity measure taking into account degree of relevance in both spatial domain and thematic domain. To support visual query and browsing, GeoVSM model recognizes the fundamental differences in the dimensionality and configuration of geographical space and thematic space, and demands multi-view visual interfaces based on a “world ” metaphor and a “desktop ” metaphor. As an example of such interfaces, GeoVIBE is presented that supports a coordinated GeoView and a VibeView for smooth integration of two browsing strategies using geographic clues as well as thematic clues provided by users. 1.

This paper explores the possibilities of visualizing document similarities and differences in both spatial and topical domains. Building on previous studies of geographical information retrieval and textual information retrieval (IR) systems, we report on the development of an information browsing tool, GeoVIBE. The system consists of two types of browsing windows, GeoView and VibeView, that work in coordination for visual navigation in the document space. GeoView imposes a geographical order to the document space based on the idea of hypermaps where "icons" and "footprints" may be embedded in maps as the clickable hotspots linking to relevant documents.

This paper raises the question of the connection between embodiment and higher-level cognition which has been eloquently addressed before, but has not yet received much focus in the AI community. The paper then proceeds to break the question down into subparts, and address how each can be approached and studied. Finally, the paper briefly overviews two directions of our work: group behavior and imitative behavior, and describes their relation to the issue of embodiment and cognition. Motivation The role of physical embodiment in cognition has long been the subject of debate. In Artificial Intelligence (AI) it is largely accepted that embodiment has strong implications on the control strategies for generating purposive and intelligent behavior in the world. However, some theories outside AI have proposed that embodiment not only constrains but may also facilitate certain types of higher-level cognition. Evidence from neuroscience allows for postulating shared mechanisms for low-level c...

The fact that concepts are grammaticalized di#erently in different languages is a major problem for translation, especially for machine translation. Two major examples of this are syntactic number, and the use of (in)definite articles (a, some, the). In languages such as English, nouns are marked for number and the choice of article (or of no article) must be made for every noun phrase. In contrast, for languages such as Japanese, number distinctions are not normally made, and there are no articles. This means that whenever a noun phrase is translated from Japanese to English, even if the denotation is perfectly understood and a good translation equivalent found, generating the noun phrase still requires two difficult choices: should the head noun be singular or plural, and which article, if any, should be generated. This thesis proposes a semantic representation and a series of three heuristic algorithms that make possible the appropriate generation of articles and number when translating from Japanese to English. The semantic representation provides a tractable set of features to represent (1) the referential use of a noun phrase, as either referential, generic, ascriptive or idiomatic; (2) the interpretation of the noun phrase's referent as either a countable individual or a mass, with seven detailed subtypes; (3) the definiteness of the noun phrase, as either definite, indefinite, definite and extensive, or possessed. The three algorithms automatically acquire values for these features from the analysis of the Japanese text and the lexical properties of the English translation equivalents, and then use them to generate English. The first algorithm determines the referential use of Japanese noun phrases, based on a defeasible hierarchy of pragmatic rules that are applie...

In this paper we present an analysis of the ways Japanese and Korean categorize long-distance communication, for example, over the telephone, or through email, fax or letters. In Korean and Japanese, numeral classifiers are used to overtly categorize such communicative acts. Japanese uses more classifiers in this field than Korean, and so there is more possible variation in classifier use. We use the framework of Pustejovsky's (1995) Generative Lexicon to make clear the di#erences between kind and shape classifiers, and explain the variation in classifier use by showing how different classifiers select for di#erent features of lexical entries. 1 Introduction In this paper, we examine how Japanese and Korean classifiers are used to pick out certain aspects of the denotations of the nouns they classify. Classifiers are used to pick out salient features of the noun they classify. These can be kinds, that is, semantic categories, or properties such as shape or size. Kind classifiers are c...

van de academische graad van doctor in de wetenschappen, in het openbaar te verdedigen op vrijdag 8 maart 2002. Acknowledgements I started as a research assistant in the Artificial Intelligence Laboratory in autumn 1996. My first interests were into behavioural robotics and robot ecosystems. As a continuation to my “licentiaats ” thesis I started building a camera system to extend the sensory perception of the lab’s robots (Belpaeme and Birk, 1997a,b; Belpaeme, 1998; Birk and Belpaeme, 1998; Birk et al., 1998, 1999; Belpaeme and Birk, 2001). It was around that time when Luc Steels got interested in the origins of language. His early experiments formed the seed for what is now one of the most important paradigms for exploring linguistic interactions with computer simulations. Luc soon wanted more and had plans to implement a language experiment in the real world, for which I delivered the visual perception (Belpaeme et al., 1998; Belpaeme, 1999). This got me interested in visual features, and my research

A certain metaphysical thesis about meaning that we'll call Informational Role Semantics (IRS) is accepted practically universally in linguistics, philosophy and the cognitive sciences: the meaning (or content, or `sense') of a linguistic expression 1 is constituted, at least in part, by at least some of its inferential relations. This idea is hard to state precisely, both because notions like metaphysical constitution are moot