As was discussed in section 22, language is central to a correct understanding of the mind. Compositional analytic models perform well in the domain and subject area they are developed for, but any extension is difficult and the models have incomplete psychological veracity. Here we explore how to compute representations of meaning based on a lower level of abstraction and how to use the models for tasks that require some form of language understanding.

The authors present a computational model that builds a holographic lexicon representing both word meaning and word order from unsupervised experience with natural language. The model uses simple convolution and superposition mechanisms (cf. B. B. Murdock, 1982) to learn distributed holographic representations for words. The structure of the resulting lexicon can account for empirical data from classic experiments studying semantic typicality, categorization, priming, and semantic constraint in sentence completions. Furthermore, order information can be retrieved from the holographic representations, allowing the model to account for limited word transitions without the need for built-in transition rules. The model demonstrates that a broad range of psychological data can be accounted for directly from the structure of lexical representations learned in this way, without the need for complexity to be built into either the processing mechanisms or the representations. The holographic representations are an appropriate knowledge representation to be used by higher order models of language comprehension, relieving the complexity required at the higher level.

According to an "external grounding" theory of meaning, a concept's meaning depends on its connection to the external world. By a "conceptual web" account, a concept's meaning depends on its relations to other concepts within the same system. We explore one aspect of meaning, the identification of matching concepts across systems (e.g. people, theories, or cultures). We present a computational algorithm called ABSURDIST (Aligning Between Systems Using Relations Derived Inside Systems for Translation) that uses only within-system similarity relations to find between-system translations. While illustrating the sufficiency of a conceptual web account for translating between systems, simulations of ABSURDIST also indicate powerful synergistic interactions between intrinsic, within-system information and extrinsic information. q 2002 Elsevier Science B.V. All rights reserved.

Vector-based semantic analysis is the practice of using co-occurrence statistics to construct vectors that represent word meanings by virtue of their direction in multi-dimensional semantic space. This paper discusses the theoretical presumptions behind this practice, and a representational scheme based on the Distributional Hypothesis is identified as the rationale for vector-based semantic analysis. A new method for calculating semantic word vectors is then described. The method uses random labeling of words in narrow context windows to calculate semantic context vectors for each word type in the text data. The method is evaluated with a standardized synonym test, and it is shown that incorporating linguistic information in the context vectors can enhance the results.

A common assumption of theories of memory is that the meaning of a word can be represented by a vector which places a word as a point in a multidimensional semantic space (e.g. Landauer &

Concepts are interrelated to the extent that the characterization each concept is influenced by the other concepts, and isolated to the extent that the characterization of one concept is independent of other concepts. The relative categorization accuracy of the prototype and caricature of a concept can be used as a measure of concept interrelatedness. The prototype is the central tendency of a concept, whereas a caricature deviates from the concept's central tendency in the direction opposite to the central tendency of other acquired concepts. The prototype is predicted to be relatively well categorized when a concept is relatively independent of other concepts, but the caricature is predicted to be relatively well categorized when a concept is highly related to other concepts. Support for these predictions comes from manipulations of the labels given to simultaneously acquired concepts (Experiment 1) and the order of categories during learning (Experiment 2). 3 Concepts seem to be simultaneously connected to each other and to the external world. On the one hand, concepts seem to gain their meaning by the role that they play within a network of concepts (Collins & Quillian, 1969; Field, 1977). The notion of a "conceptual web" by which concepts all mutually define one another has been highly influential in all of the major fields that comprise cognitive science, including linguistics (Saussure, 1915/1959), computer science (Lenat & Feigenbaum, 1991), psychology (Landauer & Dumais, 1997), and philosophy (Block, 1999). However, there is also dissatisfaction in some quarters with the circularity of this conceptual web account. Researchers have argued that concepts must be grounded in the external world rather than merely related to each other (Harnad, 1990). The British e...

Advocates of dynamic systems have suggested that higher mental processes are based on continuous representations. In order to evaluate this claim, we first define the concept of representation, and rigorously distinguish between discrete representations and continuous representations. We also explore two important bases of representational content. Then, we present seven arguments that discrete representations are necessary for any system that must discriminate between two or more states. It follows that higher mental processes require discrete representations. We also argue that discrete representations are more influenced by conceptual role than continuous representations. We end by

Current connectionist models of bilingual language processing have been largely restricted to localist stationary models. To fully capture the dynamics of bilingual processing, we present SOMBIP, a self-organizing model of bilingual processing that has learning characteristics. SOMBIP consists of two interconnected selforganizing neural networks, coupled with a recurrent neural network that computes lexical co-occurrence constraints. Simulations with our model indicate that (1) the model can account for distinct patterns of the bilingual lexicon without the use of language nodes or language tags, (2) it can develop meaningful lexical-semantic categories through self-organizing processes, and (3) it can account for a variety of priming and interference effects based on associative pathways between phonology and semantics in the lexicon, and (4) it can explain lexical representation in bilinguals with different levels of proficiency and working memory capacity. These capabilities of our model are due to its design characteristics in that (a) it combines localist and distributed properties of processing, (b) it combines representation and learning, and (c) it combines lexicon and sentences in bilingual processing. Thus, SOMBIP serves as a new model of bilingual processing and provides a new perspective on connectionist bilingualism. It has the potential of explaining a wide variety of empirical and theoretical issues in bilingual research.

This chapter focuses on unsupervised corpus-based methods of word sense discrimination that are knowledge-lean, and do not rely on external knowledge sources such as machine readable dictionaries, concept hierarchies, or sense-tagged text. They do not assign sense tags to words; rather, they discriminate among word meanings based on information found in unannotated corpora. This chapter reviews distributional approaches that rely on monolingual corpora and methods based on translational equivalence as found in word-aligned parallel corpora. These techniques are organized into type- and token-based approaches. The former identify sets of related words, while the latter distinguish among the senses of a word used in multiple contexts.

A neural network model of language acquisition is introduced, motivated by current research in psychology and linguistics. It uses both extra-linguistic perceptual features and symbolic representations of words. The network learns to auto-associate these inputs to their linguistic labels, as well as to predict the next word in the corpus. This is interpreted to model both the acquisition of a lexicon, and the beginnings of syntax or grammar (word order). Furthermore, the inclusion of the extralinguistic perceptual features is argued to be a form of direct developmental grounding in embodied concepts, which allows the later learning of more abstract concepts to be grounded indirectly in meaning through relations to the first words. Through this bootstrapping process, the entire network may be scalable to large vocabularies, and may bridge the gap between high-dimensional and embodied theories of meaning.