Mary Hare

A Connectionist Model of Phonological Representation in Speech Perception

Abstract A number of recent studies have examined the effects of phonological variation on the perception of speech. These studies show that both the lexical representations of words and the mechanisms of lexical access are organized so that natural, systematic variation is tolerated by the perceptual system, while a general intolerance of random deviation is maintained. Lexical abstraction distinguishes between phonetic features that form the invariant core of a word and those that are susceptible to variation. Phonological inference relies on the context of surface changes to retrieve the underlying phonological form. In this article we present a model of these processes in speech perception, based on connectionist learning techniques. A simple recurrent network was trained on the mapping from the variant surface form of speech to the underlying form. Once trained, the network exhibited features of both abstraction and inference in its processing of normal speech, and predicted that similar behavior will be found in the perception of nonsense words. This prediction was confirmed in subsequent research (Gaskell & Marslen-Wilson, 1994).

Default Generalisation in Connectionist Networks.

Abstract A potential problem for connectionist accounts of inflectional morphology is the need to learn a “default” inflection (Prasada & Pinker, 1993). The early connectionist work of Rumelhart and McClelland (1986) might be interpreted as suggesting that a network can learn to treat a given inflection as the “elsewhere” case only if it applies to a much larger class of items than any other inflection. This claim is true of Rumelhart and McClellands (1986) model, which was a two-layer network subject to the computational limitations on networks. of that class (Minsky & Papert, 1969). However, it does not generabe to current models, which have available to them more sophisticated architectures and learning algorithms. In this paper, we explain the basis of the distinction, and demonstrate that given more appropriate architectural assumptions, connectionist models are perfectly capable of learning a default category and generalising as required, even in the absence of superior type frequency.

The Role of Similarity in Hungarian Vowel Harmony: a Connectionist Account

Over the last 10 years, the assimilation process referred to as vowel harmony has served as a test case for a number of proposals in phonological theory. Current autosegmental approaches successfully capture the intuition that vowel harmony is a dynamic process involving the interaction of a sequence of vowels; still, no theoretical analysis has offered a non-stipulative account of the inconsistent behavior of the so-called ‘transparent’, or disharmonic, segments. This paper proposes a connectionist processing account of the vowel harmony phenomenon, using data from Hungarian. The strength of this account is that it demonstrates that the same general principle of assimilation which underlies the behavior of the ‘harmonic’ forms accounts as well for the apparently exceptional ‘transparent’ cases, without stipulation. The account proceeds in three steps. After presenting the data and current theoretical analyses, the paper describes the model of sequential processing introduced by Jordan (1986), and motivates this as a model of assimilation processes in phonology. The paper then presents the results of a series of parametric studies that were run with this model, using arbitrary bit patterns as stimuli. These results establish certain conditions on assimilation in a network of this type. Finally, these findings are related to the Hungarian data, where the same conditions are shown to predict the correct pattern of behavior for both harmonic and transparent vowels.

Semantic and Associative Relations in Adolescents and Young Adults: Examining a Tenuous Dichotomy

The constructs of semantic and associative relatedness have played a prominent role in research on semantic memory because researchers have historically drawn on the distinction between these two types of relations when formulating theories, creating experimental conditions, and explaining empirical results. We argue that the binary distinction between semantics and association is rooted in a fundamental problem in how the two are defined and contrasted. Whereas semantic relatedness has typically been limited to category coordinates, associative relatedness has most often been operationalized using the word association task. We show that meaningful semantic relations between words/concepts certainly extend beyond category coordinates, that word association is driven primarily by meaningful semantic relations between cue and response words, and that non-meaningful, purely associative relations between words generally are not retained in memory. To illustrate these points, we discuss research on semantic priming, picture naming, and the Deese-Roediger-McDermott false memory paradigm. Furthermore, we describe how research on the development of mnemonic skills in adolescents supports our view. That is, adolescents do not learn arbitrary associations between words, but develop elaborative strategies for linking words by drawing on their rich knowledge of events and situations. In other words, adolescents use existing memories of meaningful relations to ground their memories for novel word pairs, even in an associative learning paradigm. The term “semantic memory” is used to refer to people’s memory for concepts and word meanings. An important aspect of understanding semantic memory concerns delineating the ways in which knowledge of word meaning is organized, and as such, a great deal of research has been aimed at providing insight into this issue. A key goal in this regard is to uncover the relations among concepts that are encoded in semantic memory. To this end, the constructs of semantic and associative relations have been central components of theories of the organization of semantic memory, and research comparing the two has provided a substantial amount of informative data that have furthered both theory development and empirical work. However, critical issues remain with regard to how semantic and associative relations have been defined and studied in semantic memory research, and how they might best be defined and studied in future research. In an influential paper on the organization of human memory, Tulving (1972) noted an increased interest among some of his contemporaries in the kind of memory that underlies the seemingly effortless execution of skills such as language processing and memory access. Tulving’s definition of semantic memory still nicely captures some commonly held views: Semantic memory is the memory necessary for the use of language. It is a mental thesaurus, organized knowledge a person possesses about words and other verbal symbols, their meaning and referents, about relations among them, and about rules, formulas, and algorithms for the manipulation of these symbols, concepts, and relations. (p. 386) Tulving also stated that “the relations among items in semantic memory are of much greater variety” (p. 388) than the relations among the contents of episodic memories, which he believed to be organized chiefly along spatio-temporal dimensions. Since that time, a large number of theories and studies have focused on the contrast between semantic and associative relations because they are considered to be the two principle and distinguishable components of conceptual organization (Crutch & Warrington, 2010; Fischler, 1977; Hutchison, 2002; Shelton & Martin, 1992; Thompson-Schill, Kurtz, & Gabrielli, 1998; Yee, Overton, & Thompson-Schill, 2009). It has been a common working hypothesis in semantic memory research that these components are defined on orthogonal dimensions. Associative relatedness is defined typically in terms of stimulus-response combinations in a word association task (e.g., agony-pain; Nelson, McEvoy, & Schreiber, 1998). In fact, Nelson et al.’s word association norms, although not the Semantic & Associative Relations 3 sole source of word association norms in the literature, have been the most often used operationalization of association in memory research for at least the past decade. In contrast, semantic relatedness has typically been defined either as membership in the same superordinate category (e.g., horse-dog; Lupker, 1984), or as the degree to which the semantic features of two concepts overlap (horse-cow; Frenck-Mestre & Beuno, 1999). Often these two measures are treated as essentially the same, and indeed both are based on closeness in a representational structure, although featural overlap is more of a continuous dimension than is shared category. In this chapter, we outline our position concerning the relationship between association and meaning. Association in its general sense spatial and temporal co-occurrence in the world and language is an important driving force in learning, and this includes the formation of semantic representations. Furthermore, word association norms are an interesting and rich source of data. However, word associations on their own provide little if any insight into the relations that are encoded in semantic memory. Performance on word association norms is driven by meaningful semantic relations, and these relations are identifiable, and in many cases, quantifiable. We also argue that is not fruitful to attempt to understand semantic memory using a binary distinction between semantic similarity and word association (or even between semantic relatedness, broadly defined, vs. word association). On the one hand, the scope of semantic relations is much broader than similarity alone, and on the other hand, word associations are driven almost exclusively by semantic relations. Finally, a fruitful research strategy is to work toward understanding the relative importance or centrality of various types of semantic relations for various types of concepts. This approach, we believe, is the best path forward for understanding concepts and semantic memory. To provide evidence for these ideas, and to couch our arguments, we focus on four areas of research in which the semantics-word association dichotomy has played a major role. Section 4.1 deals with experiments regarding picture-word facilitation and interference. Section 4.2 concerns the DeeseRoediger-McDermott false memory paradigm. Section 4.3 focuses on semantic priming. Finally, Section 4.4 describes research concerning how the ability to learn word pairs develops across adolescence, and how this development crucially hinges on semantic knowledge, and the ability to employ that knowledge to make associations meaningful.