Discriminative learning and the lexicon: NDL and LDL

Word Knowledge and Word Usage

Over the last decades, a growing body of evidence on the mechanisms governing lexical storage, access, acquisition and processing has questioned traditional models of language architecture and word usage based on the hypothesis of a direct correspondence between modular components of grammar competence (lexicon vs. rules), processing correlates (memory vs. computation) and neuro-anatomical localizations (prefrontal vs. temporo-parietal perisylvian areas of the left hemisphere). In the present chapter, we explore the empirical and theoretical consequences of a distributed, integrative model of the mental lexicon, whereby words are seen as emergent properties of the functional interaction between basic, language-independent processing principles and the language-specific nature and organization of the input. From this perspective, language learning appears to be inextricably related to the way language is processed and internalized by the speakers, and key to an interdisciplinary understanding of such a way, in line with Tomaso Poggio's suggestion that the development of a cognitive skill is causally and ontogenetically prior to its execution (and sits "on top of it"). In particular, we discuss conditions, potential and prospects of the epistemological continuity between psycholinguistic and computational modelling of word learning, and illustrate the yet largely untapped potential of their integration. We use David Marr's hierarchy to clarify the complementarity of the two viewpoints. Psycholinguistic models are informative about how speakers learn to use language (interfacing Marr's levels 1 and 2). When we move from the psycholinguistic analysis of the functional operations involved in language learning to an algorithmic description of how they are computed, computer simulations can help us explore the relation between speakers' behavior and general learning principles in more detail. In the end, psycho-computational

Lecture Notes in Computer Science, 1995

Current approaches to computational lexicology in language technology are knowledge-based (competence-oriented) and try to abstract away from specific formalisms, domains, and applications. This results in severe complexity, acquisition and reusability bottlenecks. As an alternative, we propose a particular performance-oriented approach to Natural Language Processing based on automatic memory-based learning of linguistic (lexical) tasks. The consequences of the approach for computational lexicology are discussed, and the application of the approach on a number of lexical acquisition and disambiguation tasks in phonology, morphology and syntax is described. * Preprint of a paper to be published in Steffens (ed.) Machine Translation and the Lexicon. Springer Lecture Notes in Artificial Intelligence. I would like to thank my colleagues in the Atila project (Steven Gillis, Gert Durieux, and Antal van den Bosch) for their contributions to the approach described in this paper. The Atila (Antwerp-Tilburg Inductive Language Acquisition) project is a research corporation between the University of Antwerp and Tilburg University focusing on the application of Machine Learning techniques in linguistic engineering and in developmental psycholinguistics. Thanks also to the participants of the Heidelberg workshop on Machine Translation and the Lexicon for useful comments and suggestions.

Developmental Science, 2009

Language Learning, 2020

Chuang, Y-Y., M. J. Bell, I. Banke, and R. H. Baayen This study addresses the question of whether there is anything special about learning a third language, as compared to learning a second language, just by virtue of the third language being the third language acquired, and independently of the specific properties of the third language. We used computational modeling to explore this question for the learning of a small vocabulary of some 400 words, with English as L1, German or Mandarin as L2, and Mandarin and alternatively Dutch, as L3. For computational modeling, we made use of the mathematical framework of linear discriminative learning, which we extended with the learning rule of Widrow-Hoff to enable the modeling of incremental learning of the mappings between form and meaning when words' meanings are represented by vectors of real numbers (embeddings) rather than by abstract symbolic units. A series of simulation experiments covering single-language learning, bilingual learning, and finally trilingual learning, clarified that within the framework of discrimination learning, within-language homophones give rise to frailty in comprehension that in turn for production gives rise to semantic errors in L1, and language intrusions in L2 and L3. Our model correctly predicts production to lag behind comprehension in learning, and it clarified that, within the boundaries of discrimination learning, the properties of the L3 crucially determine whether L3 learning appears to involve a language that is 'dormant' with respect to L1 and L2. Qualitatively surprisingly different patterns of acquisition of the L3, and its interactions with L1 and L2, can arise in our simulations without any changes in the mathematics driving learning. Our simulations also show that when words' forms incorporate not only segmental but also suprasegmental information, the nature of errors that arise in production changes. In the general discussion, we reflect on the implications of our findings for the question of what is special about multilingualism.

PloS one, 2017

In this study we present a novel set of discrimination-based indicators of language processing derived from Naive Discriminative Learning (ndl) theory. We compare the effectiveness of these new measures with classical lexical-distributional measures-in particular, frequency counts and form similarity measures-to predict lexical decision latencies when a complete morphological segmentation of masked primes is or is not possible. Data derive from a re-analysis of a large subset of decision latencies from the English Lexicon Project, as well as from the results of two new masked priming studies. Results demonstrate the superiority of discrimination-based predictors over lexical-distributional predictors alone, across both the simple and primed lexical decision tasks. Comparable priming after masked corner and cornea type primes, across two experiments, fails to support early obligatory segmentation into morphemes as predicted by the morpho-orthographic account of reading. Results fit w...

Neural Network World, 2003

In this paper, we describe a self-organizing neural network model that addresses the process of early lexical acquisition in young children. The growing lexicon is modeled by combined semantic word representations based on distributional statistics of words and on grounded semantic features of words. Changing semantic word representations are assumed to model the maturation of word meaning and serve as inputs to the growing semantic map. The model has been tested on a real child-directed parental language ...

Psychological …, 2011

A two-layer symbolic network model based on the equilibrium equations of the Rescorla-Wagner model is proposed. The study starts by presenting two experiments in Serbian, which reveal for sentential reading the inflectional paradigmatic effects previously observed by Milin, Filipović Durdević, and Moscoso del Prado Martín (2009) for unprimed lexical decision. The empirical results are successfully modeled without having to assume separate representations for inflections or data structures such as inflectional paradigms. In the next step, the same naive discriminative learning approach is pitted against a wide range of effects documented in the morphological processing literature. Frequency effects for complex words as well as for phrases ) emerge in the model without the presence of whole-word or whole-phrase representations. Family size effects (Schreuder & Baayen, 1997; Moscoso del Prado Martín, Bertram, Häikiö, Schreuder, & Baayen, 2004) emerge in the simulations across simple words, derived words, and compounds, without derived words or compounds being represented as such. It is shown that for pseudo-derived words no special morpho-orthographic segmentation mechanism as posited by Rastle, Davis, and New is required. The model also replicates the finding of , that, on average, words with more productive affixes elicit longer response latencies, while at the same time predicting that productive affixes afford faster response latencies for new words. English phrasal paradigmatic effects modulating isolated word reading are reported and modelled, showing that the paradigmatic effects characterizing Serbian case inflection have cross-linguistic scope.

PLoS ONE, 2013

We propose a new model-based approach linking word learning to the age of acquisition (AoA) of words; a new computational tool for understanding the relationships among word learning processes, psychological attributes, and word AoAs as measures of vocabulary growth. The computational model developed describes the distinct statistical relationships between three theoretical factors underpinning word learning and AoA distributions. Simply put, this model formulates how different learning processes, characterized by change in learning rate over time and/or by the number of exposures required to acquire a word, likely result in different AoA distributions depending on word type. We tested the model in three respects. The first analysis showed that the proposed model accounts for empirical AoA distributions better than a standard alternative. The second analysis demonstrated that the estimated learning parameters well predicted the psychological attributes, such as frequency and imageability, of words. The third analysis illustrated that the developmental trend predicted by our estimated learning parameters was consistent with relevant findings in the developmental literature on word learning in children. We further discuss the theoretical implications of our model-based approach.