Friday, September 30th, 2011 [10:50 - 11:50]
OS_06.1 - Learning semantic representations from sequential and syntactic statistics
Andrews, M. 1, 2 & Vigliocco, G. 2
1 Nottingham-Trent University
2 University College London
In recent years, a common computational approach to the problem of the learning semantic representations has been premised on the hypothesis that aspects of the meaning of words can be inferred from their statistical characteristics across spoken and written language. Well known examples of models of this kind include Latent Semantic Analysis due to Landauer et al. One of the widely shared assumptions of these models, however, has been to treat the linguistic context in which a word occurs as an unordered set of words, and by so doing they disregard fine-grained sequential and syntactic information. In the present work, we will describe a set of Bayesian distributional models that go beyond this so-called "bag-of-words" paradigm. These models avail of information regarding the sequential order in which words occur, the argument structure and general syntactic relationships within sentences, all of which potentially provide vital information about the possible meaning of words. By reference to word-associations norms and experimental behavioural measures of semantic representation in both monosemous and polysemous words, we demonstrate that more precise and psychologically valid semantic representations can be learned when these more fine-grained sources of statistical information are used.
OS_06.2 - Detecting inherent bias in the lexical decision task
Keuleers, E. & Brysbaert, M.
Department of Experimental Psychology, Ghent University, Ghent, Belgium
A basic assumption of the lexical decision task is that a correct response to a word requires access to a corresponding mental representation of that word. However, systematic patterns of similarities and dissimilarities between words and nonwords can introduce inherent biases for a particular response to a given stimulus (e.g., word-stimuli can contain more vowels, nonword stimuli can end frequently with a certain letter). We introduce LD1NN, a simple algorithm based on the Levenshtein Distance (LD) and one-nearest-neighbor classification (1NN), which derives the inherent response bias for each stimulus in an experiment from the distribution of word and nonwords among the most similar previously presented stimuli. We first show that LD1NN is very sensitive to differences between words and matched nonwords generated according to different principles (i.e., random nonword generation, letter replacement, linguistically informed pseudoword generators). Finally, we examine participant data from three lexical decision megastudies and show that the algorithm's predicted biases for and against responses correspond to respectively faster and slower responses to stimuli. The algorithm can be used to examine and limit the degree of inherent bias when designing an experiment and to control for existing bias in statistical analysis of experimental data.
OS_06.3 - A new computational theory of mental imagery
Sima, J. F.
Cognitive Systems, University of Bremen, Germany
The nature of the mental representations and processes underlying human mental imagery has been one the most prominent open questions in cognitive science for decades and still remains unresolved today. We shed new light on this question with a new theory of mental imagery, which is able to integrate the three contemporary theories, i.e., descriptive, enactive, and quasi-pictorial, by providing a consistent explanatory framework for a range of phenomena, which are not covered by one of the other theories on its own. In contrast to the other theories, this new theory is also implemented as a computational cognitive model. We show how the model accounts for common imagery phenomena, e.g., mental scanning, cognitive penetration, eye movements, and mental reinterpretation. We discuss how the structure and components of the model offer a new take on the distinction between visual and spatial mental imagery as well as neuropsychological results, e.g., imaginal neglect.