This study seeks to find out how human beings cluster Web pages naturally. 20 Web pages retrieved by the Northern Light search engine for each of 10 queries were sorted by 3 subjects into categories that were natural or meaningful to them. It was found that different subjects clustered the same set of Web pages quite differently and created different categories. The average inter-subject similarity of the clusters created was a low 0.27. Subjects created an average of 5.4 clusters for each sorting. The categories constructed can be divided into 10 types. About 1/3 of the categories created were topical. Another 20% of the categories relate to the degree of relevance or usefulness. The rest of the categories were subject-independent categories such as format, purpose, authoritativeness and direction to other sources. The authors plan to develop automatic methods for categorizing Web pages using the common categories created by the subjects. It is hoped that the techniques developed can be used by Web search engines to automatically organize Web pages retrieved into categories that are natural to users.

Jerry Fodor has argued that virtually all lexical concepts are innate. I argue against this position, but not, as other have done, on the grounds that the arguments against lexical decomposition upon which Fodor relies are flawed. Rather, I argue that even if lexical concepts cannot be decomposed, the possession conditions for having lexical concepts are nonetheless not innately satisfied.

“Hey can you help me move? I warn you I have a lot of books, though.” When we interpret such sentences we might assume that our friend is implying that there will be some heavy lifting, because she own “a lot of books”. If you’re opposed to heavy lifting, you probably want to know how many books your friend wants you to help her move. While it is easy to determine the quantities picked out by numerals, discovering the meaning of quantifiers, such as “a lot”, is less clear. For example, in a survey people gave different quantities for “a few” and “a lot” depending on the context they were asked about, choosing to give quantities as high as 76 for “a few friends on Facebook” and as low as 4 for “a lot of houses”. I ask what role context plays in these interpretations, and propose two possible hypotheses: the Fixed Quantities Hypothesis, and the Relative Quantities Hypothesis. The Fixed Quantities Hypothesis assumes that these terms pick out an exact range of quantities (e.g., 3-5 for “a few”) and that the effect of context is due to people being pragmatically generous in their interpretations (Leech, 1983; Brown & Levinson, 1987; Lasersohn, 1999). The Relative Quantities Hypothesis instead argues that the ranges picked out by each of the terms is relative to the context, and therefore these terms might be similar to relative gradable adjectives (e.g., the meaning of the word “tall” is dependent on the reference set, e.g., what counts as “tall” is different for a mug versus a building; Kennedy, 2007; Syrett, Kennedy, & Lidz, 2010). I attempt to tease apart these two theories by looking at whether people find certain quantities implausible for some contexts leading them to think a speaker is being sloppy (Experiment 1), whether the context effect persists even under light-to-no pragmatic pressure (Experiment 2), and whether we find the effect of context even with minimal knowledge of novel contexts (Experiment 3). The results favour the Relative Quantities Hypothesis, and I discuss potential future work investigating the role of distributional knowledge on quantifier mappings.

No description available.

No description available.

No description available.

The ability to reason about categories and category membership is fundamental to human cognition, and as a result a considerable amount of research has explored the acquisition and modelling of categorical structure from a variety of perspectives. These range from feature norming studies involving adult participants (McRae et al. 2005) to long-term infant behavioural studies (Bornstein and Mash 2010) to modelling experiments involving artificial stimuli (Quinn 1987). In this thesis we focus on the task of natural language categorisation, modelling the cognitively plausible acquisition of semantic categories for nouns based on purely linguistic input. Focusing on natural language categories and linguistic input allows us to make use of the tools of distributional semantics to create high-quality representations of meaning in a fully unsupervised fashion, a property not commonly seen in traditional studies of categorisation. We explore how natural language categories can be represented using distributional models of semantics; we construct concept representations for corpora and evaluate their performance against psychological representations based on human-produced features, and show that distributional models can provide a high-quality substitute for equivalent feature representations. Having shown that corpus-based concept representations can be used to model category structure, we turn our focus to the task of modelling category acquisition and exploring how category structure evolves over time. We identify two key properties necessary for cognitive plausibility in a model of category acquisition, incrementality and non-parametricity, and construct a pair of models designed around these constraints. Both models are based on a graphical representation of semantics in which a category represents a densely connected subgraph. The first model identifies such subgraphs and uses these to extract a flat organisation of concepts into categories; the second uses a generative approach to identify implicit hierarchical structure and extract an hierarchical category organisation. We compare both models against existing methods of identifying category structure in corpora, and find that they outperform their counterparts on a variety of tasks. Furthermore, the incremental nature of our models allows us to predict the structure of categories during formation and thus to more accurately model category acquisition, a task to which batch-trained exemplar and prototype models are poorly suited.

This research is concerned with texture: a source of visual information, that has motivated a huge amount of psychophysical and computational research. This thesis questions how useful the accepted view of texture perception is. From a theoretical point of view, work to date has largely avoided two critical aspects of a computational theory of texture perception. Firstly, what is texture? Secondly, what is an appropriate representation for texture? This thesis argues that a task dependent definition of texture is necessary, and proposes a multi-local, statistical scheme for representing texture orientation. Human performance on a series of psychophysical orientation discrimination tasks are compared to specific predictions from the scheme. The first set of experiments investigate observers' ability to directly derive statistical estimates from texture. An analogy is reported between the way texture statistics are derived, and the visual processing of spatio-luminance features. The second set of experiments are concerned with the way texture elements are extracted from images (an example of the generic grouping problem in vision). The use of highly constrained experimental tasks, typically texture orientation discriminations, allows for the formulation of simple statistical criteria for setting critical parameters of the model (such as the spatial scale of analysis). It is shown that schemes based on isotropic filtering and symbolic matching do not suffice for performing this grouping, but that the scheme proposed, base on oriented mechanisms, does. Taken together these results suggest a view of visual texture processing, not as a disparate collection of processes, but as a general strategy for deriving statistical representations of images common to a range of visual tasks.

No description available.

No description available.