Search

The general theme of this dissertation is meaning and paradox. Or, more precisely, how accounts of the metaphysical determination of meaning are affected by the semantic paradoxes. I focus in particular on those accounts within the Davidsonian tradition-whose key characteristic, roughly speaking, I take to be a commitment to Davidsonian semantics. The question that drives the discussion is as follows. Which (if any) accounts of meaning within the Davidsonian tradition are compatible with the fact that semantic paradoxes arise in natural language? Throughout the dissertation, I use the liar paradox as an archetypal semantic paradox. I consider first the two central extant accounts of meaning within the Davidsonian tradition: the Davidsonian and neo-Davidsonian accounts of meaning. I argue that, in light of the semantic paradoxes, both accounts should be rejected. As a result of the discussion, I then consider, in my terminology, traditional externalist accounts of meaning. I argue that an account of meaning along the lines of David Lewis' interpretationism can be adopted by the fan of Davidsonian semantics, and that the resultant account may be compatible with the fact that semantic paradoxes arise in natural language. I show how the fan of Davidsouian semantics can adopt Lewis' interpretationism. I call the resultant account the DL-hybrid account of meaning. Roughly, the account permits the adoption of a broadly Kripkean response to the paradox: truth conditions for sentences are to be characterised as conditional upon the sentences in question being grounded. I close by arguing that the DL-hybrid account of meaning is compatible with the fact that semantic paradoxes arise in natural language.

In this work, a comprehensive study of semantic annotation has been carried out in early stage. The study focuses on the annotation requirements of human knowledge acquisition in knowledge engineering, e-Iearning and computational linguistics. Based on findings from the study, annotation of natural languages for linguistic analysis creates complicated data structures. Due to the complexity, almost all existing annotation schemes are designed to support only one application domain at one instance. Discovery of new knowledge by means of cross-domains text analysis is limited by the capability of these annotation schemes. To realize the findings in the study and provide solution to the problem, a new general-purpose annotation archival scheme has been developed but not limited to (1) Enable true cross-domain data analysis in knowledge engineering, e-Learning and computational linguistics, and (2) Organize complex structure of human knowledge annotation in an accessible manner, so they can be analyzed in multiple layers through retrieval, search, visualization and etc. To further verify the contributions of the new semantic annotation scheme in real application, experiments has been carried out in several areas, namely (1) collaborative retrieval of complex linguistic information, (2) computer-assisted production of learning material and (3) relevancy comparison between text. In (1), the annotation scheme is applied in a cloud-based platform for hosting parallel multilingual corpora leading to new applications such as computer assisted pattern visualization, speech analysis, speech-to-text transcription and statistical analysis. In (2), the annotation scheme provides support to applications that produce reader friendly learning material suites for teacher, and as a result improve learning quality. In (3), the annotation scheme provides support to a text' comparison platform that carries out writing assessment semantically. XIII

Semanticists, philosophers and logicians have standardly analyzed a range of natural language expressions on the model of the variables of formal languages. This dissertation explores variable- based semantic theories of pronounsand proper names. The first two chapters argue that a variable-based, presuppositional semantics for pronouns proposed by Cooper (1979, 1983) resolves an apparent trilemma arising from the conjunction of three plausible commitments concerning the semantics of pronouns: that some anaphoric pronouns with quantifier antecedents are bound variables, that referential pronouns have context-independent meanings, and that the relation between bound and referential pronouns is not ambiguity or homonymy. The first chapter argues that Cooper’s semantics is descriptively and empirically superior to alternative theories, including Kaplan’s account, which does not resolve the trilemma. The second addresses an important objection to the view that bound and referential pronouns make the same semantic contribution, involving cases where bound pronouns do not appear to trigger semantic presuppositions. I argue that this appearance is misleading: bound pronouns do trigger semantic presuppositions and have the same (Cooper-style) denotations as referential pronouns. The third chapter addresses the phenomenon of non c-command or ‘donkey’ anaphora. These anaphors appear to require a departure from Cooper’s semantics and, more generally, from the view that anaphoric pronouns correspond to (classical) bound variables. I consider D-type accounts of donkey anaphora and argue that these accounts must be revised in ways that bring the denotations of donkey anaphors closer to those of ordinary bound and referential pronouns. The last chapter asks whether proper names require a variable-based semantics. I consider a pair of recent anaphoric theories of proper names, according to which proper names have anaphoric uses and are thus better modeled as variables. I argue that there are important reasons to reject a variable-based model for proper names, in contrast with pronouns.

This thesis explores the processing of lexical ambiguity: words with several unrelated meanings (homonymy) or many related senses (polysemy). Chapter I provides a literature overview of studies investigating this topic. Chapters 2 and 3 pursue a first goal: to investigate whether effects are influenced by the methodology of defining lexical ambiguity. The results support the hypothesis (Rodd, Gaskell, & Marslen- Wilson, 2002) that studies using questionnaires to define lexical ambiguity (e.g. Rubenstein, Garfield, & Millikan, 1970) found a polysemy advantage rather than a homonymy advantage. Questionnaire-based ambiguity classifications are more similar to dictionary-defined polysemy than to homonymy (Experiment 1). Moreover, earlier findings (e.g. Rodd et al., 2002) of a polysemy advantage and homonymy disadvantage are replicated, and the questionnaire-based classifications result in effects more similar to the former than to the latter (Experiments 2 to 4). Chapters 4 to 6 pursue a second goal: to explore the effects of polysemy and homonymy with new stimuli. Chapters 4 and 5 indicate that polysemy effects are sensitive to concreteness (Experiments 5 & 6), frequency (Experiment 7), and level of processing (Experiment 8). Furthermore, polysemy effects seem to take place relatively late (Experiment 9). In contrast, Chapter 6 does not find any effects of homonymy (Experiments 10 to 12). Chapter 7 pursues a third goal: to test whether the relationship between senses plays a role in word processing. Sense relationship influences word recognition (Experiments 13 & 16), but not semantic categorization (Experiment 14). The temporal locus of the lexical decision effect cannot be determined (Experiment 15). Finally, Chapter 8 shows that the current findings fit reasonably well within an account by Rodd, Gaskell, and Marslen- Wilson (2004), and suggests possible directions for further research.

Background – English verbs are linguistically more complex than nouns and this has contributed to the dearth of in-depth investigation into similarities and differences between their representations within semantic memory and subsequent implications for language processing. However, recent theoretical accounts have argued that verbs and nouns are represented within a unitary semantic system. Aims – This thesis investigates the semantic representations of English verbs with particular attention to how verbs are inter-related as a consequence of semantic similarity. This is achieved through a series of psycholinguistic experiments with healthy adult speakers and an intervention study with adults with aphasia (i.e. acquired communication impairment). Throughout the thesis, comparisons are made to the semantic representations of nouns either directly (i.e. through parallel experimentation) or indirectly (i.e. through the existing literature). Methods – The experiments conducted with healthy adult speakers included: (1) category listing of verbs; (2) typicality rating of verbs within categories; (3) similarity rating of verb pairs; (4) an analysis of verbs’ semantic features; (5) category verification of verbs; and (6) semantically primed picture naming of actions. The intervention study carried out with adults with aphasia compared patterns of improvement in verb and noun retrieval following a semantically-based therapy task. Results and discussion – The results of the experiments shed light on the nature of semantic representations of verbs, in particular, in relation to the similarity between the semantic representations of verbs and those of nouns and also where they differ. These insights are considered in terms of how they provide evidence for or against a unitary semantic system for verbs’ and nouns’ semantic representations and parallel mechanisms for accessing these representations. Two themes emerged in terms of future research potential: (1) the influence of polysemy on speaker’s performance in psycholinguistic tasks; and (2) the nature and influence of typicality within categories/cluster of verbs.

In the field of formal semantics, it has become increasingly difficult to ignore that Montague grammar does well enough in formally interpreting certain features of natural languages. However, the question of 'adequacy' of this grammatical approach to the formal semantic paradigm is still one of the most significant current discussions in philosophy of language and linguistics alike. Within the formal semantic paradigm, the test of adequacy for Montague grammar (with respect to Arabic) is to demonstrate syntactic compositionality at a level which guarantees semantic compositionality via a formal interpretation. The main aim of my thesis has been to examine the adequacy of Montague grammar to discharge the requirements of the formal semantic paradigm. The central focus of the thesis has been an attempt to apply Montague grammar to Arabic quantification as a case study. The method adopted for this investigation is to use a 'fragment' of Arabic quantification. In my fragment, random expressions of Arabic quantification have been selected from Arabic textbooks. One of the more significant findings is that Montague grammar 'succeeds' in providing a compositional interpretation of a fragment of Arabic quantification. On this basis, I argue that Montague grammar is adequate to the local requirements of formal semantic paradigm. I argue that, since Montague grammar is shown to be applicable to another significant kind of case (i.e., Arabic, as an example from another language group), this increases the evidence for the hypothesis that Montague grammar is universally applicable. I argue that, because we have increased evidence for the claim that there is a universal compositional grammar, we have increased evidence for the hypothesis that all natural languages are compositional, and share a common compositional core. The most interesting non-philosophical implication of the study is that the extended Montagovian approach strengthens its position among the varieties of semantic theory for having proved applicable in another language group. I conclude by arguing that the applicability of Montague grammar merits consideration in new cases.

Hoey's theory of Lexical Priming (LP) attempts to provide a theoretical framework to explain the long-established phenomena of collocation (cf. Firth 1957; Sinclair 1991; Stubbs 1996), colligation (cf. Firth 1957; Sinclair 2004, Hunston 2001), and several kinds of semantic relationships (cf. Sinclair 1991, 2004; Hunston 2001; Louw 1993; Partington 2004). The framework generates a number of hypotheses which call for closer examination. One of them concerns the phenomenon of polysemy. According to LP, the collocations, semantic associations and colligations that a polysemous word is characteristically primed for will systematically differentiate its various senses (Hoey 2005: 81). The claim is further that the different senses avoid use of each other's collocations, colligations and semantic associations. The formulation of the claims of LP regarding polysemy was based on polysemous nouns with two or more abstract senses each (i.e. consequence, reason, immunity) on the one hand, and on the word dry looked at mainly in its use as an adjective, on the other. The present study aims to explore whether the claims of LP with respect to polysemy are equally applicable when other types of polysemous words are investigated. Research consisted in two case studies based on data drawn from a subcorpus of the British National Corpus (BNC) comprising fiction texts. The corpus consists of 432 texts amounting to some 15,000,000 words. The first case study concerned the two most frequent senses -both concrete- of the noun drive (viz. "journey" and "private road"). Findings from this case study confirm the claims of LP. The second case study focused on the senses of the polysemous verb face. Face was chosen for having senses that can be classified along the cline concrete - abstract. There were two main reasons behind this choice: the first was that it was deemed necessary to examine whether the formulation of the claims of LP with respect to polysemy may have somehow been affected by the grammatical category of the polysernous words examined so far; verb meaning relies considerably on a verb's participants and this might generate new insights -and possible challenges- regarding the claims of LP with respect to polysemy. The second reason related to the classifiability of a word's senses along the concrete-abstract cline, and to the question of in how far the claims of LP -as currently formulated- fully account for such cases of polysemy as well. The verb face has a concrete use and an abstract use, illustrated by I do not like coasts that face north and I had to face al/ my problems alone, respectively, and it has uses that appear to merge the two to varying degrees, as seen in the examples below: (1) of little consequence." He turned her to face him. "If I gave that impression then I ea (2) This was yet another of his children to face him in defiance in this very kitchen. Jon (3) not realise then that Rioja had stayed to face the killers. From the hut behind them the Given the claims of LP regarding polysemy, we would expect (i) that face will have primings that clearly distinguish not only between its concrete and abstract uses, but also (ii) that there will be primings that make a distinction among the sub-senses of the concrete, or "directional" use. While the analysis in part yielded results in support of LP claims, it also showed that collocation, colligation and semantic association are not always sufficient for the disambiguation of meaning. Indeed, the nature of the data, very often involving dialogue, called for an application of Burton's (1980) framework for the analysis of casual conversation. Evidence suggests that, beyond the phenomena of collocation, colligation and semantic association, categories from the analysis of casual conversation -specifically, Challenging Moves- can be instrumental in a systematic differentiation between the senses of polysemous words.

One of the fundamental aspects of any natural language is the set of words used within it. In addition to knowing how individual words can be combined to communicate meaning, competent language users also know a large number of specific word combinations whose grammatical or distributional behaviour or meaning is idiosyncratic. This research is concerned with computational aspects of one important type of word combination: collocation. There is no agreed formal definition of collocation but it can be informally characterised as a sequence of words that occurs more often than would be expected by chance and whose combination tends to produce an element of added meaning. One of the often-cited characteristics of collocations is that they restrict substitution for their constituent words. This thesis develops a generic framework for the extraction of collocations that exploits this restriction. Experiments exploring the performance of such techniques use frequency counts derived from the WWW as well as large amounts of analysed text from conventional corpora and show that substitution-based techniques can out-perform many existing approaches to collocation extraction. The thesis concludes with a discussion of the many ways in which further research can leverage the genericity of the framework and utilise substitution for collocation extraction.

In recent years, contradictory advice to teachers has been emerging from studies into the use of semantic links or networks in classroom materials and activities for vocabulary learning in a L2. There is some experimental evidence which suggests that learning semantically related words (e.g., body parts) at the same time makes learning more difficult (Tinkham, 1993, 1997; Waring, 1997; Finkbeiner, Nicol, 2003). There is also a theoretical framework that strongly supports the idea that it is very useful to present words of related meaning together so that learners can see the distinctions between them and gain a complete coverage of the defined area of meaning (Channell, 1981, 1990; Neuner, 1992; Dunbar, 1992). The following paradox appears: while the experimental evidence suggests that semantically related vocabulary does not help vocabulary learning, the EFL coursebook-writers present vocabulary in semantic clusters. The experimental evidence mainly derives from research using artificial language and not a natural L2. The purpose of our research is to investigate which of the two contrasting views will prove to be a useful tool in L2 vocabulary learning. The present study was influenced by action research. It was conducted in EFL classrooms with natural learners in Greece. The subjects were 31 intermediate EFL children and 32 beginners EFL adults. Two different ways of organizing new vocabulary for presentation were employed: a) presenting semantically related words (topic-related vocabulary i.e. mugging, terrorism,jorgery, synonyms, antonyms or homonyms) together at the same time, and b) presenting vocabulary in an unrelated fashion (i.e. carpenter, tornado, sage). Short and long-term tests were administered to the students. The presentation will focus on the main conclusion that semantically related vocabulary impedes L2 vocabulary learning. Adult beginners performed significantly better on the unrelated vocabulary test compared to their performance on the related vocabulary test. Word frequency (in language) when combined with unrelated presentation of new L2 vocabulary appears to make a difference in students' performance.

Requirements tracing is a universally neglected practice in industry, despite clear knowledgement that it supports high-end practices, such as change management and control and impact analysis. Requirements tracing is comprised of two principal types; pre- and post-requirement specification tracing. Post-requirement.specificadon tracing, or post-RST, is concerned with tracing requirements after they have already bncluded in the specification. Pre-RST is concerned with life before inclusion, and therefore represents the origin of each requirement. Pre-RST is less often implemented in practice than post-RST. The principal reason for this is that equirements tracing of any kind doesn't appear to offer enough tangible benefits to the organisation that is developing the software to make it worthwhile, especially given the high cost of tracing by hand. Automating the process of pre-RST is therefore likely to increase its appeal to practitioners by significantly reducing the cost ing pre-RST.

How should we best analyse the meaning of proper names, indexicals, demonstratives, both simple and complex, and definite descriptions? In what relation do such expressions stand to the objects they designate? In what relation do they stand to mental representations of those objects? Do these expressions form a semantic class, or must we distinguish between those that arc referential and those that are quantificational? Such questions have constituted one of the core research areas in the philosophy of language for much of the last century, yet consensus remains elusive: the field is still divided, for instance, between those who hold that all such expressions are semantically descriptive and those who would analyse most as the natural language counterparts of logical individual constants. The aim of this thesis is to cast new light on such questions by approaching them from within the cognitive framework of Sperber and Wilson's Relevance Theory. Relevance Theory offers not just an articulated pragmatics but also a broad conception of the functioning of natural language which differs radically from that presupposed within (most of) the philosophy of language. The function of linguistic expressions, on this conception, is not to determine propositional content, but rather to provide inferential premises which, in parallel with context and general pragmatic principles, will enable a bearer to reach the speaker's intended interpretation. Working within this framework, I shall argue that the semantics of the expressions discussed should best be analysed not in terms of their relation to those objects which, on occasions of use, they may designate, but rather in terms of the indications they offer a hearer concerning the mental representation which constitutes the content of a speaker's informative intention. Such an analysis can, I shall claim, capture certain key data on reference which have proved notoriously problematic, while respecting a broad range of apparently conflicting intuitions.

This thesis investigated the ability to process semantic information from foveal and extrafoveal vision during scene viewing. Existing research suggested that object semantics could be detected from extrafoveal vision. This suggestion was investigated using three experimental paradigms. Semantic inconsistency was defined as a target semantically incompatible with scene gist. In Experiments 1 to 4, fixation position during a brief scene presentation was manipulated relative to a target object. The target's semantic inconsistency, presented foveally or extrafoveally, influenced performance on an object identification task. Extrafoveally presented semantically inconsistent targets were facilitated when simple line drawings were displayed, although this effect was unlikely to be mediated by semantic processing. No similar effect was found with complex line drawings or photographic stimuli. Experiments 5 and 6 attempted to replicate significant advantages for inconsistent targets in a change detection paradigm. However, no significant difference was found between performance for consistent and inconsistent targets in a two-exposure, forced-choice change detection task or an alternating display change detection task. There was no evidence that changing inconsistent targets were detected more reliably or earlier than changing consistent targets. Experiment 7 investigated the proposal that the extrafoveal processing of inconsistent objects could influence saccade patterns by attracting earlier fixations. Participants freely scanned both line drawings and photographs of scenes with no task. Again, no evidence was found supporting the earlier fixation of inconsistent objects in scenes. Therefore, this thesis could not confirm previous evidence of an inconsistent object advantage in either brief scene presentations, change detection or natural scene viewing. The evidence suggested that the preferential processing of inconsistent scene objects could occur under very limited circumstances, but would be unlikely to be mediated by semantic processing. When viewing complex, realistic scenes, there was no evidence of differential processing for consistent and inconsistent objects.

In Part One, three experiments investigated the effects of the surface structure of semantic illusion sentences upon semantic illusion rate (Chapters 3 to 6), but only a comparison of question and statements revealed any significant effects, with questions leading to more semantic illusion responses. To explore the implications of this lack of effect, a rating scale study was designed to provide an overview of how semantic illusion sentences compare to sentences used in ordinary discourse: semantic illusion type sentences were found to differ significantly from other sentences along a number of salient dimensions. In Part Two, three further experiments related semantic illusions to problem solving and examined the processing requirements of semantic illusions. Findings indicated that semantic illusions are subject to a kind of ‘functional fixedness’, which prevents thorough processing (Chapters 9 and 10). This may in part be explained by the load that semantic illusion sentences place on working memory, as was indicated by the results of two further experiments, which investigated the role that the different components of working memory play in semantic illusion processing (Chapter 11).

This thesis concerns the problem of complexity in operational semantics definitions. The appeal of modern operational semantics is the simplicity of their metatheories, which can be regarded as theories of deduction about certain shapes of operational judgments. However, when applied to real programming languages they produce bulky definitions that are cumbersome to reason about. The theory of interacting deductions is a richer metatheory which simplifies operational judgments and admits new proof techniques. An interacting deduction is a pair (F, I), where F is a forest of inference trees and I is a set of interaction links (a symmetric set of pairs of formula occurrences of F), which has been built from interacting inference rules (sequences of standard inference rules, or rule atoms). This setting allows one to decompose operational judgments. For instance, for a simple imperative language, one rule atom might concern a program transition, and another a store transition. Program judgments only interact with store judgments when necessary: so stores do not have to be propagated by every inference rule. A deduction in such a semantics would have two inference trees: one for programs and one for stores. This introduces a natural notion of modularity in proofs about semantics. The proof fragmentation theorem shows that one need only consider the rule atoms relevant to the property being proved. To illustrate, I give the semantics for a simple process calculus, compare it with standard semantics and prove three simple properties: nondivergence, store correctness and an equivalence between the two semantics. Typically evaluation semantics provide simpler definitions and proofs than transition semantics. However, it turns out that evaluation semantics cannot be easily expressed using interacting deductions: they require a notion of sequentiality. The sequential deductions contain this extra structure. I compare the utility of evaluation and transition semantics in the interacting case by proving a simple translation correctness example. This proof in turn depends on proof-theoretic concerns which can be abstracted using dangling interactions. This gives rise to the techniques of breaking and assembling interaction links. Again I get the proof fragmentation theorem, and also the proof assembly theorem, which allow respectively both the isolation and composition of modules in proofs about semantics. For illustration, I prove a simple type-checking result (in evaluation semantics) and another nondivergence result (in transition semantics). I apply these results to a bigger language, CSP, to show how the results scale up. Introducing a special scoping side-condition permits a number of linguistic extensions including nested parallelism, mutual exclusion, dynamic process creation and recursive procedures. Then, as an experiment I apply the theory of interacting deductions to present and prove sound a compositional proof system for the partial correctness of CSP programs. Finally, I show that a deduction corresponds to CCS-like process evaluation, justifying philosophically my use of the theory to give operational semantics. A simple corollary is the undecidability of interacting-deducibility. Practically, the result also indicates how one can build prototype interpreters for definitions.