The Generation Of Bridging Anaphora In Stream-of-consciousness Discourse

This dissertation mainly focuses on the revision and extension of two Rules of standard Centering Theory (Pronoun Rule and Transition Rule) to develop the constraints on the generation algorithm of bridging anaphora (BA) in stream-of-consciousness (SOC) discourse from a revised Centering perspective by incorporating the revised neo-Gricean pragmatic M-principle into the standard Centering.Methodologically, the present study is mainly qualitatively oriented, supported by a self-built stream-of-consciousness corpus. It explicitly targets theoretical description rather than practical application. So it is intended to derive some general heuristic rules to address bridging anaphora generation under a revised Centering Theory, which may predict and explain promising results for the stream-of-consciousness corpus. The standard Centering Theory is a "parametric theory", which is built on local coherence and pronominalization. Its core parameters—backward-looking center (Cb) and forward-looking center (Cf) ranking vary with languages, domains and genres. Standard Centering Theory can not adequately handle language phenomena yet to be explained. For example, in complex fictional narrative—stream-of-consciousness discourse, the interior monologue segment is characterized by associative leaps in syntax and narrative non-linearization. Cb realized as bridging full noun phrases (FNP) and elliptical zero pronouns (EZP) across discourse segment boundaries or between two adjacent utterances seem incoherent, but they are actually coherent in the reader’s mind, which defies the intuition of standard Centering Theory. In other words, on the one hand, standard Centering Theory focuses on pronoun Cb realization and movement across two adjacent utterances in naturally occurring discourse; on the other hand, full noun phrases or elliptical zero pronouns are mainly responsible for Cb realization and movement in stream-of-consciousness discourse. Then, the question is how to develop a Centering-oriented generation algorithm of bridging anaphora based on its generation characteristics, i.e., bridging Cb realization and Cb movement. Hence, in terms of formalization claims of standard Centering, certain aspects of the theory need to be modified and extended if the theory is to provide a more comprehensive discourse-oriented computational account of attentional state, inferential complexity, and ultimately discourse coherence. In light of the above-stated problem, the task of modelling a bridging anaphora generation algorithm can be fulfilled by addressing research questions spelled out below:1. Why can full noun phrases and elliptical zero pronouns be employed as backward-looking centers in stream-of-consciousness discourse?2. What are the discourse functions of full noun phrases and elliptical zero pronouns in stream-of-consciousness discourse?3. How are full noun phrases and elliptical zero pronouns as backward-looking centers generated in stream-of-consciousness discourse?4. How does the revised Transition Rule contribute to stream-of-consciousness discourse segmentation, and how does it relate to coherence?Upon subjecting a stream-of-consciousness corpus to a centering analysis, we can discern that a centering description of the stream-of-consciousness data structures relies heavily on bridging centers realized as full noun phrases and elliptical zero pronouns. Therefore, it is inadequate to capture coherence this corpus displays under standard Centering.The standard Centering Theory describes all of the type "x transition tends to encode the backward-looking center in y form." There are some clear tendencies that can be observed. The main concern of the present study is the realizations that do not follow standard tendencies. The most clearly stated tendency is that the backward-looking center of a COHESIVE1or COHESIVE2transition is realized via full noun phrases and elliptical zero pronouns, which is quite against the expectation previous Centering Theory literature sets up for backward-looking center realization. In stream-of-consciousness discourse, seventy-six percent of the backward-looking centers in COHESIVE1and COHESIVE2are realized as full noun phrases and elliptical zero pronouns. That is, readers can not only rely on expressions but also largely depend on expression-induced pragmatic inferencing to fulfill the backward-looking-center-based information continuity and progression, and to construct discourse local and global coherence. Full noun phrases and elliptical zero pronouns, whose meanings are entailed or evoked by their respective antecedents, result in the highest consensus in favor of a coreferential reading due to the particular discourse functions they serve. The tendency for overspecified/overdetermined marked prolix expressions—full noun phrases or over-underspecified minimal non-conventional marked forms—elliptical zero pronouns to pick out alternative, non-stereotypical or less predictable anchors supports the revised Centering’s prediction in terms of the revised neo-Gricean pragmatic M-principle. According to the revised Centering, the writer’s choice of a marked form tends to implicate an alternative, non-stereotypical, or less predictable interpretation in a marked situation, which would not have been implicated by a more minimal, unmarked expression.The study reveals that full noun phrase bridging anaphora serves different functions both within segments and across segments in stream-of-consciousness discourse. With regard to full noun phrase bridging anaphora within segments, they are mainly used to elaborate, extend and enhance information flow, to put it another way, they are used to incorporate something new into the given entity and establish an entity integrating the given with the new one. In addition, they may be employed for disambiguation, thematic prominence, a topic shift and a signal of a rhetorical relation of contrast, whilst full noun phrase bridging anaphora across segments can serve the functions such as alteration of the participants’perspectives and informationally redundant utterances as segmentation markers. The relation between two indirectly realized backward-looking centers of adjacent utterances can also be mereologically related or frame-related via elliptical zero pronouns. They may be used for topic activation and maintenance, as a signal of a subjective point of view, and to implicate special feelings, attitudes and characters such as hidden fidgets, evasion and cowardice.Centering Theory combined with the revised neo-Gricean pragmatic M-principle is supposed to impose some grammatical and pragmatic constraints on backward-looking center realization and movement. It provides theoretical rationale for an account of constraints on full noun phrase and elliptical zero pronoun bridging anaphora generation in terms of the interaction and the division of labor between syntax and pragmatics. As to both, one can not alone handle bridging anaphora without the other. Despite the fact that syntax and pragmatics operate at distinct levels of linguistic explanation, they appear to interact systematically in the case of bridging anaphora generation. On the one hand, syntax sets certain restrictions on salience and distributions of bridging anaphora, and regulates the part of interpretation which is related to grammatical structure. On the other hand, the choice of bridging anaphoric expressions by writers and their interpretation by readers is heavily dependent on preference which is regulated by the M-principle of language use and communication. Hence, the bridging anaphora generation algorithm based on the incorporation of the revised neo-Gricean pragmatic M-principle of communication into Centering provides a neat and reasonable explanation for bridging full noun phrase and elliptical zero pronoun generation.The revision of Transition Rule of Centering allows us to come closer to understanding and comprehensively modeling coherence in stream-of-consciousness discourse. An evaluation metric comprising a battery of five-test criteria other than the standard version of "salience" and "cohesion" and Kibble’s version of "cohesion","salience","cheapness" and "no backward-looking center", is developed to involve "coherence","salience","cheapness","cohesion" and "no backward-looking center" in measuring the degree of coherence of different transition sequences on the premise of the distinction between coherence and cohesion. The addition of "coherence" and the distinction between the notion of lexical cohesion and cognitive and/or pragmatic coherence are crucial to the characterization of coherence in this corpus, which the Rule2of standard Centering Theory cannot adequately capture. Cohesion mainly dwells on semantic relatedness between two backward-looking centers, which can be resolved in frame semantics. Coherence cares more about relatedness between two backward-looking centers motivated by cognitive and/or pragmatic factors. In other words, two backward-looking centers may be semantically unrelated, but they strike up a relation with each other either temporarily or permanently due to cognitive and/or pragmatic factors.In conclusion, in this bridging anaphora generation algorithm, Centering identifies the center of attention intersententially. That is, the backward-looking center across adjacent utterances, by means of imposing grammatical constraints as filters on the selection of the appropriate backward-looking center of each utterance. Then, the candidate center is further mediated by the revised neo-Gricean pragmatic M-principle. It provides an explicit clue to delving into why the backward-looking center can be indirectly realized as full noun phrases or elliptical zero pronouns, and how different kinds of context are evoked and encoded by different anaphoric expressions, which can be predicted via different backward-looking center realizations.On the basis of a detailed analysis of stream-of-consciousness discourse, one of the most influential discourse-oriented computational models, i.e., standard Centering Theory on anaphora generation is revised. Put differently, another anaphora generation model is added to the standard Centering Theory on anaphora generation. The generation algorithm of bridging anaphora developed in the study is predicated on its generation characteristics in stream-of-consciousness discourse. This bridging anaphora generation algorithm modelling can be counted as a special contribution to the development of discourse-oriented computational linguistics.