Thesis (M.A.)--University of Hong Kong, 1988. / Also available in print.

Thesis (D.A.)--Illinois State University, 1990. / Title from title page screen, viewed November 15, 2005. Dissertation Committee: Margaret Steffensen (chair), Irene Brosnahan, Bruce Hawkins, Maurice Scharton, Sandra Metts. Includes bibliographical references (leaves 211-218) and abstract. Also available in print.

Thesis (Ph. D.)--University of Texas at Austin, 2007. / Vita. Includes bibliographical references.

Thesis (M. Ed.)--University of Hong Kong, 1997. / Includes bibliographical references (leaf 81-84).

Thesis (M.Ed.)--University of Hong Kong, 1997. / Includes bibliographical references (leaves 81-84). Also available in print.

No description available.

Understanding the process of language acquisition is a challenge that many researchers spanning different disciplines (e.g. linguistics, psychology, neuroscience) have grappled with for centuries. One which has in recent years attracted a lot of attention has been in the area of non-native phoneme acquisition. Speech sounds that contain multiple phonetic cues are often difficult for foreign-language learners, especially if certain cues are weighted differently in the foreign and native languages. Greek adult and child speakers of English were studied to determine which cues (duration or spectral) they were using to make discrimination and identification judgments for an English vowel contrast pair. To this end, two forms of identification and discrimination tasks were used: natural (unedited) stimuli and another ‘modified’ vowel duration stimuli which were edited so that there were no duration differences between the vowels. Results show the Greek speakers were particularly impaired when they were unable to use the duration cue as compared to the native English speakers. Similar results were also obtained in control experiments where there was no orthographic representation or where the stimuli were cross-spliced to modify the phonetic neighborhood. Further experiments used high-variability training sessions to enhance vowel perception. Following training, performance improved for both Greek adult and child groups as revealed by post training tests. However the improvements were most pronounced for the child Greek speaker group. A further study examined the effect of different orthographic cues that might affect rhyme and homophony judgment. The results of that study showed that Greek speakers were in general more affected by orthography and regularity (particularly of the vowel) in making these judgments. This would suggest that Greek speakers were more sensitive to irrelevant orthographic cues, mirroring the results in the auditory modality where they focused on irrelevant acoustic cues. The results are discussed in terms of current theories of language acquisition, with particular reference to acquisition of non-native phonemes.

Active learning reduces annotation costs for supervised learning by concentrating labelling efforts on the most informative data. Most active learning methods assume that the model structure is fixed in advance and focus upon improving parameters within that structure. However, this is not appropriate for natural language processing where the model structure and associated parameters are determined using labelled data. Applying traditional active learning methods to natural language processing can fail to produce expected reductions in annotation cost. We show that one of the reasons for this problem is that active learning can only select examples which are already covered by the model. In this thesis, we better tailor active learning to the need of natural language processing as follows. We formulate the Unreliable Parameter Principle: Active learning should explicitly and additionally address unreliably trained model parameters in order to optimally reduce classification error. In order to do so, we should target both missing events and infrequent events. We demonstrate the effectiveness of such an approach for a range of natural language processing tasks: prepositional phrase attachment, sequence labelling, and syntactic parsing. For prepositional phrase attachment, the explicit selection of unknown prepositions significantly improves coverage and classification performance for all examined active learning methods. For sequence labelling, we introduce a novel active learning method which explicitly targets unreliable parameters by selecting sentences with many unknown words and a large number of unobserved transition probabilities. For parsing, targeting unparseable sentences significantly improves coverage and f-measure in active learning.

Since their recent introduction, conditional random fields (CRFs) have been successfully applied to a multitude of structured labelling tasks in many different domains. Examples include natural language processing (NLP), bioinformatics and computer vision. Within NLP itself we have seen many different application areas, like named entity recognition, shallow parsing, information extraction from research papers and language modelling. Most of this work has demonstrated the need, directly or indirectly, to employ some form of regularisation when applying CRFs in order to overcome the tendency for these models to overfit. To date a popular method for regularising CRFs has been to fit a Gaussian prior distribution over the model parameters. In this thesis we explore other methods of CRF regularisation, investigating their properties and comparing their effectiveness. We apply our ideas to sequence labelling problems in NLP, specifically part-of-speech tagging and named entity recognition. We start with an analysis of conventional approaches to CRF regularisation, and investigate possible extensions to such approaches. In particular, we consider choices of prior distribution other than the Gaussian, including the Laplacian and Hyperbolic; we look at the effect of regularising different features separately, to differing degrees, and explore how we may define an appropriate level of regularisation for each feature; we investigate the effect of allowing the mean of a prior distribution to take on non-zero values; and we look at the impact of relaxing the feature expectation constraints satisfied by a standard CRF, leading to a modified CRF model we call the inequality CRF. Our analysis leads to the general conclusion that although there is some capacity for improvement of conventional regularisation through modification and extension, this is quite limited. Conventional regularisation with a prior is in general hampered by the need to fit a hyperparameter or set of hyperparameters, which can be an expensive process. We then approach the CRF overfitting problem from a different perspective. Specifically, we introduce a form of CRF ensemble called a logarithmic opinion pool (LOP), where CRF distributions are combined under a weighted product. We show how a LOP has theoretical properties which provide a framework for designing new overfitting reduction schemes in terms of diverse models, and demonstrate how such diverse models may be constructed in a number of different ways. Specifically, we show that by constructing CRF models from manually crafted partitions of a feature set and combining them with equal weight under a LOP, we may obtain an ensemble that significantly outperforms a standard CRF trained on the entire feature set, and is competitive in performance to a standard CRF regularised with a Gaussian prior. The great advantage of LOP approach is that, unlike the Gaussian prior method, it does not require us to search a hyperparameter space. Having demonstrated the success of LOPs in the simple case, we then move on to consider more complex uses of the framework. In particular, we investigate whether it is possible to further improve the LOP ensemble by allowing parameters in different models to interact during training in such a way that diversity between the models is encouraged. Lastly, we show how the LOP approach may be used as a remedy for a problem that standard CRFs can sometimes suffer. In certain situations, negative effects may be introduced to a CRF by the inclusion of highly discriminative features. An example of this is provided by gazetteer features, which encode a word's presence in a gazetteer. We show how LOPs may be used to reduce these negative effects, and so provide some insight into how gazetteer features may be more effectively handled in CRFs, and log-linear models in general.

This thesis is concerned with the investigation of directionality of difficulty in second language acquisition (SLA) by Chinese-speaking learners learning English as a foreign language (EFL) and by English-speaking learners learning Chinese as a foreign language (CFL) . Chinese allows both subject PRO in finite clauses and object pro. However, subject PRO in finite clauses and object pro give rise to ungrammaticality in English. Unlike Chinese, in which topics can be base-generated, English does not allow a base-generated topic. Chinese and English are also different in that while English reflexives can only take a local subject in finite clauses as their antecedent (thus a short-distance reflexive), the Chinese reflexive ziji can take the matrix subject as well as the embedded subject as its antecedent (thus a long-distance reflexive) . With respect to these differences between the two languages, our focus is on whether it is more difficult for CFL learners to acquire subject PRO, object pro, base-generated topics and the long-distance reflexive in the acquisition of Chinese than for EFL learners to unlearn subject PRO, object pro, base-generated topics and the long-distance reflexive in the acquisition of English. The results of our study suggest that there is no single direction of difficulty in the SLA of Chinese and English. In terms of object pro, the direction of difficulty is from Chinese to English. However, in acquiring and unlearning the subject PRO, neither CFL learners nor EFL learners seem to have much difficulty. As for base-generated topics, it is found that the acquisition of this feature by CFL learners is more difficult than the unlearning of this feature by EFL learners. The results concerning the acquisition of the Chinese long-distance reflexive ziji by CFL learners suggest that a lack of long-distance binding for ziji is fossilized in these learners' interlanguage (IL) grammars of Chinese. Based on the findings in this study, we argue that the directionality of difficulty in SLA can only be studied with respect to individual language features and that the mere existence of relevant positive evidence in the input is not a guarantee that there will be a change in the learner's IL grammar. There are many factors involved in deciding the direction of difficulty in SLA. These factors include the availability of informative evidence to the learner, the possibility that the learner makes use of the evidence available for the restructuring of his IL grammar of the target language, the learners' ability to process the relevant data in the input, and the interaction between the structure in the learners' L1 and the inherent developmental stage of the target language.