No description available.

Research into statistical parsing for English has enjoyed over a decade of successful results. However, adapting these models to other languages has met with difficulties. Previous comparative work has shown that Modern Arabic is one of the most difficult languages to parse due to rich morphology and free word order. Classical Arabic is the ancient form of Arabic, and is understudied in computational linguistics, relative to its worldwide reach as the language of the Quran. The thesis is based on seven publications that make significant contributions to knowledge relating to annotating and parsing Classical Arabic. Classical Arabic has been studied in depth by grammarians for over a thousand years using a traditional grammar known as i'rab. Using this grammar to develop a representation for parsing is challenging, as it describes syntax using a hybrid of phrase-structure and dependency relations. This work aims to advance the state-of-the-art for hybrid parsing by introducing a formal representation for annotation and a resource for machine learning. The main contributions are the first treebank for Classical Arabic and the first statistical dependency-based parser in any language for ellipsis, dropped pronouns and hybrid representations. A central argument of this thesis is that using a hybrid representation closely aligned to traditional grammar leads to improved parsing for Arabic. To test this hypothesis, two approaches are compared. As a reference, a pure dependency parser is adapted using graph transformations, resulting in an 87.47% F I-score. This is compared to an integrated parsing model with an F I-score of 89.03%, demonstrating that joint dependency-constituency parsing is better suited to Classical Arabic. The Quran was chosen for annotation as a large body of work exists providing detailed syntactic analysis. Volunteer crowdsourcing is used for annotation in combination with expert supervision. A practical result of the annotation effort is the corpus website: http://corpus.quran.com, an educational resource with over two million users per year.

This thesis introduces a general method for using information at the utterance level and across utterances for automatic speech recognition. The method involves classification of utterances into types. Using constraints at the utterance level via this classification method allows information sources to be exploited which cannot necessarily be used directly for word recognition. The classification power of three sources of information is investigated: the language model in the speech recogniser, dialogue context and intonation. The method is applied to a challenging task: the recognition of spontaneous dialogue speech. The results show success in automatic utterance type classification, and subsequent word error rate reduction over a baseline system, when all three information sources are probabilistically combined.

Parsing is the task of assigning syntactic or semantic structure to a natural language sentence. This thesis focuses on syntactic parsing with Combinatory Categorial Grammar (CCG; Steedman 2000). CCG allows incremental processing, which is essential for speech recognition and some machine translation models, and it can build semantic structure in tandem with syntactic parsing. Supertagging solves a subset of the parsing task by assigning lexical types to words in a sentence using a sequence model. It has emerged as a way to improve the efficiency of full CCG parsing (Clark and Curran, 2007) by reducing the parser’s search space. This has been very successful and it is the central theme of this thesis. We begin by an analysis of how efficiency is being traded for accuracy in supertagging. Pruning the search space by supertagging is inherently approximate and to contrast this we include A* in our analysis, a classic exact search technique. Interestingly, we find that combining the two methods improves efficiency but we also demonstrate that excessive pruning by a supertagger significantly lowers the upper bound on accuracy of a CCG parser. Inspired by this analysis, we design a single integrated model with both supertagging and parsing features, rather than separating them into distinct models chained together in a pipeline. To overcome the resulting complexity, we experiment with both loopy belief propagation and dual decomposition approaches to inference, the first empirical comparison of these algorithms that we are aware of on a structured natural language processing problem. Finally, we address training the integrated model. We adopt the idea of optimising directly for a task-specific metric such as is common in other areas like statistical machine translation. We demonstrate how a novel dynamic programming algorithm enables us to optimise for F-measure, our task-specific evaluation metric, and experiment with approximations, which prove to be excellent substitutions. Each of the presented methods improves over the state-of-the-art in CCG parsing. Moreover, the improvements are additive, achieving a labelled/unlabelled dependency F-measure on CCGbank of 89.3%/94.0% with gold part-of-speech tags, and 87.2%/92.8% with automatic part-of-speech tags, the best reported results for this task to date. Our techniques are general and we expect them to apply to other parsing problems, including lexicalised tree adjoining grammar and context-free grammar parsing.