Thesis (M. Phil.)--University of Hong Kong, 1989.

This thesis shows that existing definitions of code-injection attacks (e.g., SQL-injection attacks) are flawed. The flaws make it possible for attackers to circumvent existing mechanisms, by supplying code-injecting inputs that are not recognized as such. The flaws also make it possible for benign inputs to be treated as attacks. After describing these flaws in conventional definitions of code-injection attacks, this thesis proposes a new definition, which is based on whether the symbols input to an application get used as (normal-form) values in the application's output. Because values are already fully evaluated, they cannot be considered ``code'' when injected. This simple new definition of code-injection attacks avoids the problems of existing definitions, improves our understanding of how and when such attacks occur, and enables us to evaluate the effectiveness of mechanisms for mitigating such attacks.

Not available / text

Not available / text

Not available

The subject matter of this thesis is the problem of learning to discover grammatical structure from raw text alone, without access to explicit instruction or annotation -- in particular, by a computer or computational process -- in other words, unsupervised parser induction, or simply, unsupervised parsing. This work presents a method for raw text unsupervised parsing that is simple, but nevertheless achieves state-of-the-art results on treebank-based direct evaluation. The approach to unsupervised parsing presented in this dissertation adopts a different way to constrain learned models than has been deployed in previous work. Specifically, I focus on a sub-task of full unsupervised partial parsing called unsupervised partial parsing. In essence, the strategy is to learn to segment a string of tokens into a set of non-overlapping constituents or chunks which may be one or more tokens in length. This strategy has a number of advantages: it is fast and scalable, based on well-understood and extensible natural language processing techniques, and it produces predictions about human language structure which are useful for human language technologies. The models developed for unsupervised partial parsing recover base noun phrases and local constituent structure with high accuracy compared to strong baselines. Finally, these models may be applied in a cascaded fashion for the prediction of full constituent trees: first segmenting a string of tokens into local phrases, then re-segmenting to predict higher-level constituent structure. This simple strategy leads to an unsupervised parsing model which produces state-of-the-art results for constituent parsing of English, German and Chinese. This thesis presents, evaluates and explores these models and strategies. / text

published_or_final_version / Computer Science / Master / Master of Philosophy

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.

This thesis tests the early incorporation of prosodic information during on-line processing of ambiguous word pairs such as Packing cases. The word pair is syntactically ambiguous between a noun or verb phrase interpretation. However, the two interpretations are prosodically distinct. An on-line, cross-modal, response-time task found that subjects disambiguated the word pairs using prosodic information. Experiment 2 swapped the timing,f_o and amplitude of the noun phrase versions with the verb phrase versions. If prosodic information were guiding parsing, swapping the prosody of the alternatives should change subjects' parses of the word-pairs. Subjects interpreted the cross-synthesised noun phrases as verb phrases and the crosssynthesised verb phrases as noun phrases. This provides additional evidence in favour of early prosodic processing. Experiment 3 tested whether subjects' ability to differentiate the two forms would be affected by flattening the f_o of the word pairs. Subjects' ability to disambiguate the word pairs was reduced by flattening the f_o of the stimuli. Again, this provides evidence in favour of f_o guiding parsing. Experiment 4 investigated the perceptual salience of prosodic information in the absence of lexical information, by testing parsing of delexicalised versions of the same wordpairs. Subjects continued to disambiguate the stimuli. This indicates that prosody can guide parsing even without lexical information. The results of the four experiments provide strong evidence in favour of the early incorporation of prosodic information in parsing: prosodic information can influence on-line parsing even in the presence of contradictory syntactic and spectral preferences; and in the absence of lexical information. This thesis concludes that the results of the experiments support strong interaction models of processing.

In this thesis we shall investigate the kinds of information which the Human Sentence Processing Mechanism employs during the resolution of local syntactic ambiguity in written texts. The thesis is in three parts. In Part I we consider some current models of syntactic ambiguity resolution. On the one hand, we consider the structural approaches, in which the processor considers only syntactic information when choosing between alternatives. On the other, we consider the interactive approaches, in which different kinds of information are brought to bear during the resolution process. In Part II, we describe a number of experiments which contrast the predictions of these two approaches. In particular, we investigate the processing of sentences which are locally ambiguous between a simple noun phrase analysis and a complex noun phrase analysis. Frazier (1979) predicts that the simple noun phrase analysis is chosen because it utilizes fewest phrasal nodes in its associated phrase marker. Crain and Steedman (1985), however, predict that the interpretation of the noun phrase is determined by referential factors, such as whether or not a unique referent can be identified for the noun phrase. The results support Crain and Steedman's interactive hypothesis. In Part III, we consider some theoretical issues concerning the timing of the processor's decisions. Crain and Steedman's original model is modified in the light of such considerations. We explore the implications of the modification for the status of syntax and semantics within our model of sentence comprehension. In the final chapter, we attempt to explain the existence of parsing preferences in sentences which are presented in isolation, and for which no explicit contextual information has been provided. We conclude that contextual considerations, such as the distinction between what is and what is not already known to the hearer, are of fundamental importance during the resolution of local syntactic ambiguity by the Human Sentence Processing Mechanism.