No description available.

No description available.

No description available.

No description available.

My aim is to develop a computerised system that gives a (sight-impaired or sighted) user the ability to skim a scientific paper. Specifically, I use lexical chains (Morris and Hirst, 1991) to represent topics in text. I adapt existing lexical chain algorithms to scientific text, as opposed to news text. My hypothesis is that terms in lexical chains for scientific text require the inclusion of adjectives. For example, the term <i>rhetorical structure</i> is more characteristic of a paper about discourse structure than the word <i>structure</i> alone. However, not all adjectives have this property of making terms more characteristic. For example, the phrase<i> different experiment </i>is no more descriptive than the word <i>experiment</i>. I will present an algorithm for automatically distinguishing between characteristic adjectives (e.g., <i>rhetorical) </i>and non-characteristic adjectives (e.g., <i>different).</i> In my target application (text skimming) the user is presented with lexical chains directly. A good lexical chain by my evaluation is one that is similar to a lexical chain created by a human. For this reason I collect a corpus of human-generated lexical chains. This gold standard contains 230 lexical chains created by 13 annotators. No ready-made metrics for computing lexical chain similarity exist. Therefore I will test and choose appropriate similarity measures for this task. To calibrate them, I formalise my intuition of similarity between lexical chains by creating a gold standard of similarity judgements. My data show that humans often include adjectives in lexical chains and that lexical chains created using characteristic adjectives are more similar to human-generated lexical chains than lexical chains that contain only nouns.

Commonly used programs to search sequence databases such as BLAST, FASTA and SSEARCH identify sequence homology through pairwise alignment techniques. These programs are good at detecting closely related sequences but have problems accurately detecting homologous sequences with low sequence identity. This thesis describes a new approach that attempts to improve the detection of distantly related sequences by rejecting the assumption that all sites in a protein behave in an identical manner. This is done without the use of profile techniques, which require the preliminary collection of a set of homologs. Existing programs use general properties of proteins to generate alignment scores, which simplify calculations but may also result in a decrease in accuracy. In reality, amino acid replacement probabilities and rates, amino acid frequencies and gap probabilities all vary according to where a residue lies in a protein structure. Typical patterns of these structure-specific variations in evolutionary dynamics can be incorporated into a database search program through the use of hidden Markov models (HMMs), and hence potentially improve the detection of more distantly related sequences. In this thesis, the utility of including structure-specific evolutionary information in a database search program has been assessed. I have developed a general methodology permitting structure-based evolutionary models to be used for database searching, and specific algorithms that incorporate either solvent accessibility distinctions or protein secondary structure distinctions for globular proteins. In addition I have developed a database search algorithm for transmembrane proteins. The improvement afforded by adding the extra information has then been evaluated through the use of both simulated sequences, which exactly fit the models, and real sequences from the SCOP database. The success rate of each of these programs has been compared to a simplified model that contains the general properties of proteins but with no structural distinctions.

No description available.

No description available.

No description available.

No description available.