Manning Inferring Sequential Structure Craig G . Nevill - Manning

Structure exists in sequences ranging from human language and music to the genetic information encoded in our DNA. This thesis shows how that structure can be discovered automatically and made explicit. Rather than examining the meaning of the individual symbols in the sequence, structure is detected in the way that certain combinations of symbols recur. In speech and text, these repetitions form words and phrases. They can be concisely represented by a hierarchical context-free grammar, where each repetition gives rise to a rule. As well as exact repetitions, sequences often exhibit branching and looping structure. These can be inferred and visualised as an automaton. We develop simple, robust techniques that reveal interesting structure in a wide range of real-world sequences including music, English text, descriptions of plants, graphical figures, DNA sequences, word classes in language, a genealogical database, programming languages, execution traces, and diverse sequences from a data compression corpus. These techniques are useful: they produce comprehensible visual explanations of structure, identify morphological units and significant phrases, compress data, optimise graphical rendering, infer recursive grammars by recognising self-similarity, and infer programs.