Deducing linguistic structure from the statistics of large corpora

Two experiments that strongly suggest that largely distributional techniques might be developed to automatically provide both a set of part of speech tags for English and a skeletal parsing of free English text are described. In one experiment the authors have developed a constituent boundary parsing algorithm that derives an (unlabeled) bracketing, given text annotated for part of speech as input. In other experiment the authors have investigated whether a distributional analysis can discover a part of speech tag set which might prove adequate to support experiments. The state of a tagged natural language corpus to aid such experiments is summarized.<
>