Description
- Abstract:
- This dissertation describes two novel algorithms for fully unsupervised part-of-speech tagging (POST), both of which exceed the performance of the current state-of-the-art models in both accuracy and computational cost. The first iterative algorithm produces a collection of geometric descriptor vectors that characterize each word type along with a clustering of these vectors which provide the inferred POS labels. The second algorithm also produces a clustering of descriptor vectors, but it relies on simple biologically-plausible mechanisms which cause these vectors to self-organize in response to the presentation of natural language data. These biological mechanisms are extended for use in a related algorithm for grammar induction which is shown to perform well on simple context free grammars. Finally, a careful examination of the three most common evaluation criteria for unsupervised POST is included which demonstrates that only one of these three criteria performs satisfactorily.
- Notes:
- Thesis (Ph.D. -- Brown University (2010)
Access Conditions
- Rights
- In Copyright
- Restrictions on Use
- Collection is open for research.
Citation
Lamar, Michael Taylor,
"Unsupervised Linguistic Inference"
(2010).
Applied Mathematics Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://doi.org/10.7301/Z07M0655
Relations
Collection:
-
Applied Mathematics Theses and Dissertations
Theses and Dissertations for the Applied Mathematics department....