Full text
| |
| Author(s): |
Arnaldo Candido Junior
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | São Carlos. |
| Institution: | Universidade de São Paulo (USP). Instituto de Ciências Matemáticas e de Computação (ICMC/SB) |
| Defense date: | 2013-03-28 |
| Examining board members: |
Sandra Maria Aluisio;
André Carlos Ponce de Leon Ferreira de Carvalho;
Valéria Delisandra Feltrim;
Marcelo Finger;
Estevam Rafael Hruschka Júnior
|
| Advisor: | Sandra Maria Aluisio |
| Abstract | |
Natural Language Processing is an interdisciplinary research area that encompasses two large research avenues: language analysis and language synthesis. This thesis contributes for both of them. In what concerns language analysis, it presents an integrative model that links different levels of linguistic analysis. The evaluation of such model takes into consideration several levels: morphologic (including lexical and morph-syntactic sub-levels), syntactic and semantic. Whereas traditional analysis are undertaken from the lower levels to the upper ones, propagating errors in such direction, the model proposed herein is able to unify different levels of analysis using a bidirectional approach. The model is based on a large-scale neural network trained in corpus, which extracts its training features from tokens within the sentences. A type of recurrence denominated co-activation is applied to the model to make the analysis of a pattern able to modify (and to be modified by) the analysis of other patterns in a same context. This model may be used for purposes different from those for which it was conceived and yields satisfactory results in lemmatization and part-of-speech analysis, but still needs work on syntactic analysis. The tool associated to this model makes it possible to analyze the proposed recurrence language and the bidirectional influence of different levels on each other, including sub-level interaction. Experiments on both co-activation and bidirectional level integration were performed, and the results were considered satisfactory. On the other hand, in what concerns language synthesis, this thesis presents a rule-based model of syntactic simplification (one of text adaptation techniques), applicable to syntactically parsed texts in order to render them simpler for low literacy readers. The tool associated to this model makes it possible to carry out the task of syntactic simplification in Portuguese language. Such tool achieved 77.2% of f-measure in a task that simplified approximately 16% of the sentences of an encyclopedic text (AU) | |
| FAPESP's process: | 08/08963-4 - Bidirectional Language Analysis for Syntactic Simplification of Portuguese Texts Aiming at Digital Accessibility Improvement |
| Grantee: | Arnaldo Candido Junior |
| Support Opportunities: | Scholarships in Brazil - Doctorate |