Text Chunking using Transformation-Based Learning
About Chunking
Conclusions
- Focuses on identifying low-level noun groups by using hand-built grammars and finite state techniques
- Hidden Markov Models (HMM) statistical tool for modeling a wide range of time series data.
- Introduce as a tagging problem,able to apply transformation-based learning
- Some transformation-based learning algorithms are useful in other settings
Future Directions
Introduction
Transformational Text Chunking
Transformation-Based Learning Mechanism
- Different paths to increase systems power
- Diminish linguistic contexts
- Adding templates to chunk structure
- Involves dividing sentences into non overlapping segments on the basis of fairly superficial analysis.
- Provides a foundation for further levels of analysis such as: verb-argument identification.
- Brill 1993
- Sequence of transformational rules i learned, which improves a baseline model for interpretative text
- Part of speech tagging
- Prepositional phrase attachment
- Assigning unlabeled binary-branching tree sentences
- Can be encoded as a tagging problem
- Different Encoding Choices
- A baseline system to go off of
- Rule template to follow
Algorithm Design Issues
Results
- Organization of the computation
- Indexing Static Rule Elements
- Heuristic Disabling of unlikely rules
- Derivation of training and testing data
- Analysis of Initial Rules
- Contribution of Lexical Templates
- Frequent Error Classes