Related Experiment Videos
Offline grammar-based recognition of handwritten sentences.
Matthias Zimmermann1, Jean-Cédric Chappelier, Horst Bunke
1International Computer Science Institute, Berkeley, CA 94704-1198, USA. zimmerma@icsi.berkeley.edu
Summary
This study integrates Hidden Markov Models (HMM) with chart parsing for handwritten English sentence recognition. Syntax analysis significantly boosts recognition accuracy, enhancing performance in natural language processing tasks.
Area of Science:
- Natural Language Processing
- Computational Linguistics
- Machine Learning
Background:
- Offline handwritten English sentence recognition is a challenging task.
- Existing methods may not fully leverage linguistic structures for improved accuracy.
Purpose of the Study:
- To propose a novel sequential coupling of a Hidden Markov Model (HMM) recognizer and a probabilistic chart parser.
- To evaluate the impact of syntax analysis on handwritten sentence recognition rates.
Main Methods:
- Utilized a Hidden Markov Model (HMM) for initial recognition of handwritten English sentences.
- Employed a probabilistic bottom-up chart parser with Stochastic Context-Free Grammars (SCFG) derived from a text corpus.
- Sequentially coupled the HMM recognizer with the chart parser.
Main Results:
- Syntax analysis demonstrably improved recognition rates.
- The sequential coupling approach showed significant gains in accuracy.
Conclusions:
- Syntax analysis is a crucial component for enhancing handwritten English sentence recognition.
- The proposed HMM and chart parsing integration offers a promising direction for improving performance in this domain.