Can Syntactic Knowledge Be Ported from One Domain to Another?

Title: Can Syntactic Knowledge Be Ported from One Domain to Another?

Research Question: Can syntactic knowledge, which is essential for understanding and generating human language, be effectively transferred from one domain to another?

Methodology: The study investigates the impact of porting syntactic knowledge from the Wall Street Journal (WSJ) domain, where a treebank is available, to the Air Travel Information System (ATIS) domain, which is different in style and structure. Two methods were used to compare the results: using an automatic parser trained on the treebanked-domain and applying the SLM as an automatic parser.

Results: The SLM initialized on the WSJ domain outperformed the other methods, achieving a significant 0.4% absolute and 7% relative reduction in word error rate (WER), which is a measure of the accuracy of the generated text. This improvement was more than twice the minimum WER achievable on the N-best lists worked with.

Implications: The results suggest that syntactic knowledge can be effectively ported from one domain to another, improving the performance of language models in the new domain. This finding has significant implications for natural language processing and could lead to more accurate and efficient language generation and understanding systems.

Link to Article: https://arxiv.org/abs/0108022v1 Authors: arXiv ID: 0108022v1