Web-scale Surface and Syntactic n-gram Features for Dependency Parsing
This work addresses parsing accuracy improvements for natural language processing applications, but it is incremental as it builds on existing feature-based approaches.
The paper tackled dependency parsing by developing novel first- and second-order features based on syntactic n-grams from the Google Syntactic Ngrams corpus and extending surface n-gram features from Web1T to the Google Books corpus, resulting in up to 0.8% absolute UAS improvements on newswire and 1.4% on web text.
We develop novel first- and second-order features for dependency parsing based on the Google Syntactic Ngrams corpus, a collection of subtree counts of parsed sentences from scanned books. We also extend previous work on surface $n$-gram features from Web1T to the Google Books corpus and from first-order to second-order, comparing and analysing performance over newswire and web treebanks. Surface and syntactic $n$-grams both produce substantial and complementary gains in parsing accuracy across domains. Our best system combines the two feature sets, achieving up to 0.8% absolute UAS improvements on newswire and 1.4% on web text.