Parallel Aligned Treebank Corpora at LDC: Methodology, Annotation and Integration

The interest in syntactically-annotated data for improving machine translation quality has spurred the growing demand for parallel aligned treebank data. To meet this demand, the Linguistic Data Consortium (LDC) has created large volume, multi-lingual and multi-level aligned treebank corpora by aligning and integrating existing treebank annotation resources. Such corpora are more useful when the alignment is further enriched with contextual and linguistic information. This paper details how we create these enriched parallel aligned corpora, addressing approaches, methodologies, theories, technologies, complications, and cross-lingual features.