JAVELIN IV: Cross-Lingual Question Answering System

Two papers from CMU’s Language Technologies Institute describing the JAVELIN IV QA system, presented at NTCIR-7 Workshop (Tokyo, December 2008).

Papers Merged

1. Complex Cross-lingual Question Answering as a Sequential Classification and Multi-Document Summarization Task — Hideki Shima, Ni Lao, Eric Nyberg, Teruko Mitamura
2. Query Expansion and Machine Translation for Robust Cross-Lingual Information Retrieval — Ni Lao, Hideki Shima, Teruko Mitamura, Eric Nyberg

Key Contributions

System Architecture

JAVELIN IV treats complex QA as sequential classification + multi-document summarization. Language-independent design with distributed computing — all algorithms use uniform module interfaces, loading language-specific resources at runtime.

IR Subsystem (Paper 2)

• Query term extraction, translation, filtering, expansion
• Novel alias expansion using lexico-syntactic patterns learned with weakly-supervised algorithm
• Achieved 59% MAP in both Chinese-to-Chinese and Japanese-to-Japanese NTCIR-7 IR4QA subtasks

QA Pipeline (Paper 1)

• Linguistic annotation, answer extraction, summarization algorithms
• Comparison of extraction units, syntactic features for classification, summarization strategies
• Analysis of MT and IR performance impact on answer quality

Connection to Vault

Early QA system work by Ni Lao before his Path Ranking Algorithm research. Demonstrates cross-lingual retrieval challenges also relevant to Efficient Reading Comprehension methods.