Effect of Language Pair Complexity on Zero-Shot Retrieval Performance Gaps with Code-Switched Training Data

Transferring information retrieval (IR) models from a high-resource language (typically English) to other languages in a zero-shot fashion has become a widely adopted approach. In this work, we show that the effectiveness of zero-shot rankers diminishes when queries and documents are present in different languages. Motivated by this, we propose to train ranking models on artificially code-switched data instead, which we generate by utilizing bilingual lexicons. To this end, we experiment with lexicons induced from (1) cross-lingual word embeddings and (2) parallel Wikipedia page titles. We use

Research goal: What is the effect of varying the language pair complexity (e.g., typologically similar vs. dissimilar languages) on the zero-shot retrieval performance gaps when using code-switched training data, quantified by MRR on standard multilingual IR benchmarks?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 8.5/10.