Phrase Retrieval for Open-Domain Conversational QA with Conversational Dependency Modeling via Contrastive Learning

Paper: arXiv:2306.04293 Authors: Soyeong Jeong, Jinheon Baek, Sung Ju Hwang, Jong C. Park Published: 2023-06-07 Code: https://github.com/starsuzi/PRO-ConvQA

Summary

Open-Domain Conversational Question Answering (ODConvQA) traditionally relies on a retriever-reader pipeline: passages are retrieved first, then a reader predicts answers from them. This design suffers from three problems:

• Error propagation: reader mistakes stem from retriever failures.
• Development overhead: both retriever and reader must be built separately.
• Latency: the two stages cannot run in parallel.

Proposed Approach

The authors propose directly predicting answers via phrase retrieval over a sequence of words, collapsing the two subtasks into one. This is the first study of phrase retrieval's capability for ODConvQA.

A naive adoption is problematic because a conversation has dependencies between previous and current turns. To address this, they introduce a contrastive learning strategy:

• Maximize representational similarity of consecutive turns in a conversation;
• Minimize similarity with irrelevant conversational contexts.

This ensures that phrase retrieval for the current turn reflects prior turns of the dialogue.

Results

Experiments on two ODConvQA datasets show the model substantially outperforms relevant retriever-reader baselines.

Relevance

This work sits at the intersection of conversational search and retrieval-augmented systems, illustrating a broader trend: merging cascaded retrieval/ranking/generation stages into unified, end-to-end architectures. It complements surveys and follow-up work on conversational search systems, agentic retrieval, and LLM-based retrievers (e.g., ChatRetriever, CoSearchAgent).

Original Abstract

> Open-Domain Conversational Question Answering (ODConvQA) aims at answering questions through a multi-turn conversation based on a retriever-reader pipeline, which retrieves passages and then predicts answers with them. However, such a pipeline approach not only makes the reader vulnerable to the errors propagated from the retriever, but also demands additional effort to develop both the retriever and the reader, which further makes it slower since they are not runnable in parallel. In this work, we propose a method to directly predict answers with a phrase retrieval scheme for a sequence of words, reducing the conventional two distinct subtasks into a single one. Also, for the first time, we study its capability for ODConvQA tasks. However, simply adopting it is largely problematic, due to the dependencies between previous and current turns in a conversation. To address this problem, we further introduce a novel contrastive learning strategy, making sure to reflect previous turns when retrieving the phrase for the current context, by maximizing representational similarities of consecutive turns in a conversation while minimizing irrelevant conversational contexts. We validate our model on two ODConvQA datasets, whose experimental results show that it substantially outperforms the relevant baselines with the retriever-reader. Code is available at: https://github.com/starsuzi/PRO-ConvQA.