Phrase Retrieval Learns Passage Retrieval, Too

Abstract

Dense retrieval methods have shown great promise over sparse retrieval methods in a range of NLP problems. Among them, dense phrase retrieval-the most fine-grained retrieval unit-is appealing because phrases can be directly used as the output for question answering and slot filling tasks. 1 In this work, we follow the intuition that retrieving phrases naturally entails retrieving larger text blocks and study whether phrase retrieval can serve as the basis for coarse-level retrieval including passages and documents. We first observe that a dense phrase-retrieval system, without any retraining, already achieves better passage retrieval accuracy (+3-5% in top-5 accuracy) compared to passage retrievers, which also helps achieve superior end-to-end QA performance with fewer passages. Then, we provide an interpretation for why phrase-level supervision helps learn better fine-grained entailment compared to passage-level supervision, and also show that phrase retrieval can be improved to achieve competitive performance in document-retrieval tasks such as entity linking and knowledge-grounded dialogue. Finally, we demonstrate how phrase filtering and vector quantization can reduce the size of our index by 4-10x, making dense phrase retrieval a practical and versatile solution in multi-granularity retrieval. 2 * This work was done when JL worked as a visiting research scholar at Princeton University.1 Following previous work (Seo et al., 2018(Seo et al., , 2019, the term phrase denotes any contiguous text segment up to L words, which is not necessarily a linguistic phrase (see Section 2).2 Our code and models are available at https:// github.com/princeton-nlp/DensePhrases.

Augmenting Document Representations for Dense Retrieval with Interpolation and Perturbation

“…We observe that our retrieval result with small retrieved documents (i.e., K = 10) significantly improves the performance of the reader. This implies that a more accurate retrieval on smaller K in Table 1 helps achieve the improved QA performance as Lee et al (2021a) described. Furthermore, our reader performance may be further enhanced with advanced reading schemes (Mao et al, 2021a;Mao et al, 2021b).…”

Learning to Retrieve Passages without Supervision

“…Phrase-based retrieval (Seo et al, 2018(Seo et al, , 2019 eliminates the need for a reader during inference, as it directly retrieves the answer span given a query. Lee et al (2021a) demonstrated strong end-to-end ODQA results with this approach, and Lee et al (2021b) showed that it is also effective for passage retrieval. Our pretraining scheme can be seamlessly used for those architectures as well.…”

Visual Explanation for Open-Domain Question Answering With BERT

“…As a baseline for recent work, dense passage retriever (DPR) [32] uses a sophisticated sample mining and training approach to enable the potential of the dual-encoder retriever architecture, that is, using two independent BERTs to encode the question and the context separately and calculate the similarity between the two to select relevant passages. On the basis of this architecture, subsequent work includeed improvements in the calculation of similarity [33], [62], developing highly efficient training methods [60], [63], [66], and hierarchizing the retrieval process [40], [49]. However, such models act like a "black box," and it is difficult to understand what knowledge has been accurately stored and how the models internally filter and process passages and output the final results.…”