ReQA: An Evaluation for End-to-End Answer Retrieval Models

Amin Ahmad; Daniel Cer; Noah Constant; Yinfei Yang

RECORD

ReQA: An Evaluation for End-to-End Answer Retrieval Models

Amin Ahmad, Noah Constant, Yinfei Yang, Daniel Cer

Computation and Language · Submitted 10 Jul 2019

abstract compiled · 2026-09-23 · no language-model review

The authors' main claim In this paper, we introduce Retrieval Question-Answering (ReQA), a benchmark for evaluating large-scale sentence-level answer retrieval models.

how it is stated · asserted 1.00 read the record →

Comment (0) Read on arXiv ↗

Cite ▾

arXiv ID1907.04780 (version 2)

Pith IDBBKICF7X, Pith's permanent record of this paper and its review

CitationAmin Ahmad, Daniel Cer, Noah Constant, Yinfei Yang. "ReQA: An Evaluation for End-to-End Answer Retrieval Models." arXiv:1907.04780 (2019). https://pith.science/paper/BBKICF7X

Claim paper

The compiled record

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Open the compiled record

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-09-23: 1 request, 4,803 input tokens.

What the abstract itself states, as probabilities: claims first 0.07 · states limitation 0.12 · human participants 0.05 · controlled comparison 0.72 · randomized assignment 0.01 · states quantitative result 0.04 · compares with prior results 0.52 · mentions code or data availability 0.97.

The abstract, sentence by sentence

arXiv listing abstract v2. Results and the main claim are underlined; show every role.

Popular QA benchmarks like SQuAD have driven progress on the task of identifying answer spans within a specific passage, with models now surpassing human performance. However, retrieving relevant answers from a huge corpus of documents is still a challenging problem, and places different requirements on the model architecture. There is growing interest in developing scalable answer retrieval models trained end-to-end, bypassing the typical document retrieval step. In this paper, we introduce Retrieval Question-Answering (ReQA), a benchmark for evaluating large-scale sentence-level answer retrieval models. We establish baselines using both neural encoding models as well as classical information retrieval techniques. We release our evaluation code to encourage further work on this challenging task.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#	sentence	role	the authors' own work	states the main conclusion	strength
s0	Popular QA benchmarks like SQuAD have driven progress on the task of identifying answer spans within a specific passage, with models now surpassing human performance.	background 1.00	0.04	0.03	asserted 1.00
s1	However, retrieving relevant answers from a huge corpus of documents is still a challenging problem, and places different requirements on the model architecture.	gap 0.65	0.06	0.03	asserted 1.00
s2	There is growing interest in developing scalable answer retrieval models trained end-to-end, bypassing the typical document retrieval step.	background 0.97	0.04	0.04	asserted 0.98
s3	In this paper, we introduce Retrieval Question-Answering (ReQA), a benchmark for evaluating large-scale sentence-level answer retrieval models.	contribution 1.00	0.92	0.19	asserted 1.00
s4	We establish baselines using both neural encoding models as well as classical information retrieval techniques.	method 0.88	0.91	0.06	asserted 0.99
s5	We release our evaluation code to encourage further work on this challenging task.	contribution 0.62	0.90	0.05	asserted 0.97

What kind of paper the classifier reports this is: dataset or benchmark 1.00 · evidence from observation or experiment 0.86.

What was examined, and what was not

sources read

arXiv listing abstract v2

not examined

full text
figures and tables
references and citing sentences
relations to other papers

levels run

P abstract

not run

S sections
T body sentences
C citations
R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

The paper

Original abstract

Popular QA benchmarks like SQuAD have driven progress on the task of identifying answer spans within a specific passage, with models now surpassing human performance. However, retrieving relevant answers from a huge corpus of documents is still a challenging problem, and places different requirements on the model architecture. There is growing interest in developing scalable answer retrieval models trained end-to-end, bypassing the typical document retrieval step. In this paper, we introduce Retrieval Question-Answering (ReQA), a benchmark for evaluating large-scale sentence-level answer retrieval models. We establish baselines using both neural encoding models as well as classical information retrieval techniques. We release our evaluation code to encourage further work on this challenging task.

Human review

How Pith checks papers → · Full audit of this paper →

References

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

SPARTA: Efficient Open-Domain Question Answering via Sparse Transformer Matching Retrieval
cs.CL 2020-09 conditional novelty 6.0 of 10

SPARTA learns a sparse, token-level matching representation for answer passages that can be indexed like a search engine, achieving strong retrieval QA performance.
Using the Hammer Only on Nails: A Hybrid Method for Evidence Retrieval for Question Answering
cs.IR 2020-09 conditional novelty 6.0 of 10

A router that picks BM25 or a neural retriever from the top BM25 scores improves QA evidence retrieval on two of three benchmarks and runs up to five times faster.
Embedding-based Zero-shot Retrieval through Query Generation
cs.IR 2020-09 conditional novelty 5.0 of 10

Synthetic queries generated by BART finetuned on MS-MARCO, used to pretrain two-tower and Siamese BERT retrievers, yield strong zero-shot retrieval performance, beating BM25 on five of six datasets.

How the literature uses it

11 arXiv papers cite this in 23 sentences

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

Papers citing it · 11 sentences without a confident label

Uses its method or tool 1 paper

USE-QA is pre-trained on large scale retrieval tasks and, as used in [ref] , we do not fine-tune it in the retrieval tasks.

Using the Hammer Only on Nails: A Hybrid Method for Evidence Retrieval for Question Answering · Individual IR Models · 0.93

Uses its data or values 1 paper

We show that using this hybrid strategy generally improves evidence retrieval performance in three QA datasets: OpenBookQA [ref] , ReQA SQUAD, and ReQA Natural Questions (NQ) [ref] .

Using the Hammer Only on Nails: A Hybrid Method for Evidence Retrieval for Question Answering · Introduction · 0.83

Compares with it 1 paper

A summary of the results are shown in <ref>[Our BM25 baseline significantly outperforms those reported in [ref] .

Embedding-based Zero-shot Retrieval through Query Generation · Ablations · 0.85

Cites it as support 1 paper

Similar observations are also found in prior research [ref] .

SPARTA: Efficient Open-Domain Question Answering via Sparse Transformer Matching Retrieval · In-domain Performance · 0.98

Cites it as background 5 papers

[ref] proposed additional pretraining tasks and showed improved performance in the ReQA benchmark [ref] .

Embedding-based Zero-shot Retrieval through Query Generation · Related Work · 0.84
Recent research has shown promising results [ref] on developing neural models for passage retrieval tasks, including Retrieval Question Answering [ref] , Open Domain Question Answering [ref] , and MS MARCO [ref] .

Neural Passage Retrieval with Improved Negative Contrast · Introduction · 0.97
One family of approaches involves the dual encoder architecture where the query and document are encoded independently of each other, and efficient retrieval is achieved using approximate nearest-neighbour search [ref] or by employing other data structures, such as learning an inverted index based on latent representations [ref] .

Conformer-Kernel with Query Term Independence for Document Retrieval · Full retrieval with deep models · 0.88

What it cites · 14 sentences without a confident label

Uses the method or tool of 1 paper

As our primary neural baseline, we take the recently released universal sentence encoder QA (USE-QA) model from [ref] [<https://tfhub.dev/google/universal-sentence-encoder-multilingual-qa/1>].

Multilingual Universal Sentence Encoder for Semantic Retrieval · Neural Baselines · 0.92

Discussion

Know something about this paper? Leave a comment, or highlight any passage above to comment on it.