skip to main content


Title: Can NLI Models Verify QA Systems’ Predictions?
To build robust question answering systems, we need the ability to verify whether answers to questions are truly correct, not just “good enough” in the context of imperfect QA datasets. We explore the use of natural language inference (NLI) as a way to achieve this goal, as NLI inherently requires the premise (document context) to contain all necessary information to support the hypothesis (proposed answer to the question). We leverage large pre-trained models and recent prior datasets to construct powerful question conversion and decontextualization modules, which can reformulate QA instances as premise-hypothesis pairs with very high reliability. Then, by combining standard NLI datasets with NLI examples automatically derived from QA training data, we can train NLI models to evaluate QA models’ proposed answers. We show that our approach improves the confidence estimation of a QA model across different domains, evaluated in a selective QA setting. Careful manual analysis over the predictions of our NLI model shows that it can further identify cases where the QA model produces the right answer for the wrong reason, i.e., when the answer sentence cannot address all aspects of the question.  more » « less
Award ID(s):
1814522
NSF-PAR ID:
10380028
Author(s) / Creator(s):
; ;
Date Published:
Journal Name:
Findings of the Association for Computational Linguistics: EMNLP 2021
Page Range / eLocation ID:
3841 to 3854
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Current textual question answering (QA) models achieve strong performance on in-domain test sets, but often do so by fitting surface-level patterns, so they fail to generalize to out-of-distribution settings. To make a more robust and understandable QA system, we model question answering as an alignment problem. We decompose both the question and context into smaller units based on off-the-shelf semantic representations (here, semantic roles), and align the question to a subgraph of the context in order to find the answer. We formulate our model as a structured SVM, with alignment scores computed via BERT, and we can train end-to-end despite using beam search for approximate inference. Our use of explicit alignments allows us to explore a set of constraints with which we can prohibit certain types of bad model behavior arising in cross-domain settings. Furthermore, by investigating differences in scores across different potential answers, we can seek to understand what particular aspects of the input lead the model to choose the answer without relying on post-hoc explanation techniques. We train our model on SQuAD v1.1 and test it on several adversarial and out-of-domain datasets. The results show that our model is more robust than the standard BERT QA model, and constraints derived from alignment scores allow us to effectively trade off coverage and accuracy. 
    more » « less
  2. Question Answering (QA) naturally reduces to an entailment problem, namely, verifying whether some text entails the answer to a question. However, for multi-hop QA tasks, which require reasoning with \textit{multiple} sentences, it remains unclear how best to utilize entailment models pre-trained on large scale datasets such as SNLI, which are based on sentence pairs. We introduce Multee, a general architecture that can effectively use entailment models for multi-hop QA tasks. Multee uses (i) a local module that helps locate important sentences, thereby avoiding distracting information, and (ii) a global module that aggregates information by effectively incorporating importance weights. Importantly, we show that both modules can use entailment functions pre-trained on a large scale NLI datasets. We evaluate performance on MultiRC and OpenBookQA, two multihop QA datasets. When using an entailment function pre-trained on NLI datasets, Multee outperforms QA models trained only on the target QA datasets and the OpenAI transformer models. 
    more » « less
  3. Natural language inference (NLI) datasets (e.g., MultiNLI) were collected by soliciting hypotheses for a given premise from annotators. Such data collection led to annotation artifacts: systems can identify the premise-hypothesis relationship without observing the premise (e.g., negation in hypothesis being indicative of contradiction). We address this problem by recasting the CommitmentBank for NLI, which contains items involving reasoning over the extent to which a speaker is committed to complements of clause-embedding verbs under entailment-canceling environments (conditional, negation, modal and question). Instead of being constructed to stand in certain relationships with the premise, hypotheses in the recast CommitmentBank are the complements of the clause-embedding verb in each premise, leading to no annotation artifacts in the hypothesis. A state-of-the-art BERT-based model performs well on the CommitmentBank with 85% F1. However analysis of model behavior shows that the BERT models still do not capture the full complexity of pragmatic reasoning, nor encode some of the linguistic generalizations, highlighting room for improvement. 
    more » « less
  4. Estimating the quality of a result list, often referred to as query performance prediction (QPP), is a challenging and important task in information retrieval. It can be used as feedback to users, search engines, and system administrators. Although predicting the performance of retrieval models has been extensively studied for the ad-hoc retrieval task, the effectiveness of performance prediction methods for question answering (QA) systems is relatively unstudied. The short length of answers, the dominance of neural models in QA, and the re-ranking nature of most QA systems make performance prediction for QA a unique, important, and technically interesting task. In this paper, we introduce and motivate the task of performance prediction for non-factoid question answering and propose a neural performance predictor for this task. Our experiments on two recent datasets demonstrate that the proposed model outperforms competitive baselines in all settings. 
    more » « less
  5. null (Ed.)
    Recent work on Question Answering (QA) and Conversational QA (ConvQA) emphasizes the role of retrieval: a system first retrieves evidence from a large collection and then extracts answers. This open-retrieval setting typically assumes that each question is answerable by a single span of text within a particular passage (a span answer). The supervision signal is thus derived from whether or not the system can recover an exact match of this ground-truth answer span from the retrieved passages. This method is referred to as span-match weak supervision. However, information-seeking conversations are challenging for this span-match method since long answers, especially freeform answers, are not necessarily strict spans of any passage. Therefore, we introduce a learned weak supervision approach that can identify a paraphrased span of the known answer in a passage. Our experiments on QuAC and CoQA datasets show that although a span-match weak supervisor can handle conversations with span answers, it is not sufficient for freeform answers generated by people. We further demonstrate that our method is more flexible since it can handle both span answers and freeform answers. In particular, our method outperforms the span-match method on conversations with freeform answers, and it can be more powerful when combined with the span-match method. We also conduct in-depth analyses to show more insights on open-retrieval ConvQA under a weak supervision setting. 
    more » « less