Multi-Matching Network for Multiple Choice Reading Comprehension

Abstract

Multiple-choice machine reading comprehension is an important and challenging task where the machine is required to select the correct answer from a set of candidate answers given passage and question. Existing approaches either match extracted evidence with candidate answers shallowly or model passage, question and candidate answers with a single paradigm of matching. In this paper, we propose Multi-Matching Network (MMN) which models the semantic relationship among passage, question and candidate answers from multiple different paradigms of matching. In our MMN model, each paradigm is inspired by how human think and designed under a unified compose-match framework. To demonstrate the effectiveness of our model, we evaluate MMN on a large-scale multiple choice machine reading comprehension dataset (i.e. RACE). Empirical results show that our proposed model achieves a significant improvement compared to strong baselines and obtains state-of-the-art results.

Towards Medical Machine Reading Comprehension with Structural Knowledge and Plain Text

“…Multi-Matching (Tang et al, 2019) applies the Evidence-Answer Matching and Question-Passage-Answer Matching module to gather matching information and integrate them to get the scores of options.…”

“…We also pretrained word embeddings on a large-scale Chinese medical text. (Wang et al, 2018) 56.1 45.8 BiDAF (Seo et al, 2017) 52.7 43.6 SeaReader 58.2 48.4 Multi-Matching (Tang et al, 2019) 58.4 48.7 BERT-base 64.2 52.2 ERNIE (Sun et al, 2019) 64.7 53.4 RoBERTa-wwm-ext-large (Cui et al, 2019)…”

Learning to Distract: A Hierarchical Multi-Decoder Network for Automated Generation of Long Distractors for Multiple-Choice Questions for Reading Comprehension

“…It turns out that these interactions are helpful in finding potential candidate sentences of the article for DG. Such interactions can be achieved using SoftSel operation [27], which encodes the most relevant aspects of a sequence to another sequence. The input to SoftSel operation are two sequences, and output is an encoded sequence.…”

“…Layer. Inspired from the work on sequential variant of highway network [27], we adopted gated mechanism ( _ ℎ ) to control and encode information flow between contextual representation (i.e. from LSTM network) and evidence representation (i.e.…”

DCMN+: Dual Co-Matching Network for Multi-Choice Reading Comprehension

“…To well handle multi-choice MRC problem, an effective solution has to carefully model the relationship among the triplet of three sequences, passage (P), question (Q) and answer candidate options (A) with a matching module to determine the answer. However, previous unidirectional matching strategies usually calculate question-aware passage representation and ignore passage-aware question representation when modeling the relationship between passage and question (Wang et al 2018b;Tang, Cai, and Zhuo 2019;Chen et al 2019).…”