In this paper, we investigate the tendency of end-to-end neural Machine
Reading Comprehension (MRC) models to match shallow patterns rather than
perform inference-oriented reasoning on RC benchmarks. We aim to test the
ability of these systems to answer questions which focus on referential
inference. We propose ParallelQA, a strategy to formulate such questions using
parallel passages. We also demonstrate that existing neural models fail to
generalize well to this setting.

探讨了端到端神经机器阅读理解模型倾向于匹配浅层模式而非推理，同时提出了利用平行语段来进行参照推理问题的 ParallelQA 策略，并证明了现有神经模型在此设置中无法很好地泛化。

面向推理的阅读理解：ParallelQA

Towards Inference-Oriented Reading Comprehension: ParallelQA

Arabic is a widely-spoken language with a rich and long history spanning more
than fourteen centuries. Yet existing Arabic corpora largely focus on the
modern period or lack sufficient diachronic information. We develop a
large-scale, historical corpus of Arabic of about 1 billion words from diverse
periods of time. We clean this corpus, process it with a morphological
analyzer, and enhance it by detecting parallel passages and automatically
dating undated texts. We demonstrate its utility with selected case-studies in
which we show its application to the digital humanities.

本文介绍一种收集和处理阿拉伯语历史语料库的方法，所得到的大规模语料库包括约 10 亿个单词，使用形态分析器进行处理，并检测出了平行段落并自动标注了年代。该语料库可用于数字人文学研究的实际应用。

Shamela：一个大规模历史阿拉伯语语料库

Shamela: A Large-Scale Historical Arabic Corpus

We propose a method for efficiently finding all parallel passages in a large
corpus, even if the passages are not quite identical due to rephrasing and
orthographic variation. The key ideas are the representation of each word in
the corpus by its two most infrequent letters, finding matched pairs of strings
of four or five words that differ by at most one word and then identifying
clusters of such matched pairs. Using this method, over 4600 parallel pairs of
passages were identified in the Babylonian Talmud, a Hebrew-Aramaic corpus of
over 1.8 million words, in just over 30 seconds. Empirical comparisons on
sample data indicate that the coverage obtained by our method is essentially
the same as that obtained using slow exhaustive methods.

提出一种高效、能够识别不完全相同的并行段落的方法，通过使用语料库中每个单词的两个最不常用的字母，找到由四到五个单词组成、相差不超过一个单词的匹配字符串对，然后识别这些匹配字符串对的聚类，该方法在约 30 秒内在数百万字的希伯来 - 阿拉姆语语料库中识别出 4600 多个并行段落，与缓慢、耗时的穷举法相比，该方法的覆盖范围基本相同。