We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CL

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computation and Language

Title: Improving Low-resource Reading Comprehension via Cross-lingual Transposition Rethinking

Abstract: Extractive Reading Comprehension (ERC) has made tremendous advances enabled by the availability of large-scale high-quality ERC training data. Despite of such rapid progress and widespread application, the datasets in languages other than high-resource languages such as English remain scarce. To address this issue, we propose a Cross-Lingual Transposition ReThinking (XLTT) model by modelling existing high-quality extractive reading comprehension datasets in a multilingual environment. To be specific, we present multilingual adaptive attention (MAA) to combine intra-attention and inter-attention to learn more general generalizable semantic and lexical knowledge from each pair of language families. Furthermore, to make full use of existing datasets, we adopt a new training framework to train our model by calculating task-level similarities between each existing dataset and target dataset. The experimental results show that our XLTT model surpasses six baselines on two multilingual ERC benchmarks, especially more effective for low-resource languages with 3.9 and 4.1 average improvement in F1 and EM, respectively.
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Cite as: arXiv:2107.05002 [cs.CL]
  (or arXiv:2107.05002v2 [cs.CL] for this version)

Submission history

From: Gaochen Wu [view email]
[v1] Sun, 11 Jul 2021 09:35:16 GMT (650kb)
[v2] Tue, 28 Sep 2021 09:14:48 GMT (650kb)

Link back to: arXiv, form interface, contact.