Unsupervised Monolingual Word Alignment

Analysis of unsupervised neural methods for monolingual word alignment. Based on SimAlign, we designed several monolingual word alignment methods.

Data:

• MultiMWA-Benchmark: a manually annotated benchmark, Multi-Genre Monolingual Word Alignment (MultiMWA) proposed by Lan et al. (2021).

Model:

• SimAlign: high quality unsupervised word alignment model without parallel training data using static and contextualized embeddings, which was proposed by Sabet et al. (2020).