An Evaluation of Sindhi Word Embedding in Semantic Analogies and Downstream Tasks

TL;DR

This paper introduces a new Sindhi word embedding corpus and evaluates various embedding algorithms, demonstrating that CBOW and skip-gram outperform GloVe and fastText in semantic analogy and downstream tasks.

Contribution

It presents a large Sindhi corpus and a comprehensive evaluation of embedding algorithms, highlighting the effectiveness of CBOW and skip-gram for Sindhi language tasks.

Findings

CBOW and skip-gram outperform GloVe and fastText in evaluations

New Sindhi corpus with over 61 million words

Effective preprocessing pipeline developed for Sindhi text

Abstract

In this paper, we propose a new word embedding based corpus consisting of more than 61 million words crawled from multiple web resources. We design a preprocessing pipeline for the filtration of unwanted text from crawled data. Afterwards, the cleaned vocabulary is fed to state-of-the-art continuous-bag-of-words, skip-gram, and GloVe word embedding algorithms. For the evaluation of pretrained embeddings, we use popular intrinsic and extrinsic evaluation approaches. The evaluation results reveal that continuous-bag-of-words and skip-gram perform better than GloVe and existing Sindhi fastText word embedding on both intrinsic and extrinsic evaluation approaches