# Empirical Evaluation of Sequence-to-Sequence Models for Word Discovery in Low-resource Settings

## TL;DR

This paper empirically compares CNN, RNN, and Transformer sequence-to-sequence models for word discovery in low-resource language settings, finding RNNs outperform others in alignment quality and proposing an entropy-based confidence measure.

## Contribution

It provides the first comprehensive empirical evaluation of different sequence-to-sequence models for word discovery in low-resource scenarios, introducing an entropy-based confidence measure.

## Key findings

- RNNs outperform CNNs and Transformers in low-resource word discovery tasks
- Average Normalized Entropy effectively evaluates alignment quality
- Entropy confidence measure improves word discovery accuracy

## Abstract

Since Bahdanau et al. [1] first introduced attention for neural machine translation, most sequence-to-sequence models made use of attention mechanisms [2, 3, 4]. While they produce soft-alignment matrices that could be interpreted as alignment between target and source languages, we lack metrics to quantify their quality, being unclear which approach produces the best alignments. This paper presents an empirical evaluation of 3 main sequence-to-sequence models (CNN, RNN and Transformer-based) for word discovery from unsegmented phoneme sequences. This task consists in aligning word sequences in a source language with phoneme sequences in a target language, inferring from it word segmentation on the target side [5]. Evaluating word segmentation quality can be seen as an extrinsic evaluation of the soft-alignment matrices produced during training. Our experiments in a low-resource scenario on Mboshi and English languages (both aligned to French) show that RNNs surprisingly outperform CNNs and Transformer for this task. Our results are confirmed by an intrinsic evaluation of alignment quality through the use of Average Normalized Entropy (ANE). Lastly, we improve our best word discovery model by using an alignment entropy confidence measure that accumulates ANE over all the occurrences of a given alignment pair in the collection.