Evaluating Morphological Plausibility of Subword Tokenization via Statistical Alignment with Morpho-Syntactic Features

TL;DR

This paper introduces a new metric for assessing the morphological plausibility of subword segmentation using probabilistic alignment with morpho-syntactic features, enabling broader cross-linguistic evaluation.

Contribution

It proposes a language-agnostic metric based on statistical alignment with morpho-syntactic features, avoiding the need for gold-standard segmentation data.

Findings

The metric correlates well with traditional morpheme boundary recall.

It is applicable across diverse languages with different morphological systems.

Abstract

We present a novel metric for the evaluation of the morphological plausibility of subword segmentation. Unlike the typically used morpheme boundary or retrieval F-score, which requires gold segmentation data that is either unavailable or of inconsistent quality across many languages, our approach utilizes morpho-syntactic features. These are available in resources such as Universal Dependencies or UniMorph for a much wider range of languages. The metric works by probabilistically aligning subwords with morphological features through an IBM Model 1. Our experiments show that the metric correlates well with traditional morpheme boundary recall while being more broadly applicable across languages with different morphological systems.