Detection of phonetic units of the Mexican Sign Language

Current research in Sign Language (SL) recognition has been centered around lexical sign identification. However, SLs tend to exhibit lower vocabulary sizes than vocal languages, as signers codify part of their message through highly iconic signs that may never be lexicalized. In this regard, having methods to detect meaningful linguistic units other than signs may be necessary to improve results in Natural Language Processing (NLP) technology for SL. This paper proposes a method for the unsupervised identification of potential phonetic units in raw SL videos, through the expectation maximization algorithm. The objective is to identify the smallest possible linguistic units that may carry relevant information, as a means to find new features for SL-based NLP. As of the authors knowledge, little to no work exists in the detection of phonetic characteristics in SL, and there is little consensus on what may constitute a phoneme in SL linguistics. In this regard, this work may contribute to further understand the building blocks of SL utterance formation, which may have direct repercussions on how SLs are represented, transmitted and processed electronically.