Every time I see another article breathlessly announcing a new high-tech glove that will “translate” American Sign Language into English, I wonder what that really means, but I’ve neglected to follow up with research.
A 2017 article from the Atlantic, “Why Sign-Language Gloves Don't Help Deaf People,” answers my questions; it turns out that all of those gloves created over the past thirty years just recognize the 26 letters of the ASL manual (fingerspelling) alphabet. They don’t recognize ASL words; they don’t recognize the parts of ASL that don’t involve finger positions (such as facial expressions); and they certainly don’t translate from one language into another.
(The article also discusses cultural appropriation, and the lack of involvement of Deaf people in the development of these technologies, and technologies that would be more useful to Deaf people; those are important points, but not what I’m focusing on in this post.)
All of those issues are serious problems with the claim that gloves can translate from ASL to English. Alphabet-recognizing gloves are essentially transliterating devices, sort of like a system to convert katakana (or other non-Latin writing systems) into Latin letters. The gloves are slightly more useful than some transliteration systems, because the ASL manual alphabet does correspond exactly to the English alphabet; so if someone who’s signing in ASL spells out a word, then transliterating that word does result in the English equivalent. But most words in ASL are not spelled out with the manual alphabet, and the idea that transliterating the manual alphabet into English letters is the same as translating ASL is a major misunderstanding of ASL.
The article discusses a bunch of the problems with the idea, and I recommend reading the whole article. But one specific point that I want to highlight is only touched on briefly: “ASL (and other sign languages, such as British Sign Language, Chinese Sign Language, and dozens of others) are distinct languages with their own grammars and phonologies, not word-for-word reformulations of a spoken language.”
And so even if a computer system were to address the other linguistic issues (such as looking at words rather than letters, looking at facial expressions, and dealing with coarticulation), it would still not be doing a good job of translating ASL into English any more than a word-for-word translation of any language is a good translation.
That issue could presumably eventually be addressed using AI machine-translation systems, in the same way that such systems are beginning to work well for translating between non-sign languages. But as the article points out, we don’t have the “large data sets of people signing that [would be needed] to train machine-learning algorithms.”
So, in summary: it’s not impossible that computer systems in the future could translate from ASL into English. But gloves that recognize the 26 hand positions of the ASL manual alphabet are very far from being such a system.