Sign Languages, Diglossia And Other Diversions

While working on speech/language technology for other languages sometimes I come across linguistic diversions which are endlessly fascinating. I also quite enjoy reading things where people investigate and learn about a topic unfamiliar to them and feel I should get into the habit of writing more things so here we go. I hope you enjoy reading a bit about sign language and related linguistics

Disclaimer: I’ve never formally studied linguistics or any sign language so I may make mistakes. Feel free to reach out to me on any communication channel if you have some corrections or comments on what I’ve written!

Initial Glances

Sometimes customers have asked about automatically generating sign language animations for some text output of an NLP system as an accessibility feature instead of using subtitles. I have a feeling this is more complicated than simply mapping each word to it’s literal sign and playing them back in order. But this isn’t how spoken or signed communication works either. We have the whole range of expressiveness of body language, facial expressions and tone. It only makes sense that sign languages would take advantage of that to make communication simpler and more effective.

And from an initial investigation this proves true. Sign languages often use spatial grammars, body language and expressions to communicate. We can see that with Arabic sign languages how the hand positions differ can be used to communicate whether one object is in front of or behind another[^1]. You don’t sign the literal sentence for a person being in front of a car, you do the person sign with a hand in front of a hand doing the car sign.

There’s also one interesting definition mentioned in the source, that signed Arabic unlike spoken Arabic isn’t diglossic. And deaf speakers may not be aware of the different forms of Arabic that the non-Deaf community use.

Note “not aware” seems to be because the study of Arabic Sign Languages and accommodating the Arabic Deaf community in education and general life seems to be a more recent effort.

Diglossia

Wait what is diglossia? Well a language is diglossic when there’s two forms of a language a High and Low form. So in Arabic the High form is Modern Standard Arabic, whereas Low forms are the local/regional dialects. This is kind of like register and it’s a bit confusing reading things online on how to make the judgement on when a difference is a change in register or a language being diglossic.

Changing register of speech based on context is often called code-switching and we all do it sometimes e.g. when changing between formal and informal settings. I know I don’t speak in a job interview the same way I speak with friends in terms of language but that’s a register shift not English being diglossic.

How can I mentally define diglossia then? Well here’s an attempt based on things I read here[^2]:

Diglossia is when the gap between the two forms of the language are so wide they are both distinct varieties of the language in their own right. Often one form is used typically for writing and the other form is typically for speaking.

This matches well with Arabic, from our work with the language the vast majority of text data is Modern Standard Arabic, whereas speech data you only tend to find it with some TV and Movie data not in casual speech. There are Arabic dialects closer to MSA than others but that’s a bit of a diversion.

One thing I saw was some resources saying British Sign Language is diglossic and a suggestion that diglossic languages tend to have non-diglossic sign languages and vice versa. But reading further [^3] it seems this doesn’t hold up. The research that said BSL was diglossic seems flawed/outdated.

Village Sign Languages

In places where a large Deaf community have gathered either due to a concentration of Deaf people either schools or institutions, or a high prevalence of genetic deafness.

There’s likely a number of Arabic Sign Languages which are village sign languages and most definitely undocumented. For a documented one I found out about Mardin Sign Language[^4]. A sign language with ~40 speakers currently, it evolved in a family with a lot of Deaf members and arose as a way for the family to communicate.

A lot of resources I wanted to read on this were paywalled so I’ve not really got much other than they exist and I’m curious on how they’ve evolved and any examples where they’ve evolved beyond a village sign language to something more widely spoken. Is BSL an example of this? It seems to have evolved from the 15th Century to the present day with a clear Northern and Southern dialect.

Concluding Thoughts

So using your whole body to sign and combining all of that together to create meaning I have vague thoughts on detecting the timestamps of sign boundaries. Tokenising the human form and creating systems to aid in sign translation. Identifying when some motion or component is a modifier on a meaning and when it applies seems insanely difficult. I’m probably going to try and read up on what people in computer vision and machine translation communities are doing with sign language.

As someone just standing at the entryway peeking curiously in it feels like encoding a sign language in a way it could be an output or input to computer translation systems you’ll encounter a complexity that makes unicode look simple.

Also, we’ve seen LLMs are great at taking all written language and creating links and transformations between them via glorified pattern matching. Part of me wonders if similar unsupervised learning on the signed languages of the world with a visual model could lead to interesting insights or exploration of signed language.

One thing is clear though. All languages are interesting to study and figure out a bit about how they work. I spent about 4 hours reading up on things which is barely any time and I’ll probably spend more time on it. I just wanted to capture some initial learnings.

Initial Glances

Diglossia

Village Sign Languages

Concluding Thoughts

References