Can Video Language Translators Handle All Languages?

Do video language translators work with every language? Existing video language translator support a good number of the most spoken languages but are limited when dealing with specific dialects, little known languages, and nuanced local differences. More advanced platforms, such as the one that Google uses for its translation models, cover close to 100 languages and dialects, with high accuracy in large languages like English, Spanish, Chinese or Arabic. But, this covers just a small segment of 7,000+ languages spoken on this planet today which has rendered some smaller languages not been catered for properly. Google says these 100 languages account for about 90% of the world's online population, highlighting not only how far their net casts in terms of geography, but also how limited by definition inclusivity can be.

It is easy to be assimilated when language has case-like grammar or basis on culture, and has almost zero dot digital visibility. As a result, automated translation is nearly impossible for many indigenous languages' dense databases, like Quechua or Xhosa. Take, for example, the case of Inuktitut (spoken by about 40,000 people in Canada) which Microsoft Translator recently began supporting as part of a broader effort to fill linguistic blindspots. Nonetheless, it required months of working with native speakers to generate the necessary data, highlighting how expensive (in terms of resources) expanding language support is.

With the reality created by machine learning and neural machine translation (NMT) technology, this changes to include accuracy for supported languages across a context of translations as it needs more nuance for dialectal variations or colloquialisms. Languages such as Arabic — with substantial regional variation — are difficult for machine translation to get right without being trained on a wide range of idiosyncratic data. OpenAI claims that models with more than 175 billion parameters and exhaustively trained on expanded datasets function great when using the standard language but are still unable to process nonstandard forms or colloquialismsified languages—the OpenAI model called GPT-3 is one of these.

The effect of this constraint on business models is profound. While even Netflix invests in the localization of content, it is obviously limited to languages with a large audience and market potential. Major streaming services turn a blind eye to less widely spoken languages, as indicated by a report from Statista that has only ~30% of localized content being translated into lesser-known languages. While this is a cost effective method, it does limit access to speakers of languages with smaller populations.

But educational platforms have their challenges too. Translating educational materials to even the 10-15 major languages, as done by platforms like Coursera, is beyond inadequate for students in remote environments with dialects far removed from standardized orthography. Rural populations in regions like South Asia and Sub-Saharan Africa are particularly affected, as this language gap affects engagement and accessibility.

Video language translation tools open up multilingual content to billions, yet still encounter obstacles. Language support is likely to grow as the technology evolves, though complete coverage of all languages may require significant uptake, collaboration and time.