India is a lucrative market for all technology companies but what seems to have been stopping their progress is the diverse number of languages in the country.
Hence it is a no-brainer that the big three -- Google, Facebook and Microsoft -- are working on translation technologies to solve people’s problems but they seem to be lacking behind when it comes to Hindi or other languages being translated in to English (what we tried) or vice versa.
Despite most of these companies trying to get as accurate as possible, the technology is yet to be full proof and sometimes the results are funny. While Google holds translatathons and has a natural language understanding team that focusses on teaching machines to understand language, Microsoft is working on a cloud-based solutions to cater to customers and enterprises to help wherever translation is required.
Both Google Translate and Microsoft Translator, available on browser and as apps, offer a variety of languages as options along with photo, speech, keyboard input and conversation modes for translation.
However, what seems to be happening is that the machine learning tool or the translation engine either literally translates all words or in an effort to do it better fails and creates funny sentences without a meaning.
Hindustan Times tried to translate a speech given by Prime Minister Narendra Modi on his radio show Maan ki baat and here are the results:
While both the engines tried, neither could give us a good complete sentence. Google which seemed to have most of it right also failed in translating the opening line meaningfully. Interestingly, Microsoft has a bit of ‘hinglish’ -- an effort to keep certain terms such as ‘deshvashiyo’ in this case intact in the transalted speech to deliver more meaning.
HT also tried the photos translate as well as speech. Both the apps need active voice for translation. If you are thinking of playing a recorded clip, it won’t work. Also, Microsoft can translate Hindi to English in audio but can’t do vice versa. But here are some results: