Language Translation – Voice to text technology: far from perfect

Any type of automatic translation gives iffy results, but voice to text has an especially long way to go.

In my language teaching career, I have been confronted with software that is supposed to measure the correction of students’ accents and intonation.

The students repeat a sentence into a microphone, and the software turns out a sound graph showing how close to the “right” pronunciation and intonation they are.

For me, the problem is that I can’t even come close -- in my own language!

The programs I have worked with have been honed to British English, which explains my dismal failure – indeed, I got very poor scores -- at speaking my own tongue.

The problems of capturing utterances so that they are linguistically clear (to a machine, anyway) are compounded when the goal is to actually transcribe spoken language into written language.

A recent report from the Sydney Morning Herald makes this point very well. Having “put a message service to the test to see how it fared translating spoken quotes into texts,” the journalists show the sometimes hilarious results when famous movie quotations are run through voice-to-text transcription systems.

The report includes a slideshow and an article, but I would especially recommend the video found on the same page as the text.

By listening to the original extract, you can start to understand why some lines “translate” into text better than others. They are spoken more clearly, and do not include unusual proper names.

By the way, the tests were done in “English to English” voice to text. I can only imagine what would have come up in “English to Spanish,” for example.

I suspect that true voice to text language translation is decades away, and perhaps even an impossible dream.

Betty Carlson

See Also