I know how much you pride yourself on journalistic integrity and I know how keen you are to show your contribution to society, given current debates over the license fee. I also know how hard your staff work. It’s because I know all these things that I have to regretfully point out that your recent piece “Can Computer Translators Ever Beat Speaking in a Foreign Tongue?” contains several serious flaws and is therefore incoherent and misleading.
Translation and Interpreting are different professions
Unlike The Guardian, you often use the words “translator” and “interpreter” interchangeably in your reporting. While this is passable in articles on what someone “spoke through a translator”, it’s really not acceptable when talking about the achievements of computers in the two very different professions.
Conflating translation and interpreting is rather like conflating neurology and urology. No-one would ask a urologist to diagnose a brain tumour. Similarly, it’s not a good idea to ask a literary translator to interpret at a sales meeting.
The difference between translation and interpeting is simple (and has been explained here before). Interpreters work in the moment, at events, meetings or conferences. Translators work on texts after they are written.
Computer Aided Translation tools don’t fit in your ear
That flaw is the source of the greatest proportion of inaccuracies in the article. For example, it speaks about Computer Aided Translation tools as if they can do everything from detecting repetitions to providing in-ear “translation”.
That is simply false. CAT tools are software packages used by translators. Among other functions, they store previously translated sentences and glossaries of terminology. The latest tools include all sorts of features, from the ability to handle a wide range of file formats to integration with machine translation services. This makes them indispensable for many translators.
CAT tools do not and cannot provide in-ear interpreting. That is the sole preserve of in-ear speech translation gadgets. Yes, I know the term “speech translation” is rather confusing but I spent a year writing a book that explains why those gadgets don’t really deserve the title “machine interpreting” yet.
Speech Translation and Credibility
Aside from that error, it is surprising that Andrew Ochoa of Waverly Labs was approached for quotes on the capabilities of speech translation.
You may not be aware but Waverly Labs have made claims that have turned out to be less impressive in the real world. Their first product, the Pilot Translating Earpiece was claimed to have taken us to “life untethered, free of language barriers ” in its glitzy marketing video. At the same time, their own Chief Engineer wrote that the underlying technology was at a place where “two conversation partners can understand reasonably well what each one is saying“.
With their new product, The Ambassador, they claim to offer “professional grade translation“. When I approached them for the data behind their claim, such data was not forthcoming. At the time of writing this, they still have not produced any publicy available data against which to test their claim.
Indeed, that claim has to be questioned since, in your article, Mr Ochoa is quoted as saying “When it comes to expressing emotion and intonation, we need sentiment analysis, which is not there yet but may well be in ten years time”. Either his company has already matched the quality of professional human interpreters, who take such tasks in their stride, or it hasn’t. Which is it?
Suffice to say, I would argue that we should be sceptical of their claim and subsequent predictions. His idea that we just need to add in sentiment analysis ignores the fact that research done since the early 1990s shows that interpreters explain, co-ordinate who can talk, summarise, change language register, reformulate, adjust pronoun use and clarify. And that’s just the tip of the iceberg. I would invite Mr Ochoa and his company to spend more time learning what interpreters do.
What actually is the question?
The final flaw in the article is its imprecision. The title of the article asks “can computer translators ever beat speaking in a foreign tongue?” The answer to that will be “no” for as long as computer translation or interpreting doesn’t give people the same access to the culture, politeness, and expectations that we gain when putting in the effort to learn a language.
If you had asked “will computers ever replace human translators?”, the answer would depend on the skill, field, and expertise of the human translator and what the translation will be used for.
If you had asked “will computers ever interpret better than professional human interpreters?”, the answer would depend on how we understand what interpreters do and whether we believe computers can ever gain the situational, social and cultural awareness that should be the stock in trade of any professional human interpreter.
Three very different questions would have led to three very different answers and three fascinating articles. It is just a pity that your article didn’t manage to cover any other these topics fairly, leading to confusing and misleading results.
Fixing the Issue
Apart from taking the example of The Guardian and making sure that interpreting and translation are no longer conflated, I would like to ask that you review the sources you approach for such articles.
For quotes on machine translation, it’s worth going to universities like Dublin City University (Prof Andy Way and his team are superb). For quotes on interpreting or translation, the Institute of Translation and Interpreting is an excellent UK source. For a realistic appraisal of speech translation technology, go to the team behind the IWSLT2020 conference.
I am happy to suggest sources for future articles myself, especially as I had to hunt them down for my recent book. Feel free to drop me an email at email@example.com and I will either put you in touch with an expert (with their permission) or give you a quote myself.
It’s worth getting this stuff right. People’s jobs and future are at risk.