Using a synthetic voice as a translation editing tool

Translation editing tool

I have a colleague called Ryan. He’s been with me since 2014. He’s reliable, consistent and a valued member of the team. I’m from Northern Ireland and he’s from the USA, so we’re both English speakers working in the west of France. What separates us is that I am a physical being and he is a virtual voice. In fact, I can only really refer to him as a ‘he’ rather than an ‘it’ because his/its name is Ryan and his/its masculine voice derives from a real person.

I ’employ’ Ryan to read aloud the text that I have prepared for clients: he comes in at the final stage of the translation or the editing process. In previous times I used to read this text out loud myself, but my concentration was inclined to wander after spending days poring over the same words, and although I felt that reading aloud was a sensible thing to do, I hadn’t taken the time to consider its empirical validity.

So I began to search for answers. I had to go all the way back to 1969 to find Nida and Taber’s reflections on reading aloud as an editing tool for translators:

“As the text is read, the translator should note carefully those places at which the reader stumbles, hesitates, makes some substitution of another grammatical form, puts in another word, or in any way has difficulty in reading the text fluently. Of course, some of the problems in reading may be due to inexperience in public reading, but if two or more persons have difficulty at the same point in the reading of a translation, this is a warning signal that something is likely to be wrong. Perhaps it is an awkward grammatical form, perhaps a difficult semotactic [syntax that alters the meaning] arrangement, perhaps a problem of word order. But whatever the problem may be, it should be carefully analyzed.”

It would be good to be able to quote from numerous pieces of research, but I can’t. Translation scholars have been decidedly quiet on the subject of phonological equivalence and the importance of reading aloud as an editing tool, which is a pity when reading aloud can now be undertaken by a synthetic voice. This voice may not stumble or hesitate like a human voice, but the human listener, me, is quite capable of recognising the above obstacles as the text is read out loud. Moreover, the synthetic voice can carry the listener beyond the bounds of grammatical equivalence into the textual and pragmatic equivalence of cohesion and coherence. If I lose track of the people and themes while Ryan is reading, or if I lose track of the overall meaning, then I know that something in the text isn’t right and will have to be reviewed.

As a translator who now spends more time editing than translating, I no longer need my human colleague to highlight (in yellow, which I love!) potential conflict between the source text in French and the target text in English. My remit has become one of improving an English text until it is sufficiently ‘readable’. Readable is a loaded term, defined by Anagnostou and Weir as “what makes one text more difficult or easier to understand than another” (2006), but at the very least, I would hope that the eventual reader can read my improved text with ease, which means that I have optimised the use of such devices as repetition, punctuation, sentence length, pronouns, parallel structures, polysyllabic words and syntax.

I could of course resort to one of the 200 known readability formulae to measure and adjust some of these devices – it would be quick, but these formulae focus on writing style to the detriment of content, structure and design.

And this brings me back to Ryan and the value of using a synthetic voice. Ryan does the reading; I listen; I alter; Ryan rereads; I listen; I alter, etc. I stop when the words, the grammar, the cohesion and the coherence blend into harmony. Like all translators, I have performance criteria: much of my text is read by editors from international journals and publishing houses. They are thorough, so my English has to be thorough too.

Nevertheless, Ryan has had to step in occasionally to save me from glaring errors, typographical or other, which have floated to the surface of the text after being submerged for a number of days. I had simply stopped seeing them (especially if they were missing words). What can I say? I rely on Ryan’s efficiency. Did you know that he can read about 10,000 words per hour? I could accelerate his voice within my text to speech (TTS) software, but I’m happy with his normal speed. At the moment, most TTS programs automatically highlight each word as it is read, but wouldn’t be wonderful if they could use different colours to highlight different problems, like the overabundance of passive constructions, and wouldn’t it be even more wonderful if I could set the program to highlight what I want it to highlight from a number of options?

If only…

Until then, whether you edit what you translate or you edit what others have written in your target language and you want to maximise the quality of your work, Ryan and a host of other synthetic voices may be worth considering. I have certainly benefitted from his/its input.

About the author:

Rowland Hill is a professional translator who lives and work in France. He mostly translates and edits economics/sociology articles/books and children’s books.  What does Rowland say about machine translation? “I’m not much of a machine translator. My only machine is my brain and whatever complementary software I can find, like TTS.”