mojilove

dictionary jockey

JA→EN translator. Overeducated and unlearned. Writing systems / shmups / nanoloop / lumines / puns / nonsense / memories / banality

avatar stamp made by my thrice-verified soulmate. header made using The Death Generator


日英翻訳者 / ユダヤ人 / 文学バカせ / 文字マニア / 小並感の塊 / 人間(堕落者)

身体の104%が文字と文字愛でできおり、残りの29%は肉体。STG・PZLとFM音源も好き。

日本語垢: @mojilove-j


website button



and you know what I said about what to do if I have a Take...
(seriously, any take i have has usually already been presented better and more concisely by someone else)
but still

there is a bit of buzz going on about a certain tl agency that is implementing reduced rates for "fuzzy matches" - parts of the text that are not complete repetitions but that resemble other parts of the same text (I am assuming they are talking about the same text. charging fuzzy rates based on matches with other texts would be absurd but not inconceivable coming from a company)

I think there is a case to be made for for fuzzy matches if (a) the text in question is highly repetitive and informative rather than expressive, like an instruction manual, and (b) the people working on it are getting a decent base rate to begin with. but neither (a) nor (b) seem to apply in this particular case as far as I can gather (it's not like that one agency is the only bad company in the industry though)

imagine you were translating an EN fictional text. consider how many different meanings or inflections there could be in the single-word utterance of "Fine." even the exchange "Fine?" "Fine." would be treated as a "75%-99% match" regardless of the characters saying each line

fuzzy matching is a homogeneous approach to dealing with texts, and is not compatible with any kind of expressive or emotive work


You must log in to comment.

in reply to @mojilove's post:

I don't know so much about Lionbridge, but they don't seem to have a good reputation at all based on what people had to say in the wake of this thing... it's sad to see

Also, let's say a work has 10 instances of "Fine.", and it turns out 8 of them, or even all 10 of them, are functionally the same utterance and should/could be translated identically.

How do you know which 8(10) they were? You had to read them? In their contexts? And translate them?

oh yeah this is a big thing too - the 10% rate for total matches doesn't really cover the effort needed to consider the context and make sure that an identical translation would work (not to mention adjusting or rewriting it if it doesn't work)