Don’t fall for AI ‘translation’
There has been such a proliferation of ‘AI translation’ tools in recent years that many people might reasonably wonder whether they actually need a professional to translate their documents.
Here’s a quick way to find the answer to that question in any given case. Just ask yourself the following two questions:
Does accuracy matter?
If the answer is yes, then you need a professional.
Does confidentiality matter?
If the answer is yes, then you definitely need a professional.
Translation is a profession that is rarely well understood by people who are not translators or interpreters themselves. The result of this has been that people have long had trouble understanding the difference between a bilingual person and a translator (or, indeed, realising that there is a difference), and, more recently, the difference between a translator and a large-language model or LLM (what is commonly marketed as ‘AI’, in order to misleadingly associate it with concepts from science fiction).
LLMs are software. They are not conscious. They do not think. They do not comprehend. They merely do what they are programmed to do, which is to calculate the most probable sequence of words based on the ‘training data’ they have been fed, data that now include a growing percentage of utterances generated by other LLMs. Sometimes, that may result in an accurate rendering of the source phrase in the target language; often, it results in output that is superficially coherent – and thus looks accurate to someone who does not know the language of the original – but is readily recognisable as garbage to anyone familiar with both languages involved.
Translators are professionals who carry out a highly intricate operation involving comprehending the original and producing a rendering in the target language that is as close to the original in all respects – literal meaning, legal effect, the overall vibe of the thing –as possible. This is rarely a straightforward process, and often requires a good working knowledge of the subject-matter and familiarity with research methods, both general and specific to the field in question.
Translators, being humans who have made human language their profession, understand how human language works in a way that no LLM can.
Human language is inherently context-dependent. That context may be linguistic and provided explicitly within the text being translated itself. It may, however, also be implicit. When communicating with others, we often omit context that we assume (correctly or otherwise) is known to the person we’re talking to. After all, if the discussion concerns something or somebody everybody in the conversation knows, there’s no need to remind people of who or what that is. The result of this is that a lot of things can end up sounding ambiguous to those who are not privy to the unstated context.
To take an example I’ve encountered more than a few times, let’s assume that you’ve been handed a banker’s box full of e-mail correspondence and chat logs in discovery that now need to be translated into whatever language the court is using. These communications will often involve a lot of unstated context, such as who the participants are talking about, what their relationship with those people is, the nature of the discussion, and other things that everyone participating in the conversation can be assumed to know.
This puts the reader in the position of a person who is eavesdropping on a conversation. You can’t very well go over and ask the people who are having this conversation you weren’t meant to hear in the first place what they meant by a particular turn of phrase, even more so if the conversation happened several years ago between people who may have very good reasons not to provide full and accurate information about that conversation to outsiders such as yourself.
The result is ambiguity and uncertainty. What does it mean, for example, when one of the participants in the exchange remarks that there’s ‘a lot of dough on the table’? Does this mean that there is a physical table with a large amount of that classical substance used in the process of sandwiching food between other food on it? Does it mean that there is a physical table with a large amount of cash money on it? Or is the table perhaps metaphorical? Is this person saying that a large amount of money has been offered? Or are they saying that this table needs to be cleaned?
Without clear context, say, someone else asking ‘How much?’ and the first person giving a number of loaves, the reader can only guess at the meaning.
The trouble for translators is that ambiguity is not equally distributed between languages. Not every language, for example, has a word for dough that can also mean money, nor does every language use on the table to indicate that something has been offered. Some languages even use different expressions depending on whether something is on the table in the sense of sitting on top of a table or whether it has in fact hardened and become stuck on the table. In cases like these, and they come up all the time, translators will be forced to commit to one reading or another.
A good professional translator will seek clarification in such cases, where possible, or, at the very least, flag the ambiguity so that the client doesn’t rely unduly on language that may actually mean something totally different.
LLMs are not conscious, and they are incapable of comprehending language, much less understanding the intricacies of unspoken context that are challenging even to humans. An LLM will simply calculate the most probable combination of words in the target language based on its training data, leaving the client blissfully unaware that they may be about to base an entire filing on what in another context would have been a humorous misunderstanding.
If accuracy and confidentiality matter, you need a pro.