Author: admin

No single answer can definitively claim that Chomsky’s views on AI are either entirely correct or incorrect, as his perspective has evolved over time and encompasses a range of considerations. Chomsky has been critical of certain approaches in AI, particularly those based on statistical learning methods, arguing that they lack a true understanding of human cognition. However, his critiques have spurred valuable discussions within the field. Ultimately, opinions on AI vary widely, and it’s essential to consider multiple perspectives when evaluating its development and impact.

Back before “AI” became the buzzword, we called it “Machine Translation” (MT), or “Computer Assisted translation” (CAT). Though strictly speaking, CAT is a hybrid form of MT. A human is still conducting and controlling the translation process, but he makes use of databases and pattern matching software to find partial or complete matches among a body of previous translations.

When MT started to become a “thing”, stingy clients would rely solely on it, but the slightly more enlightened ones would get a human to carry out post-editing in order to remove the more obvious errors and to render the text into something that was more natural. When my colleagues started to realise this was happening, some of them increased their rates for checking / compare checking / editing to bring them close to their rates for actual translation. Their rationale was because such post-editing work was often just as time-consuming for them as doing the translation themselves to begin with.

It is probable that the demand for actual translators will continue to dwindle into the future as AI improves and becomes more widely used, but there will still be a demand for human post-editors with essentially the same linguistic and subject-knowledge skills as translators. However I think the net volume of work of whatever type for humans in my field will tend to decrease overall.

If the text to be translated is poorly written, then the output will tend to be be equally as poor. It follows from this that Machine Translation tends to work better if the source text is prepared in a particular way, one that the MT database is accustomed to work with. This means you have to unify your terminology, structure all sentences with similar meaning in a similar way, write grammatically (and avoid spelling mistakes), and perhaps simplify your writing, i.e. cut long sentences into two or more shorter ones.

And this means that the source text for translation has to be pre-edited before it can be machine translated, whereas the need for pre-editing tends to diminish if a human is doing the translation. And since pre-editing requires more time and adds to the expense, it may work slightly in the human’s favour too.