Gary, developer of MT systems

I use Google Translate to translate Chinese spam messages which regularly get posted on my website – just in case any of them are of interest!

Actually most of my interest in MT is not about free online MT tools like Google Translate, but about professional tools. I created my first rule-based MT system over 25 years ago and have been involved in developing machine translation since then. My main project is a commercial neural MT system for translating  between Dutch and English. But I also experiment in neural systems for other languages, most recently Indonesian and Turkish.

Since my own knowledge of Indonesian and Turkish is very poor, I’ve found a use for online MT in this experimental work too. For example, I could translate something into Indonesian with my own engine, then back-translate it through Google into English to see how well my model has learned the patterns of a new language. Other times I simply translate the same thing using both my own and Google’s engines and compare the results. It helps me check that my neural network building is going in the right direction.

I really do also Google-translate Chinese spam messages. The translations are usually very basic but tell me what I want to know about the spam – mostly that this is not something in which I could be remotely interested. Let’s not forget that before the arrival of online MT, companies would spend large sums of money each year translating documents that turned out to be completely useless. Online MT saves money.

In general I’m happy to recommend MT to friends for general information purposes: double-checking the translation of a menu in a Chinese restaurant, translating the instructions sent with self-assembly furniture, or reading spam. Most general online translators are quite capable of fulfilling such tasks nowadays. I also recommend MT as a language learning tool.

However, I would not rely on unchecked, unedited MT for information I need to make important decisions. Apart from the obvious fact that the machine translation may be fluent yet incorrect, we have to allow for the possibility that a decision may be challenged in the courts. (Example case. See MT use in the news for more links to news stories on the use of MT).

One thing I’d like people to understand and remember is that the best results are obtained when the MT system is “trained” to handle the type of texts being translated. In online tools like Google Translate, this is not the case.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s