Thursday, 24 March 2011

EPO and Google to break the language barrier - but not yet

The EPO announced some news today, which has been in the offing for some time. With the help of Google, they will soon be offering a much improved machine translation service for patent publications.  This is very good news, and I for one am looking forward to seeing how good Google's methods of translation can be when applied to patents (I am actually quite hopeful that it will be good).

I couldn't help looking to see if the EPO were practising what they had in mind, by taking a look at the German version of the EPO's news page.  This is what I saw:

Oh dear.

1 comment:

  1. I have a problem with having translations performed by Google, as they log and correlate any user interaction, creating grave privacy issues.

    In my previous incarnation as a patent examiner, I often looked up on the internet the names of the inventors and of their employers, to find out who they were and what they were up to. Not too rarely I gained some precious insights about the invention (e.g., a description of the actual products), and sometimes even novelty destroying documents. For instance, I was confronted with the case of a company hiring an inventor straight out of university, and attempting to reformulate his doctoral thesis into a patent application. Such a document wouldn't have been found in the internal databases.

    I was made very wary of Google after I entered a very specific search request at work. Sometimes later, while at home, I had an inspiration about the file, and started entering a similar request in the search box. As I was typing I was rather horrified to see my previous request, which had been entered from a completely different computer connected to a completely network, being proposed as a possible search query by the auto-completion feature. Fortunately the information was not too sensitive. Did Google make the connection between my home computer and the one at my workplace? All trafic from the EPO come out of a specific block of IP addresses.

    Serving the translations through an EPO operated Chinese wall might be better. In any case I feel that it belongs to the EPO's missions to collect, classify and distribute information. This function shouldn't be externalised.