Saturday, October 6, 2018

Language Barriers and Machine Translation

One of the more expensive parts of acquiring global patent protection is having a patent application translated into the relevant language for local patent offices. This is typically viewed simply as an administrative cost of the patent system, though my survey of how scientists use patents suggested that these translation costs may improve access to information about foreign inventions. As I wrote then, "[t]he idea that patents might be improving access to existing knowledge through mandatory translations and free accessibility is a very different disclosure benefit from the one generally touted for the patent system and seems worthy of further study." E.g., if researchers at a U.S. firm publish their results only in English but seek patent protection in the United States and Japan, then Japanese researchers who don't speak English would be able to read about the work in the Japanese patent.

I've also been interested in the proliferation of machine translation tools for patents—which can make patents even more accessible, but which also might limit this comparative advantage of patents over scientific publications if machine translation of journal articles becomes commonplace.

I don't know of much data on the actual economic impact of any of these translation tools, so I was intrigued to spot a new empirical study about the benefits of machine translation for international trade: Does Machine Translation Affect International Trade? Evidence from a Large Digital Platform. Three researchers from MIT and Washington University, Erik Brynjolfsson, Xiang Hui, and Meng Liu, found that the introduction of eBay Machine Translation increased international trade on the platform by 17.5%. They conclude: "Our results provide causal evidence that language barriers significantly hinder trade and that AI has already begun to improve economic efficiency in at least one domain."

Of course, this trade benefit of machine translation is different from the effect on patent disclosure, but the study made me wonder if a similar methodology could be applied by the hosts of patent translation tools (e.g., WIPO, EPO, SIPO) to study the increase in access to patent documents from different countries. Such a study could be a nice complement to survey-based work about how researchers in different countries access information about foreign work, and how machine translation fits into this picture. I'm not currently planning any of this work myself, but if the topic interests you, feel free to email—it seems like a fruitful area for a number of studies, and I'd love to share more thoughts and advice.

No comments:

Post a Comment