Language Barriers and the Speed of Knowledge Diffusion

Language barriers are a well-known obstacle to knowledge diffusion, but quantitative evidence of such barriers is very rare. For domestic dissemination of novel technological developments that are only described in a foreign language, there must be someone with the ability to directly understand the literature, or its translation, both of which are expensive and slow the diffusion process. The question is how important such a barrier is. If they are indeed very important, then the social value of investment in reducing these barriers, through automated translation, for example, will be significant.

To fill this evidence gap, this study uses the pre-grant publication system introduced by the American Inventors Protection Act (AIPA) in 2000 to provide causal evidence on the impact of language barrier on the speed of knowledge diffusion. The patent system mandates publication as a pre-condition for protection of a patent right, and requires that the application, if filed in a foreign jurisdiction, be filed in a language that is accepted by that jurisdiction. Because acceptable languages differ by jurisdiction, the patent system has the function of forcing translation and disseminating technical knowledge worldwide. Pre-grant publication systems often mandate that this be done as early as 18 months after filing an application in a country.

Since the United States introduced their pre-grant publication system with the American Inventors Protection Act, the timing of publication in English of not only domestic applications but also applications from foreign countries accelerated in the year 2000. At this time, applications filed in foreign countries were already published in the local language 18 months after the application. Therefore, the impact of language barriers in the context of technological knowledge can be quantitatively assessed by examining the extent to which the U.S. Inventors Protection Act (AIPA) accelerated the timing of the use of foreign inventions by US inventors as input into their own inventions. This study does just that.

In particular, we focus on the use of knowledge disclosed in the patent applications by Japanese inventors (filed only in the U.S. and Japan). We use the time until the first inventor citation made in a U.S. patent as the measure of the delay to the start of knowledge use. Prior to November 2000 (when the AIPA was implemented), patent applications filed only in the U.S. and Japan were published in English only upon their U.S. grant, and 18 months from their first-filings thereafter. However, all of these applications were published in Japanese 18 months after their first-filings, even before the AIPA. Within this natural experiment, therefore, US inventors are the treatment group, as they receive Japan-originating knowledge significantly earlier than before, and the Japanese inventors are the control group, whose access has not changed.

As the figure below shows, before the American Inventors Protection Act (AIPA) became effective, the average time to the first citation was on average significantly longer for U.S. inventors than for Japanese inventors (by 0.27 in logarithmic terms, or about 31%). After the American Inventors Protection Act (AIPA) became effective, the difference dropped to 0.14 on a logarithmic scale. In other words, we can estimate that the difference (0.13 on a logarithmic scale (about 13%)) is the policy impact of the AIPA, which is large, amounting to around a half of the citation delay of the US-based inventors before the AIPA.

Figure 1 also shows that the Average Time to First Citation has become shorter, not only for U.S. inventors but also for Japanese inventors due to the American Inventors Protection Act (AIPA). There are several potential explanations for this. First, the fact that an application is also filed in the U.S. signals that the invention is important in the U.S. market, which is very likely to increase the frequency of citation in U.S. patents. In addition, under the U.S. patent system, applicants are obliged to disclose known prior art to the Patent Office. If there are relevant patent applications published only in Japan and relevant applications published in both Japan and the U.S., and either can serve as prior art, the latter is considered to be the more likely choice. Therefore, the application which is published in both Japan and the U.S. has a higher probability of being cited. Both effects apply not only to Japanese inventors but also to U.S. inventors. Thus, a part of the reduction in the time to the first citation observed for the U.S. inventors in Figure 1 reflects the preferences of the inventors to cite the local prior art.

We can control for the above effects on early citations, which are unrelated to the removal of language barriers, by introducing Japanese-U.S.-European trilateral applications in which an application in English is published at the European Patent Office 18 months after first filing. Since the only crucial difference between the two samples is the existence of pre-grant publication of the English, differencing the means of the two samples for both the US inventors and the Japanese inventors gives us an estimate of the effect of early publication in the US, controlling for the above preferences of citing local prior art. The result gives an estimate of 0.12, which is very close to that obtained from Figure 1.

Upon extending these analyses, we find that the impact of these language barriers on knowledge diffusion is nonexistent for inventors in large firms, but is significant for inventors in small firms as well as firms with little investment in the Japanese market. Further, we find that the impact of the AIPA is almost completely confined to citations to (or, conversely, knowledge diffusion originating from) high quality patent applications. This is consistent with incentives for, and constraints on, translation. Firms with higher expected revenues from translations understand the content of the Japanese application before the US grants the English-language equivalent, even before the AIPA. In addition, without translation, it is difficult to know the quality of the patent, making it difficult for firms to implement targeted translation of patents that describe the most important or relevant inventions.

Our results suggest that the pre-publication of patent applications provides an important public good for cumulative innovation through the early translation of foreign patents into domestic patent literature. They also suggest that sophisticated and technologically specialized machine translation of technical information has the potential to have a large impact on the global diffusion of knowledge, and that ensuring easy access by small and medium-sized firms and individual researchers to such service is important for realizing such potential.