Patents In, Garbage Out? When machine translation mangles the prior art

Voltaire had it right when he said: “Woe to the makers of literal translations, who by rendering every word weaken the meaning.” Was he foreseeing the unfortunate results of feeding a Japanese specification through Google Translate? Probably not, but he had a point nonetheless.

"Shoes not allowed" - Michael Cornelius
EPO decision T 1343/12 concerns a machine translation of a prior art Japanese reference used in an inventive step objection. The Board, examining the ambiguous translation, was unable to reach a conclusion as to its meaning, and remitted the case back to the Examining Division “for re-consideration of the issue of inventive step in the light of a certified translation of D1.” What intrigued the IPKat was that the decision is silent on who has the legal responsibility of procuring this certified translation and who is expected to pay for it.

The Decision
The claims required that a surfactant in an oil-based composition was present in an amount of from 10 to 30% of the composition. The machine translation of document D1 had a single sentence which the Board held to be hopelessly ambiguous:

"The amount of the surface-active agent used besides the above of below 10 mass % is 0.1 to 8 mass % still more preferably preferably among the oils of this invention." [sic]
On one interpretation this could mean "up to 10%, more preferably 0.1% to 8%," teaching away from any values higher than 10%. On another interpretation (which is perhaps not convincing when the sentence is read in isolation, but is more plausible when read in context as the decision explains) this could mean 10% of one surfactant component plus 0.1% to 8% of another surfactant, i.e. giving a total of up to 18%, i.e. within the claimed range.

The Board held that this ambiguity could not be resolved from the machine translation, concluding:
4.8 Remittal to the examining division 
4.8.1 Since it appears that the assessment of inventive step hinges to a great extent on the true disclosure of D1 in terms of surfactant contents and the corresponding effects, a definitive conclusion regarding the issue of inventive step in the light of D1 cannot be reached based on the machine translation thereof, which appears to be unclear and/or ambiguous in several relevant aspects. 
4.8.2 Therefore, the board considered it appropriate to remit the case to the examining division for re-consideration of the issue of inventive step in the light of a certified translation of D1.
There then immediately followed the operative part of the decision:
For these reasons it is decided that: 
1. The decision under appeal is set aside. 
2. The case is remitted to the examining division for further examination of inventive step based on the claims 1 to 4 according to the request submitted during oral proceedings and on a certified translation of document D1.
Insofar as it goes, the decision is sensible. To evaluate a prior art document, one must rely on the content of that document or, as in some older cases, a prior published English abstract. If the content of the document cannot be unambiguously established but it appears relevant then it makes sense to establish that content definitively before proceeding further.

But did they need to remit to the Examining Division? An alternative approach was seen in an older decision: in T 243/96 the Board had doubts over the fidelity of the abstract to an original Japanese document, so they did the needful thing by supplying the applicant with a full translation of the original document. Presumably the Board here could have done the same, assuming that the doubts over D1's content were known to the Board in advance of the hearing. (In the interests of even-handedness, which is not normally her style, Merpel presumes that the applicants, being a couple of Japanese companies, could equally have resolved the ambiguity over D1’s content much earlier and at not too much expense, had they wished to do so. But, says the IPKat, why should they be expected to do so? And who's to say that a good translation will not reveal ambiguity in the original?)

But who pays?
"Unrecycling" - Drakhrub
It may be that the simplest answer to the IPKat's questions over who is responsible for getting the translation prepared is that the IPKat is imagining difficulties where none exist. Presumably when a Board of Appeal makes an order, this is addressed to the EPO department of first instance and not to the applicant. It would therefore be up to the Examining Division to implement that order, which presumably means that they must procure a certified translation at the EPO’s cost. This would also chime with the IPKat’s understanding that whoever alleges that a document deprives a claim of patentability would normally be expected to put forward their complete case without help from the applicant. 

But perhaps the Board was being more pragmatic and deliberately omitted to address who should procure the translation. It may have been wary of either (a) requiring the applicant to incur an additional cost for which there is no basis in the EPC, and which might be ultra vires, or (b) explicitly and publicly saying that in cases where a machine translation is sufficiently ambiguous and potentially relevant to patentability, the EPO is obliged to provide a certified translation to the applicant. While the cost of one translation won’t break the bank for either the EPO or the applicant in this case, the costs would add up for the EPO considering the number of foreign language references (i.e. not English, French or German) relied on per annum across the organisation. 

Merpel says maybe there's another way of figuring out what the Board intended. If machine translations can turn a meaningful text into an ambiguous one, might the program not sometimes translate ambiguity into clarity? It may be worth putting the final paragraphs of this Decision through Google Translate to see if any answer emerges as to who will fund the missing translation.

An interesting footnote, and a counterpoint to the assumption that the EPO should procure the translation, can be seen from the file wrapper. The EPO search cited the Derwent abstracts for three Japanese citations. Surprisingly (at least to this Kat) the search opinion started by requesting the applicant to provide the Examining Division with a translation of the three documents into an EPO language because their content seemed to be "of the utmost relevance". Rather than human translations, the applicant submitted the machine translations at issue in the appeal, which of course they were entitled to do. Indeed it is hard to see why the applicants were under any obligation (other than one of courtesy) to comply with a request to provide the Division with full translations.

Maybe readers have thoughts on whether applicants should be expected to supply the EPO with better ammunition to help invalidate their applications?
Patents In, Garbage Out? When machine translation mangles the prior art Patents In, Garbage Out? When machine translation mangles the prior art Reviewed by David Brophy on Friday, November 14, 2014 Rating: 5


  1. I personally think that, if the EPO cites it, the EPO should bloomin' well pay for it to be cited properly. I mean, they c should know what a disaster are machine translations, which sometimes appear designed to amuse rather than inform. I mean, what would you do when you encounter, in the middle of an application on washing detergent, a lady called Lily Earle? Co-inventor? Co-applicant? No, the common translation from the Japanese of "lilial", a common perfume aldehyde.

  2. I've certainly had the EPO cite specific paragraphs of Japanese prior art documents (for which they only provided the English abstract). The onus was on me (representing the patentee) to provide the translation and proove that the document didn't actually disclose what the examining division alleged it did. So the ED alleged something based on a document in a non-EPO language, and yet I had to provide the translation to resolve the ambiguity - still seems unfair to me.

  3. Once the BoA have decided that the translation is ambiguous with regards to a key passage, then I would expect the EPO to have to foot the bill to provide a proper basis for the objection (or use a different document).

    In first instance proceedings, I would say it is all about who has discharged their burden of proof. This is of course initially with the EPO, but a reasoned objection based on a translation would probably pass the burden on to the applicant.

    If in turn the applicant can make a reasonable argument against the clarity of the specific passage, then they could well pass the burden back to the EPO.

    On the other hand, simply stating that machine translations cannot be trusted is unlikely to convince. In such a case the applicant would have to file their own authorised translation to prove their point.

    All IMVHO, of course.

  4. I had something similar a few years ago. In the written opinion that accompanied the search report, the examiner had referred to what turned out to be paragraphs of a machine translation obtained from the Japanese Patent Office web site. If you know how to navigate around its sometimes obtuse English-language terminology, they are available for the later specifications that had to be filed in electronic form. As I knew from experience, these translations are not downloaded from a database of translations, but generated on an ad hoc basis, and so translations downloaded on different dates are seldom identical due to translation engine updates. The examiner had been unaware of this. In my case the accuracy of the translation was not really at issue, but to ensure that we were both working from the same document one examination got underway, with the examiner's agreement, I sent the examiner a copy of the translation I was working from to ensure that we were not at cross-purposes. Had there been prospective problems with accuracy, I would have got it translated by a Japanese patent attorney that our firm used to use.

    The most "interesting" Japanese machine translation I experienced was an electrical case where an electrode of an integrated circuit was described as having the shape of a XXXXX, where XXXXX was a Latin word that I cannot now remember. When I looked it up, top of the list was a medical term for the region of the anus where the surface tissue changes from external skin to internal tissue! The correct rendition in the context of the invention was found much further down, and was a fern-like structure.

  5. I agree with AnonAnon. Note that the board in this decision seems to trust the machine translation in so far as it is sufficiently understandable and unambiguous. On the other hand, the appellant seems not to have contested its accuracy.

    But I noticed something strange. The ED started out by citing three WPI abstracts and REQUESTING the applicant to file translations... That's just ridiculous.

    The applicant complied by filing machine translations.

    So who knows, this ED might now again REQUEST, obviously without any legal basis, the applicant to file a certified translation.

  6. In my case, thee examiner had not mentioned in the written opinion that he was working from a machine translation obtained from the Japanese Patent Office web site: this was only established after I had spoken to him on the telephone to clarify things. Paragraphs of the original Japanese text had been referred to without actually reciting the text of the passage relied on.

  7. I believe that the EPO should pay for the translations - Applicants pay endless renewal fees while the EPO lets applications sit 'pending' for five, six, seven years or more. Instead of spending Applicant's money on artworks and fancy buildings, is it too much to ask that the EPO spends some of it on the functions it was put in place to perform?

  8. not only Japanese documents "suffer": look at machine translations of the claims of EP 0 921 916 B....

  9. As a developer of machine translation technology, we're constantly trying to preach about correct usage - specifically about NOT using the output as is in circumstances like this. In these cases, the machine translation is used to determine that, yes, this document is relevant and a "human" translation is needed. (in this case, the machine translation can be used as a draft based on which to produce the final translation - so called post-editing which costs less than full translation from scratch).

    If the offending text was not identified by the provider of the data as a machine translation, then surely the liability (for cost?) lays with them UNLESS they indemnified themselves in the T&Cs or usage policy, in which case the end user is responsible for their own course of action.

  10. I revisited this matter and chased my tail on finding the cited D1 patent until I discoved that the PN cited in the decision report by the EPO was wrong, should be JP2002-069435. Now, Japanese patent numbers can be challenging, especially the older ones, but mixing up a 5 with a 6.....
    Adittionally, with all of hoopla about computer translation, presumably the two Japanese appellants thought it was acceptable to cite a computer translation ??? And the EPO in preparing the hearing, thought it was OK to start gussing based on the computer translation????


All comments must be moderated by a member of the IPKat team before they appear on the blog. Comments will not be allowed if the contravene the IPKat policy that readers' comments should not be obscene or defamatory; they should not consist of ad hominem attacks on members of the blog team or other comment-posters and they should make a constructive contribution to the discussion of the post on which they purport to comment.

It is also the IPKat policy that comments should not be made completely anonymously, and users should use a consistent name or pseudonym (which should not itself be defamatory or obscene, or that of another real person), either in the "identity" field, or at the beginning of the comment. Current practice is to, however, allow a limited number of comments that contravene this policy, provided that the comment has a high degree of relevance and the comment chain does not become too difficult to follow.

Learn more here:

Powered by Blogger.