It is widely expected that the Unitary Patent Court (UPC) will open for legal proceedings in early 2023, and then also Unitary patents (UP) can be registered. For a transitional period of 6 to 12 years, the registration of a UP is to be filed together with one full translation of the specification within 1 month after grant: if the specification is English, into any other official language of the European Union (not limited to UPC-participation states), and if the specification is French or German, into English. In the event of a dispute, a further full-text translation may be required at the request of an alleged infringer or a competent court. Hence, statements that the translation issue of EP patents after grant would be resolved by the UP are not completely true. Only additional translation requirements otherwise needed for validation of the participating member states are avoided from the beginning.
The majority of granted EP patents have an English specification, and for those a daunting question is which language should be selected for the UP translation. Most effective would be one that can be re-used to a significant degree:
- If an EU state not participating in the UPC (see first post of the series) with full-text translation requirement such as Spain shall be validated in addition to the UP, the translation in that language can be used for both without any further adaptation.
- If the EP patent claims priority from a first filing in a language of an EU member state, apparently the priority application prompts to be used as basis for the translation. The same applies if the EP application was originally filed in such language and only later translated into English.
- If none of this is the case, the German or French claims filed in response to the intention to grant can serve as a partial translation and may prompt to choose one of them. A further consideration is then the language of the most likely jurisdiction for proceedings in case of a dispute, which is often Germany (see third post).
- A further option may be a re-use for obtaining parallel protection in addition to a UP as safety measure. Validation of a national bundle patent from the same EP patent in a UPC member state is excluded by the unitary effect, and a divisional application for basically the same patent is rather expensive. A low-cost alternative would be to branch-off a German utility model with same scope, which thereby effectively becomes an examined right. In technical fields accessible to protection by such right, this allows to cover the most often validated EPC member state for 10 years, even if the UP gets revoked (or if there are concerns that validity of the UPC system as such will be challenged). Such utility model requires a full-text translation into German which could be the corresponding UP translation.
Another relevant question refers to the type of the translation. The transitional period of at least 6 years (extendable up to 12 years) is supposed to last until high quality machine translations into all official languages of the EU are available. The translation to be filed during this period is legally for information only, but according to the preamble of the respective Council Regulation (EU) No 1260/2012, “translations should not be carried out by automated means and their high quality should contribute to the training of translation engines by the EPO”. Hence, downloading a machine translation from EPO Patent Translate and filing it as UP translation would appear to be a misuse of the system. Neural networks do not really improve by being trained with their own translations. From my perspective, it can be seen as a general principle of our profession that any reduction of quality is a step backwards. Even though not legally binding, nobody can prevent a judge from being influenced by an incorrect translation in her/his mother language and a conflicting party can be expected to point to that. Overly aggressive replacements of human translators should also consider potential job losses and counter reactions. Full-service providers have translation charges included in their overall price calculation, and in certain countries, IP firms still live to a significant degree from translations. This already prevented some states from signing the London agreement and may be a driving factor for extending the transitional period or not ratifying the UPCA.
On the other hand, the difference in costs of a fully manual human translation compared to a free computer translation is significant so that intermediate solutions are worth to consider. The reality is already no black-and-white world of either human or computer translations. While machine translation tools clearly do not reach the quality of professional human translators, they can well support them by providing preliminary translations in a quick and convenient way, which are then to be optimized by a human user.
Currently, however, most machine translation software suffers from consistency issues. A well-known problem is that it may translate individual terms in a manner which does not fit to the specific technical field of a patent or which changes over different parts of the document. And when the human translator adapts the wording in one sentence, the limited context scope of the algorithm does not apply the adaptation to remote sentences, so that the next iteration of the same word may be translated differently. Glossaries known from other computer aided translation (CAT) tools are barely combined with neural networks or limited in flexibility. In the end, it often requires numerous manual steps which reduce efficiency.
I admit that I am not neutral in what follows, but intend to spread awareness and will try to stick to the facts. IP.Translator is a software developed by European patent attorneys which addresses the aforementioned issues via a deep neural network specifically trained for patent translations only. It implements a proprietary AI algorithm that searches the already translated text for similar segments to provide translation proposals for the remaining parts which correspond to earlier user adaptations. Similarly, already existing claim translations (e.g. filed under Rule 71(3) EPC) or translations of patent family members can be loaded as translation memory. This way, repeated manual adaptations are significantly reduced and consistency is improved. Such approach allows to obtain a translation at the quality level of the human translator who is using the software – while increasing throughput significantly. This has been confirmed by users from some major German patent firms, who previously relied on other popular translation platforms.
If the transitional period shall be a continuous one rather than a single step at the end, it might be time to reconsider certain translation processes.