Patent Translation

But, weren’t you a biologist?

Years ago, by the time I had been working full time and with satisfaction on patent translation on a freelance basis, people were surprised to hear me say that I translated patents all day long. I remember their befuddlement, which was expressed more or less in such terms: “But, weren’t you a biologist? And why in the world should a patent be translated? And, are there really that many patents to translate, such as to warrant a full-time commitment?”. Well, for my part, the answers were fairly obvious and predictable.

In Italy and other countries in Europe, patents need to be translated in order to be validated, that is, to be valid in the country in question. With the London Agreement (on May 1, 2008), some countries ratified some rules to streamline the procedures relating to the translation of European patents. By the Agreement, which is optional for the member States of the European Patent Convention, the Signatories waive their right to require the mandatory translation of the text of European patents as a necessary step to validate patents in their country. The Agreement establishes that Member States with one of the three official languages of the European Patent Convention (EPO), namely French, English, and German, no longer require translation into their own language at the time of national validation. The other Member States, whose official language is other than French, English, and German, have the right to request, if not otherwise available, the translation of the European patent into one of the three official EPO languages of their choice, with only the translation of any claims being in their own official language. To date, the Agreement has been signed by thirteen EPO Member States. As of May 1, 2008 it is no longer necessary to translate European patents for validation in France, Germany, Luxembourg, Monaco, the United Kingdom, and Switzerland/Lichtenstein, while Croatia, Denmark, Iceland, Latvia, Netherlands, Slovenia, and Sweden only require the translation of claims in their national language. In all other countries, the translation of the entire patent text is still required.

The correct translation of the patent is, therefore, a “bottleneck” for the purposes of giving it legal value in the countries concerned. Although I am not a trained linguist, I nevertheless plunged into this discrete sector, which, over the years, afforded me the opportunity to hone the valuable linguistic and technical skills (think of CAT tools and machine translation) that I originally lacked; such experience primed me to tackle any “technology sector” (typical patent jargon): from biotechnology to mechanics, from organic chemistry to information technology. Conceivably, my scientific training combined with a certain stubbornness have allowed me to fill the linguistic and technical gaps, thereby enabling me, over time, to deal with disparate topics. However, just to take a step back, what is a patent?

The patent: a mix between the legal lexicon and technical knowledge at the frontier of technology

The patent is a legal title under which the owner is granted an exclusive right of economic exploitation of the invention, in a country and for a set period of time, and that prevents others from manufacturing, marketing, or using the invention without authorization. The word patent comes from the Latin word patere, which means “to lay open”, i.e. to make available for public inspection. From the macro-textual point of view, the patent consists of two fundamental elements: description and final set of claims. The description, which is the most substantial part, is in turn constituted by the background, summary of the invention, and detailed description thereof. The second portion of the patent, i.e. the set of claims, defines and delimits the scope of protection of the patent in an unequivocal yet succinct manner. In terms of volume, a patent text can range from a few thousand words to more than 100,000 words, depending on technique sector considered and type of patent. Well, many people may lick their chops upon hearing about such volumes! Regretfully, these volumes are compensated at relatively low per word rates, which, at first glance, could discourage many people. But let’s not despair. Later on, we will indeed see how, thanks to the use of CAT tools, termbases, and machine translation (especially if populated with patent data), it is possible to make patent translation a very interesting business, notwithstanding the relatively low rates. 

But what are the critical issues in dealing with the translation of a patent text? First and foremost, it is necessary to know the patent jargon, that is, a series of words and expressions used in the intellectual property (IP) sector, which must be translated in the appropriate way, just as they were translated by a patent attorney. Secondly, it is essential to stay true to the original text, replicating, as far as possible, the syntactic structure of the text and avoiding, for example, the use of synonyms. The consistency between source and target texts, at the lexical, morphological, and syntactic level, must be as unambiguous as possible.

However, in terms of translation, some patent texts can be decidedly difficult precisely because they stand at the technological frontier of that sector at that particular time, and it is sometimes difficult, if not almost impossible (except for those who are considered to be “skilled in the art”), to find the appropriate terminology and fully understand the original text in all of its parts. To reassure any translator who may want to give patent translation a try, let me point out that such challenging cases are few and far between, and that it is always possible, in the face of a serious impasse during a translation, to seek the advice of a real “skilled in the art” person, in other words, a technician who will certainly be able to clear doubts and suggest the correct terminology. Over time and with daily practice on these texts, it will be possible to realize that, despite the wide range of subjects, patent texts, no matter how seemingly disparate, share the same similarities, at a minimum, as it pertains to how they should be approached for translation.

Patent translation: controlled language and automation

It is impossible to know everything about anything. No one can have a foundation in Computer Science, Biochemistry, Organic Chemistry, Mechanics, and Electronics that is so extensive that it is always effortless to translate any text dealing with any topic. You cannot humanly expect from a translator to be innately able to manage a rich variety of lexicons, terminological worlds, and expressive ways so far apart from each other. However, there is a translation sector in which the translator is required to have familiarity with an array of sectors in virtually any technological and scientific field having a potential industrial impact. This is, indeed, the world of patent translation. Some indulge in high-sounding expressions, such as “the art of translating patents” or “the craft of patent translation”, but emphasis and rhetoric aside, it is undeniable that the translation of patents is an often difficult task, sometimes at the very limit of one’s translation skills. This is because the technology that is the subject of the patent is itself at the frontier of technology; hence, it is difficult to find reliable terminological sources to draw on to translate a patent. Luckily, such cases are few and far between, and the use of termbases, of validated translation memories, and of machine translation (we will discuss it more in detail below) is most of the time more than enough to complete a flawless patent translation from any point of view.

A patent text, if written correctly, is a classic example of controlled natural language. Controlled language has the aim of reducing the ambiguity and complexity of a text, simplifying grammar, and consolidating its lexicon as far as possible. Controlled language encodes a set of rules that must help the writer who is processing the text in terms of syntax, semantics, and structure. Therefore, the purpose of controlled language is to improve:

  • legibility and clarity: the structural and lexical ambiguities in the text are considerably reduced by establishing rules that already limit the possibility of incurring in syntactic, semantic, or pragmatic ambiguities at the time that the document is drawn up;
  • editability: any text that is easier to read and understand is also easier to update because the structure of the document can be more easily recognized.

It is easy to understand that such a text, by virtue of its extremely simplified, unambiguous, and often redundant structure becomes an ideal text to be processed automatically. And indeed it is: for example, at times, the use of CAT tools and termbases significantly increases the speed of the translation process, thanks to leverageable matches and fuzzy matches within the text and those stored in the translation memory that have been harvested from segments of other translations. In fact, it often happens that the patent texts of the same inventor or company have similar or sometimes equal segments, and this is an obvious advantage in terms of speed as well as to maintain intratextual and intertextual terminological consistency. Over time, it will be possible to accumulate large and sector-specific translation memories (for example, memories for biotech, mechanic, construction, chemistry sectors, etc.), and those memories can be leveraged for the construction (training) of automatic translation engines. 

The patent translation market

In Europe, the patent market – and by extension patent translations – has been growing in the last five years: in 2018 alone, 147,317 patent applications were submitted to the European Patent Office. And Italy, with 4,399 applications and a growth of 0.9% compared to 2017, ranks sixth for the number of requests among the EPO Member States.

Trend of patent applications filed in the last 5 years and the distribution by geographical area

Data relating to the origin of patents

We do not know if the current pandemic crisis will have a negative effect on global validation processes: from our vantage point at ASTW (we translate approximately 2 million patent words every month), we do not see any decrease; indeed, we have recorded an appreciable increase in requests from our clients. It is reasonable to expect an increase in patents and related validations (and, therefore, translations) in discrete sectors, such as medical and biotech, as well as sectors relating to cellular communications and information technology, and perhaps a decrease of requests in other industries.

The most active technical sectors in recent years are those that involve new technologies, and, namely biotechnology, digital communication, information technology, energy, and transport, and the largest presence on the market is registered by large multinationals (which represent a 71% market share), with Siemens, Huawei, and Samsung being the predominant applicants in 2018.

A significant figure emerges from a large representative sample of patent applications submitted to the EPO: in 2018, one in five applications were submitted by SMEs or individual inventors. 

Data released by the EPO in the 2018 Annual Report

In this scenario, and in light of current relevant legislation, to protect the intellectual property of an invention in industrialized countries, an applicant must involve all of the countries in which they would like to obtain a patent. Therefore, it is easy to understand how translation costs are a determining factor in the validation procedure. It often happens that the costs to be incurred for the translation of a patent exceed the patent filing costs, and it is for this reason that the fees generally paid to LSPs and translators are relatively low. However, by leveraging CAT tools and machine translation, an expert patent translator can make patent translation profitable and professionally lucrative and fulfilling.

Quality is what matters

At ASTW, we handle around 2 million patent words each month, which is an appreciable amount for a relatively small LSP like ASTW. To ensure the quality of each patent translation, over the years we have established a workflow that has enabled us to be confident about the quality of our patent translations. 

Immagine che contiene screenshot

Descrizione generata automaticamente

Sample workflow

For the purpose of increasing the productivity of translators, we have long since included machine translation in our workflow. All of our translators and patent reviewers, both freelance and in-house, are essentially post-editors. A critical factor in the quality of the output of the automatic translation engines was to identify a quality MT solution that could be trained with our patent memories and learn the terminology and style of each of our translations. The solution that we chose was ModernMT ( ModernMT, which was developed by the Italian company Translated (, leverages Neural Machine Translation and relies on the context of the entire document in order to predict and provide a better translation of every sentence. In addition, the model learns in real time from each correction made by translators. We determined that, compared to other MT solutions, ModernMT offered us better quality (less post-editing effort by post-editors), with a consequent increase in the productivity and quality of our linguists’ work. Under normal conditions, for example, an expert linguist can translate up to 6,000 or even 7,000 words per day (if not more, for example, in the case of structurally simple and repetitive texts), with the use of a CAT tool and working as a post-editor on an output of an automatic translation machine trained with patent data.

The importance of linguist training

To be a good patent translator, however, it is not enough to just use technology, automatic translation, and CAT tools. Indeed, these tools must be handled by an expert hand. The training of patent translators is, therefore, critically important. For this reason, at ASTW we provide our linguists with revolving training, consisting of courses and continuous feedback on their work from more experienced colleagues. In fact, we have shown that experience in the field together with the use of validated terminological sources (and also of automatic translation) greatly shortens the learning curve of novice patent translators: on average, after six months of full-time patent translation, we can be quite confident that a budding patent translator has mastered the basics of translating patents well. It is still necessary to shadow in-training translators with a more experienced linguist for at least six more months, but, generally, the post-work feedback is good because translators have many resources at their disposal in case of doubts. Moreover, from our experience, the assist offered by machine translation is significant, especially if trained with its own patent data: the MT output will not only offer terminology that is probably appropriate, but the translation style is also generally adopted and validated in-house for the translation of these texts. Furthermore, the personal help provided by colleagues and teamwork are very important for the professional growth of patent translators: creating a positive and collaborative business environment is, therefore, of the essence, although this precondition is sometimes overlooked in many companies, where competitive behaviors prevail to the detriment of productive teamwork. In the patent translation sector, the know-how harvested personally and at the company level are the core condition upon which to build the appropriate workflow to guarantee our customers quality, speed, and capacity.


Domenico Lombardini (BS in Biological Sciences) is ASTW’s Founder and CEO ( ASTW is an Italian Language Service Provider specialized in the intellectual property (patents and trademarks), legal, medical and scientific sectors.


English translation and adaptation by Raffaella Luciani