Google Translate vs. Bing Translator, Part 1: Which Produces Better Patent Machine Translations?

[tweetmeme source=”Intellogist” only_single=false]  Sonja Olson,  Landon IP’s Director of Translation Services, summed up the main problem with machine translations for patent documents in a previous blog post:

Machine translation will get the gist of the document, but it will lose the nuance.  Machine translation can plug words together, but it can’t understand the sentence as a whole.

Sonja suggested that “if you need to know if a document contains information on a topic, you can go with a machine translation.  If you need to know how that topic relates to the document overall, then go with a professional translation.”

So, users should be forewarned, neither of the free translation services discussed in this post should be used as a substitute for professional human translations.  As you’ll see in the following tests, neither translations (from German to English and Korean to English) for Google Translate nor Bing Translator are entirely coherent.  Although the user can get an overall idea of the content of the documents from these machine translations, both Google and Bing translations would never be acceptable in any legal or official context.

Google Translate and Bing Translator (powered by Microsoft Translator) are useful, however, for evaluating the general content of a patent document during a prior art search.  Which free service produces a better machine translation of a patent document?  Continue reading as we compare the two services!

Comparison of Languages and Interfaces

First, let’s look at the easiest way to test the quality of machine translation service: the number of language options.  According to “Inside Google Translate“, the service supports 58 languages (7 of which are “alpha languages”, which “may have less-reliable translation quality” than the other supported languages).  Bing Translator only supports 37 languages, according to its Help section.

While Google Translate clearly has a larger selection of supported languages than Bing Translator, there are only minor differences between the interfaces of the two systems.  On the homepage of both translation services, the user can select the original language and target language of the text or webpage from drop down menus at the top of the page.  The user can then paste the text or URL for the web page into the first of two adjacent text boxes and select to translate the text/webpage.  If text is entered in the first text box for either service, a translated version of the text will appear in the second text box.  In Google, the user can mouse over individual terms in the translated text to view alternative translations (this feature isn’t available in Bing).

The Google Translate interface (Korean to English) with the alternate translation function.

The Bing Translator interface (Korean to English).

If a user pastes a link in the first text box in Google, another link will appear in the second text box that links to the translated web page, with a Google Translate toolbar situated above the webpage.  Mousing over the translated text of the webpage will display the original language text in a pop-up window. The user can enter a new URL through the toolbar, change the original and target languages, and select to view the translation or original webpage.

If the user enter a URL in Bing Translator, the service will automatically generate a translated version of the web page within the “Billingual Viewer”, which provides the following features:

• Options to enter a new web page URL or select a new translation language.
• Four bilingual layouts: “side by side”, “top/bottom”, “original with hover translation”, and “translation with hover original.”
• View the original web page by selecting “Go to original page.”

Both Google Translate and Bing Translator can automatically detect the language of text or a web page, and a “Detect language”/”Auto-Detect” option is available in the drop-down list of original languages on both services.

 

Testing Translation of Abstracts

Now that we have an overview of the language options and interface functions for both systems, let’s try testing the translation of actual patent document abstracts.  The following abstracts were first located in their original languages through various national authority or subscription databases.  The original-language abstract is displayed below, followed by two versions of the abstract translated to English (one by Google Translate and one by Bing Translator).  Finally, to highlight the differences between the two translations, a text comparison feature from the file history tool Patent Workbench® was used to compare the translations.  Text that is in the Google translation but absent from the Bing translation is highlighted in strike-through red, text that is unique to the Bing translation is highlighted in underlined green, and text common to both translations is displayed as normal black text.

Example 1: German Abstract

Original Abstract (Found on DEPATISnet – DE102009045236A1):  Die Erfindung betrifft ein Fahrzeug (1), insbesondere Kraftfahrzeug, mit einem elektrischen Energiespeicher (5), einer elektrischen Maschine (10) und einem Antriebsstrang (11), wobei der elektrische Energiespeicher (5) mit der elektrischen Maschine (10) elektrisch und die elektrische Maschine (10) mit dem Antriebsstrang (11) mechanisch verbindbar ist. Die Erfindung sieht vor, dass das Fahrzeug (1) einen pneumatischen Energiespeicher (3) aufweist, der mit dem Antriebsstrang (11) und/oder mit dem elektrischen Energiespeicher (5) in Wirkverbindung bringbar ist. Die Erfindung betrifft weiterhin ein Verfahren zum Betreiben eines Fahrzeugs (1).

Google Translation: The invention relates to a vehicle (1), particularly motor vehicle with an electrical energy storage (5), an electric machine (10) and a drive train (11), wherein the electrical energy storage device (5) with the electrical machine (10) electrically and electric machine (10) with the drive train (11) is mechanically connected. The invention provides that the vehicle (1) a pneumatic energy storage (3) which, with the drive train (11) and / or with the electric energy storage (5) is brought into operative connection. The invention further relates to a method for operating a vehicle (1).

Bing Translation: The invention relates to a vehicle (1), in particular motor vehicles, with an electrical energy storage (5), an electrical machine (10), and a drive train (11), where electrical energy storage (5) with the electrical machine (10) electric and the electric motor (10) with the drive train (11) is mechanically interconnected. The invention provides for that the vehicle has pneumatic energy storage (3) (1), that is items with the drive train (11) and/or with the electric energy storage (5) in active connection.The invention concerns a method for operating a vehicle (1) continue.

Comparing Google (red text) and Bing (green text) Translations through Patent Workbench® Text Comparison:

The invention relates to a vehicle (1), [particularly]  in particular motor [vehicle]  vehicles, with an electrical energy storage (5), an [electric]  electrical machine 10 [)]  ), and a drive train (11), [wherein the]  where electrical energy storage [device] (5) with the electrical machine (10) [electrically]  electric and  the electric [machine]  motor (10) with the drive train (11) is mechanically [connected.]  interconnected. The invention provides  for that the vehicle [(1) a]  has pneumatic energy storage (3) [which,]  (1), that is items with the drive train (11) [and / or]  and/or with the electric energy storage (5) [is brought into operative connection. The]  in active connection.The invention [further relates to]  concerns a method for operating a vehicle 1 [).]  ) continue.

Example 2: Korean Abstract

Original Abstract (Found on KIPRIS – App no#: 10-2008-7001230 (20080116), Unexamined Pub #: 1020080065577 (20080714) ) : 의료적 치료를 위해 경두개 또는 내두개에 전기적 에너지를 최적으로 제공하는 방법이 결정된다. 환자의 뇌에 대해 MRI나 CAT 스캔 데이터, 또는 이들 모두의 데이터를 획득한다. 획득한 데이터에 기초해서 환자의 뇌의 일부 부위에 상이한 전기적 저항값을 할당한다. 전극 위치를 선택한다. 이러한 할당과 선택에 기초해서, 경두개 또는 내두개에 치료를 위한 최적의 전기를 제공하기 위한 하나 이상의 인가된 전기적 입력을 산출한다.

Google Translation: For medical treatment or naedugaee transcranial electrical energy to provide the optimum is determined. About the patient’s brain MRI or CAT scan data, or all of these data are obtained. On the basis of data obtained in the patient’s brain, some areas are assigned a different electrical resistance. Select a location to the electrode. Based on these assignments and selections, the best treatment for transcranial or naedugaee to provide electricity to more than one licensed electrical input is calculated.

Bing Translation: Medical treatment for two or my two electrical energy to determine how best to deliver. The patient’s brain MRI or CAT scan data, or for all of the data acquisition. Acquired data are part of the patient’s brain based on different parts of the electrical resistance. Electrode position is selected.Select these two assignments, based on my two best electric for treatment or for the provision of one or more licensed electrical input and output.

Comparing Google (red text) and Bing (green text) Translations through Patent Workbench® Text Comparison:

[For medical]  Medical treatment  for two or [naedugaee transcranial]  my two electrical energy to [provide the optimum is determined. About the]  determine how best to deliver. The patient’s brain MRI or CAT scan data, or  for all of [these]  the data acquisition. Acquired data are [obtained. On the basis]  part of [data obtained in] the patient’s brain [, some areas are assigned a]  based on different  parts of the electrical resistance. [Select a location to the electrode. Based]  Electrode position is selected.Select these two assignments, based on [these assignments and selections, the]  my two best  electric for treatment  or for [transcranial]  the provision of one or [naedugaee to provide electricity to] more [than one] licensed electrical input [is calculated.]  and output.

Japanese and Chinese Abstract Examples to Come…

Both Google and and Bing translations for the German abstract have more words in common, as illustrated by the Patent Workbench® Text Comparison.  The German translations for both services also appear to be more coherent overall than the Korean translations.  It’s possible that both services are more adept at translation of  Latin-based text than non-Latin text.  The translation of non-Latin text by both services will be further tested in the next blog post, when we look at examples of Japanese and Chinese abstracts translated by both Bing and Google.

Google has a longer list of languages which the user can translate, and although both services have similar interfaces, Google Translate has the additional useful feature of viewing possible alternative translation terms.  I’ll wait to make any judgement on the overall quality of the actual translations until we see a few more example translations.  You can be the judge, though….which translation service works better for you when translating patent documents during prior art searches?  Are the above examples good representations of the quality of machine translations produced by each service?  Let us know in the comments!

If you’re interested in downloading the Patent Workbench® Reader for free and testing the text comparison tool (as well as many other features), visit the Patent Workbench® website to request a free trial.

This post was contributed by Joelle Mornini. The Intellogist blog is provided for free by Intellogist’s parent company Landon IP, a major provider of patent searches, trademark searches, technical translations, and information retrieval services.