The Ropes

A Few Postediting Tips

By Barbara McClintock, Certified Translator

According to the philosophy of “if you can’t beat ’em, join ’em,” language professionals are starting to test various automatic translation software programs, also referred to as machine translation. Why use machine translation? For a translator, the main reason is to save time, but, if it is not properly revised, the output could lead to comical or even disastrous results. The pre-translation may be described as a sketch or unfinished painting. It needs an artist’s touch to finish it. The purpose of this article is to discuss some common-sense tips for postediting using MT.

Prepare your document

You first need to prepare your source document. As MT programs can usually handle only a limited number of words at a time, you may need to separate your source document into paragraphs or chunks if the paragraphs are very long. Open a draft version of your document to do this and to cut out any sensitive information in the original. Delete or change proper names if they could be compromised. You can then copy the MT version into this draft. Only use a safe MT system, not a public system. Be careful not to put confidential information in a program that may copy and store it in an undisclosed location.

Postediting stage

Once you have the machine translation output, it is time to revise or postedit it. Two screens are helpful for seeing both the French (or other source language) and the MT output. Ideally, you should type your own translation on top of the French text, rather than copying and pasting the MT version into it, mainly because of the hidden codes. It is also advisable to do this in order to distance yourself from the MT version while editing it. You can use the MT version as a terminology reference while you are translating your first draft and then do searches and replacements in your translation afterward.

Often the MT looks good on the surface, but, when you review it in depth, it may have changed meanings and terms. I have noticed that the MT may even give you the opposite meaning, and it can change species names for other species.

The following is a sample source text:

Original source: Gérer le programme qui régit la façon dont l’ensemble des activités liées à la Loi sont planifiées, conçues, mises en œuvre et évaluées, processus au terme duquel sont prises les décisions finales concernant l’application de la loi et la divulgation publique d’information.

MT: “Manage the program that governs how the range of activities related to the Act are planned, designed, implemented, and evaluated, processes, at the end of which are made the final decisions concerning the application of the act and the public disclosure of information.”

Analysis
French word order is often the reverse of English and is not necessarily corrected by MT. For example, processus au terme duquel sont prises les décisions finales is machine-translated as “processes at the end of which are made the final decisions.”

Knowing that the sample is from a job description, the translation could be edited as follows:

Postedited: Manages the program that governs how all of the activities related to the Act are planned, designed, implemented and evaluated—a process at the end of which the final decisions concerning the application of the Act and the public disclosure of information are made.

Another sample from the job description:

Original source: Planifier, concevoir, élaborer, mettre en œuvre et évaluer les stratégies, politiques, outils et autres moyens connexes, dont la formation et la sensibilisation, concernant le programme.

MT: “Planning, design, develop, implement and evaluate strategies, policies, tools and other related means, including the training and awareness, concerning the program.”

Analysis
The MT version is often too literal. We do not know what autres moyens connexes refers to, which may call for a “fudge” or a switched word order as in this case. Here you need to adapt the verbs for the style of an English job description and ensure parallelism.

Postedited: Plans, designs, develops, implements and evaluates strategies, policies and other related means and tools, including training and awareness-raising related to the program.

Words have nuances just as colours do. MT can be useful for understanding purposes or to prepare a first draft, providing a canvas which, ideally, a language professional should work with, touching it up or repainting it, where necessary.

From statistical machine translation to neural machine translation (NMT)


Google Translate was released in 2006, so we have had statistical machine translation for over 10 years. However, scientists have recently made breakthroughs with a technology referred to as neural machine translation (NMT) using artificial intelligence. The term AI was actually coined in 1956 according to a highly informative New York Times Magazine article entitled “The Great AI Awakening, How Google Used Artificial Intelligence to Transform Google Translate, One of Its More Popular Services — and How Machine Learning Is Poised to Reinvent Computing Itself.”1

Machine translation research has necessarily raised questions about the nature of translation. Scientists initially thought that they could collect the data, run some tests and voilà. Well, it did not quite work out that way. Humans have to think to translate. NMT tries to mimic the human brain. AI research is now being conducted into how to design machines to learn other applications than translation (deep learning).

. “The Great AI Awakening” by Gideon Lewis-Kraus, The NY Times Magazine,

This article solely reflects the author’s opinions.