\\ home
\\ resources
\\ events
\\ about

Interview by Susan Harkus

The paper describes the translation outcomes that Susan achieved through disciplined writing which is a "writing for translation" approach that applies the good writing practices readily acknowledged by most technical communicators, and mixes in some simple rules that optimise source for translation.

Andres Heuberger is the President of ForeignExchange Translations, Inc. He is an active participant in industry discussions about machine translation and translation-memory software. Susan wrote to Andres to ask him to assess when machine translation becomes a viable option for delivering commercial translations.

His responses to the following questions provide valuable insights into the commercial application of machine translation (MT).

Q1. Does ForeignExchange use MT?

ForeignExchange does not currently use machine translation. The reason for this is lack of client demand. In our area of expertise (medical devices, financial services), most companies are too concerned with compliance issues to consider using machine translation.

However, in a "previous life" at another translation company, we developed our own, very basic machine translation tool. It consisted of:

a) a source-text evaluator
b) a string lookup table
c) some rudimentary MT logic

The project involved the one-time translation of several million words stored in a proprietary format. The text didn't include complete sentences but was a collection of product features and descriptions.

At the time, we decided to develop our own MT mechanism because the cost of commercial systems was prohibitive and the quality too low. For approximately US$5,000, we had a programmer code a system to our specifications (probably in less time than it would have taken to purchase, install, and train for a commercial system).

Q2. How do you identify an opportunity for MT?

Clearly, MT requires a significant investment in equipment, employee training, source writing, and the MT system itself.

Criteria for the successful deployment of MT fall into two categories: financial considerations and liability considerations.

Financial considerations

• large volume of source text available electronically
• ongoing need (in this case, weekly updates)
• availability of MT with appropriate source/target language combination
• clearly defined domain/terminology base
• appropriately written source text

Liability considerations

• no compliance aspect to translation
• client is aware of MT use

At ForeignExchange, we would not hesitate to use MT for a project that fitted these criteria.

The liability items in the above list may deserve some explanation. MT has benefited from significant advances recently. However, it is still limited and the decision to use the technology must involve a risk/reward analysis.

We would not consider employing MT if our client was not aware of its use. Likewise, we would not use MT to do a first-draft translation for compliance-sensitive projects, such as legal, financial and medical documents.

The "quality" of the source text is an issue only as far as its consistency and ambiguity are concerned. Efforts such as Controlled English et alia have shown that MT quality improves significantly if the source text is controlled and limited.

Q3. What is the impact of MT on turnaround times?

In my opinion, the main reason to employ MT is to speed up translation processing and delivery. My exposure has been too limited to generalize on the extent of time savings. However, the project mentioned in my answer to question 1 above could not have been done in the time frame stipulated by using human translators. On that project, I would guess that MT cut the total turnaround time in half (this is including the time needed to program the MT tools themselves).

Q4. What improvements have you noticed in MT products?

My limited exposure to commercial MT systems prevents me from basing my opinion on extensive experience of the tools. However, I do believe that, yes, MT quality has improved drastically.

I believe that "MT quality" really needs to be measured as two, not always related measures:

1. Linguistic quality (actual quality of MT output)

Linguistic quality is hard to generalize and I don't think that this should be rated on a scale from "bad" to "good." Rather, the key question is: Is the output appropriate? Technology (hardware and software) advances over the past decade have made it much more possible for MT output to be of appropriate quality.

2. Perceived quality

At least as important, is the quality as it is perceived by practitioners and lay persons. For many years, everybody inside and outside the translation industry snickered whenever the words "machine translation" were uttered. Some translation software companies didn't help. Their systems were laughable, their presentation poor, and their high management turnover made them a running a joke.

Thanks to the technology advances, there have been more and more successful MT deployments. Today there are more people that talk to their colleagues about their successful experiences and that is changing the public perception of MT.

In my opinion, MT-over-the-Internet is hurting the cause of machine translation. Because users have no understanding of the relationship between the quality of the source text and the rendered translation, many sites return poor quality translations and they risk setting back public perception by 10 years.

Q5. How do you measure the translation quality of MTs?

Measuring translation quality is a tricky business. In my opinion, there is no single correct way to do this. Rather, the importance lies in determining a set of metrics, tracking them, and using them to improve quality over the long term. The use of MT would impact some of the quality system's components but the overall system would remain unchanged.

Many companies, governments, and organizations have developed quality standards. For instance, the SAE supports initiative J2450 that grapples with this exact issue. Because many automotive manufacturers use MT, their quality standard assumes MT involvement.

The German government last year published the DIN-2345 standard, attempting to set standards for translations of print documents.

At ForeignExchange, we have developed our own definition of quality. For us, quality is in the eye of the beholder and as a result, our quality system is client-driven. We measure client changes during reviews and after final deliverables as a percentage of words/pages/documents translated.

18 March 2000

Susan Harkus, Information Manager of the Australian software development company, Cards Etc, has developed a disciplined writing solution that makes translation software a viable part of localising the first software release. Susan is presenting a paper about localisation and translation at the Australasian Online Documentation Conference being held in Brisbane, Australia, 12-14 April, 2000. Susan can be reached at susanharkus@telstra.com.

back to top