Susan Harkus, Independent Consultant

Tomoko Gondow, Business Consultant & Lecturer in English, Seijo University, Japan

In 2000, we needed to reduce localization time

In 2000, we were developing a product for several European markets, and we knew that we would only have a short time to prepare the product for each market. Would a combination of automatic translation and human translators help us reduce localization time by reducing translation time?

I knew that the English text had a major influence on the quality of the automatic translation, so I reviewed available research and developed a set of disciplined writing guidelines.

During a period of six months, everything that I wrote conformed to the guidelines. No one noticed any difference in what I wrote – no one, not the customers, not the engineers, not the marketing team.

When the documentation was complete, I did an automatic translation of each document and assessed the quality of the output using measures of completeness, correct meaning, and syntactical accuracy. The results were encouraging.

By writing for translation and by maximizing the efficiency of translation software, we reduced the time required to produce a first foreign-language edition of the documentation by two-thirds.

In 2006, we need to increase the accessibility of web content

This article re-examines automatic translations from a new perspective. It’s no longer a question of localization but of instant access to web content.

The World Wide Web has changed the mathematics of translation. The web is an environment where millions of people are publishing millions of pages, and where millions of users are visiting millions of pages every hour.

For most web content, human translation is impractical.

  • No one has the human translation resources to translate even high-value pages into the languages of anonymous readers.
  • Translation takes time. Even when human translators use computerized tools such as Translation Memory, translation extends publication times to days rather than seconds.

Real-time translation is essential

My first experience of real-time translation was AltaVista’s Babel Fish. Babel Fish was the source of constant laughter for my development centre colleagues who used the software to produce ridiculous translations of idiomatic English expressions.

English-speaking consumers often laugh at the awkward instructions that they find with the goods they buy, but who laughs when the translation makes it impossible to configure a purchase?

By creating web pages that can be translated efficiently, everyone wins. An imperfect but acceptable translation exposes our messages to non-native speakers because any reader at all can use their preferred translation engine to access our information.

The key to the quality of real-time translation is the original text

Designers of translation software know that the quality of a translation depends on the input text as well as on the software. Their conundrum is that engineers can improve the software, but they have no control over the quality of the text. Writers produce the text.

Improve the quality of the translations by controlling how the text is written

At an early stage in the development of translation software, engineers and their research associates tried to improve the quality of the input text by proposing strict rules for writing.

For the most part, writers found their approach impracticable. Communication implies that writers write to engage their readers, and strict controls can inhibit the writer’s ability to communicate effectively.

The solution? Help the software do its job

Most people think that a good translation depends on using the correct words of the second language.

That assumption is incorrect. Vocabulary is important, and the dictionaries of most translation products can be extended by user dictionaries, but the first task of translation software is to parse the text structure.

That’s where the text really influences the result. If you write for translation, you help the translation software understand how your meaning is constructed! And there’s a bonus. Think. Don’t your readers have the same objective as the software? To find meaning in what you write? Won’t they also benefit if you write for translation?

Revise the guidelines to support Asian languages

When I decided to prepare this update, I wanted to validate my 2006 guidelines against an Asian language, such as Japanese. Fortunately, I had a wonderful collaborator in Tomoko Gondow, a good friend and a teacher and lecturer in English in Tokyo, Japan.

You don’t have to write every sentence for translation

The guidelines have a strong influence on translation quality. They express general ways of thinking and should be easy to remember.

Do you need to apply the guidelines to EVERY sentence on a web page? Not at all, unless you think that your online reader will read every word on your page!

I recommend

  • that you apply maximum effort to the scan elements of your page: the page title, the opening paragraph(s), sub-headings, and link text
  • that you check your writing style from time to time, by testing different sentence structures in an online translation engine

Guidelines – Write thoughtfully, version 2006

1. Make the structure of your sentences clear to the software.

  • Write well-constructed sentences. It is the structure of the sentence, not the sentence length that confuses the software. Make sure that you have meaningful connections between the clauses and phrases of your sentences.
  • Include definite articles, as well as indefinite articles, wherever possible.
  • Repeat structural elements. For example, “by writing for translation and by maximizing the efficiency of translation software.”
  • Repeat a noun instead of using a backward-pointing pronoun like “it,” “they,” “this,” or “these”. Software analyses each sentence discretely, so the software cannot reread the preceding sentence to establish the context of the pronoun.
  • Wherever possible, convert each list item into a phrase, sentence or clause. The software does not remember the structure that each item inherits from the introduction to the list – and many readers have the same problem!
  • Always include the relative pronouns or conjunctions. For example, “I knew that the English text had a major influence…” or “the car that he bought.”
  • Connect phrases to the nouns that they modify.
  • Use punctuation correctly. For example, separate the main clause from subordinate phrases and clauses by commas.

2. Use language and terminology thoughtfully.

  • Use words with their primary dictionary meaning. For example, the primary meaning of “meet” is “encounter,” not “satisfy.” If you use words with their primary dictionary meaning, you will use them consistently and you will also avoid homographs. You will still be able to use metaphors because metaphorical language usually uses words with their primary dictionary meaning.For example, the Japanese and French translation engines recognized the meaning of “bridge” in the sentence, “We can help bridge the differences between Asian and Australian business.”
  • Don’t use idiomatic or culture-based expressions.
  • Identify terminology that is unique to your products or services. If you are using a translation dictionary, you can include the terms in the dictionary. Otherwise, you can simply make sure that the understanding of particular sentences doesn’t depend on the exact meaning of a term.

3. Avoid abstractions. Use concrete expressions.

  • Express actions as verbs rather than nouns. For example, “they can relocate their office to Asia” rather than “through relocation.”
  • Identify the subject if there is one. Style rules advise writers to avoid passive voice. However, passive voice is essential when the doer of the action cannot be identified. The real value in active voice is that you identify the doer.
  • Wherever possible, avoid using neutral words that take their meaning from the sentence context. For example, words such as “implement,” “provide,” “address,” “leverage,” “manage,” “situation,” and “environment” may cause problems.

4. Avoid actions that are expressed as a noun + preposition.

Verb combinations, such as “set up” or “shut down,” are called phrasal verbs. Most of these verbs create problems for translation software, but they are also hard to avoid in English.

Sometimes they can be replaced by a verb with a concrete meaning. For example, “establish an office” instead of “set up an office.

Why make an effort to write for translation?

An example – but who missed the opportunity?

During one of our analysis sessions, Tomoko told me about a friend who imports photos from overseas. He does not speak or read English. He needs to translate websites related to his work because his business has overseas connections.

The translations of web-based software frustrated him. He bought his own software. Happy outcome? No. He is not using the software at all because “the translations are not grammatically correct.”

Tomoko is amused that her friend is discouraged by incorrect grammar, because he doesn’t usually read every word. He just scans the text, looking for significant technical words.

As with most translation software users, Tomoko’s friend thought that the poor translation was the result of the inadequacies of the software. He didn’t realize that the website content itself contributed to the poor translations.

However, it is irrelevant whether a web user understands why a translation is unintelligible. Web sites only achieve outcomes when users achieve their objective. Tomoko’s friend was unable to read company web pages. For him, the result was frustration; for his suppliers, the result was potentially the loss of sales.

Business has three options

By writing for translation, an organisation can harness the power of instant translations in three ways:

  • They can simply publish their pages, confident that web site visitors will use a preferred translation engine to read their web content.
  • Their websites can offer automatic translations in key customer languages.
    For example, the City of Augusta, Georgia and the Washington Metropolitan Area Transit Authority.
  • They can utilise the user dictionaries of their installed translation software to overcome terminology problems.

Put this page to the test!

If you are familiar with another language, put the URL of this page into a translation engine and test the translation of a web page that has been written for translation. Here are two useful engines – SYSTRAN and EXCITE (Japanese version).