Write for machine translation

Problems with WWW pages

You have no control over how a machine translator deals with your World Wide Web pages. This causes two types of problem:

Words are translated which should not be translated

Names of people and companies often make no sense when translated as words. This kind of problem is very common. My name, Jon Miles, is translated into French as milles de Jon, or thousands of Jon. There are ways to prevent words being translated:

Proposed solution

To avoid having to use such tricks as underlines, we need an HTML tag to tell the machine translator not to translate, such as: <notrans>Jon Miles</notrans>. This is a simple solution that would be easy to implement, both for the machine translator and for the webmaster.

Words are translated wrongly

A website about the history of railroads might have references to the gauge of a rail line, where 'gauge' means the distance between the rails, not a measuring instrument. The railroad use of the word 'gauge' is not the most common meaning of the word, so the word will be wrongly translated. 

Proposed solution

We need to have special HTML tags to allow for the translation of particular words to be specified. The contents of the tag would tell the translator that the meaning of 'gauge' was one of the meanings listed for that word in a particular online dictionary.

For instance: <dictdef=noun7>gauge</dictdef>. This code says that the meaning of 'gauge' required is the seventh noun definition in the dictionary. The html page header would specify the online dictionary to be used, such as wikitionary. If you look up the definitions of 'gauge' as a noun in wikitionary, you will find: "7. The distance between the rails of a railway". The html page header would need to specify wikitionary as the dictionary to use for that page.

Of course this is not a full solution for a machine translator, as wikitionary does not have parallel numbered definitions in different languages. This proposed solution to the mistranslation of words or phrases would require such a parallel dictionary to be set up.


Updated 2006-08-13