What are the disadvantages of Machine Translation and why should I manage the users expectations to avoid and eliminate disappointments?

1. Quality of language pairs is not the same from each of the vendors in terms of speed, engine accuracy, dictionary development and integration of MT into front-end applications.

2. It will never produce perfect quality translations but as a tool for getting the gist, or generating the first draft of a translation, it is excellent.

3. Although machine translations do not select the right connotation for each context, unless one first recognizes the context and incorporates the correct language pair dictionary for each subject domain (online vertical market lexicon). So machine translation is not very useful for any text in which there are more than 3 topics but is ideal for text in which there is only one, two or three topics since dictionaries can the be customized for translation of dynamic content in this topical area.

4. If one runs a translation job through a MT engine and then reverses the translation by sending it through another directional language pair's machine translation engine back in order to translate it again back into the source language, there could be a compounding of an error rate which will make translations sound too "literal" and therefore more awkward than a human translation, and sometimes quite silly, although still useful for getting the gist.

5. English is still the hub language for most machine translation engines, although by 2005, there will be as many cross language pairs (French-German, German to Spanish, Japanese to Korean) as translation engines in which English is the common language.

6. Some language pairs are still unidirectional although by early 2004, most were bi-directional (e.g. English to Farsi and Farsi to English).

7. A human translator is still needed for editing the machine translation draft.