Archive for category Semantics

Google Translate: Friend or Foe?

I was recently asked by a researcher if I could comment on the Google Translate software for te reo Māori. The following was my short response.

The examples 1-3 below are from English to Māori, which I generated using the Google Translation service and provide some commentary about each one in the hope that it may be of use to you.

1. My name is Michael. he Michael toku ingoa. Ko toku ingoa Michael.

In #1, the software recognizes all the single words in the English (my, name, is, Michael) but returns an incorrect particle ‘he’ on the first Māori sample, which is given in lower case where writing conventions say that a sentence always begins with a capital. The second sample forces Māori to conform to an English word order ‘ko tōku ingoa Michael = is my name Michael’ however, ‘ko’ is in the right location, but as it translates to ‘is’ it can produce a question in the English translation rather than a statement but doesn’t make sense in the Māori. It doesn’t place the diacritic ‘macron’ above the appropriate vowel ‘ō’ as in ‘tōku’. The proper response should be ‘Ko Michael tōku ingoa.’ Notice that ‘ko’ replaces ‘he’; which is orthographic standards and should begin the sentence. The software needs to identify correct particles and word order and how they string together to form the appropriate translation.

2. I am from New Zealand. He uri hou au, ko ahau i zealand hou

In #2, it is confused altogether and though the first sample in Māori is grammatically and descriptively ok, it is not the correct translation in the English, ‘he uri hou au’ means ‘I am a new descendant’ but I wouldn’t say this in Māori as it sounds unnatural. You can also see that when its translated back into English ‘I am a new descendant’ semantically, it wouldn’t make sense and is not something you would hear in everyday conversation. This, in turn, will cause bi-directional translation problems, that is, translating from English to Māori and vice versa. If I were to translate the second Māori example ‘ko ahau i zealand hou’ it would say ‘I am zealand new’ and this doesn’t make sense both grammatically nor semantically. The particle ‘ko’ is incorrect and must be replaced with ‘nō = from’ (micronized). Although it’s picked up on the word ‘hou = new’ being an adjective, it’s using it in the wrong way in this context and assumes it to be the word ‘new’ as in New Zealand. It is trying to string together individual words to construct a translation as, in this case, ‘hou = new’. You have to treat NZ as a name and translate that as a whole, and so we can use a transliteration ‘Niu Tīrene’ where you would now have a correct translation ‘Nō Niu Tīrene au = I am from New Zealand.’ It seems the software is encountering issues associated with translating texts within the context in which they appear and writing algorithms to accommodate this will be a monumental effort.

3. I am a student. He akonga au, Ko ahau te ākonga

In #3, the first translation is right, however, ‘akonga = student’ is more appropriately written with a macron on the vowel ‘ā’ as in ‘ākonga as shown correctly in the second sample generated by Google Translate. It’s important to use macrons in written Māori to show appropriate pronunciation of the word in question. Let me demonstrate ‘mata = face’, and ‘matā = bullet.’ Telling the programme – to macron or not to macron – will be a challenge. The second sample, ‘I am the student = ko ahau te ākonga’ is grammatically ok however, the English uses ‘a’ as in ‘a student’ but translates to ‘te = the’ as in ‘the student in the Māori and depending on what one is trying to say it is a matter of interpretation in context. Making distinctions between ‘a’ and ‘the’ is useful to facilitate a proper interpretation. However. I wouldn’t say ‘I am the student’ in English nor Māori. This is a matter of pragmatics, that is, what does one mean by what one says.


Google translation (GT) can handle simple phrases but can get lost in translation with longer sentences. It has a tendency to force Māori to conform to English word orders. Bi-directional translations can produce nonsensical results. It uses particles incorrectly and confuses macron placements. Context and pragmatics must be considered to determine as near as possible natural translations in the target language and how this handle single sentences that can be translated in many different ways. Anyone can suggest a translation or even edit an existing one from the drop-down list, and this can make the programme vulnerable and open to error. GT has potential, though I don’t use it.

Leave a comment