Translating between two languages is work that was previously only done by humans. These days, this can be done by a machine, a computer or program, which has the ability to perform translation, machine translation.

I expect that many of you have had the chance to use programs such as Google Translate which have been created for this task. When I was a French student, I relied a lot on Google Translate, especially in my homework that needed me to write or to read French stories or in speaking exercises with my classmates. When reading, whenever I encountered a word that I did not recognise, it was easy to type it on the web platform and get a translation. And when it came to speaking, it was easier for me to think in English then translate to French, so when I wanted to use a word whose translation I did not know, I would also type it up and get what I needed.

This ease of translation is not equal among the languages of the world. Particularly when we get to African languages, there is a lot that is missing. Recently, I have had a desire to advance my proficiency in my mother tongue, which is one of the Luhya languages. I am saddened by the lack of tools that would make my efforts easier, as in the French scenario.

Kiswahili is one of thirteen African languages available on Google Translate. It is a language with many speakers, approximately between one hundred million and one hundred and fifty million. It is this state that gives the language an advantage on digital platforms as technology companies focus on and build software that they believe will be used by many people. Beyond the number of speakers, we find that Kiswahili is better resourced than other languages, particularly when it comes to digital tools. We also find that the performance of these tools is better for Kiswahili, as with other languages that have more resources. This status has contributed towards the focus on Kiswahili in language technology and in the research to advance these technologies.

It is important to understand that the existence of these translation software does not mean that they cannot make mistakes. Tasks that require the ability to think, meaning those that are done by human beings, are hard to ape. Computer Science experts working in topics involving the replication of human intelligence, a field known as Artificial Intelligence, make advances everyday towards this ability to ape human intelligence but the technology has not advanced to a point of being able to think like humans or to replace us.

In the process of preparing articles for this blog, which we are making available in Kiswahili and also in English, I have found myself doing translations here and there. This gave me the desire to know how translations that I would get from Google Translate would compare to translations that I would do myself. Here is an article that I translated, “Powering Local Innovation in the Global South: Prospects and a Way Forward”.

I found some mistakes here and there which signal to what the translation software does not understand well.

The first is in word forms(yaani ngeli). I have found that Google Translate is not sure of or does not understand word forms and the way they are supposed to align with different words in use of the language.

Example

Sentence in English: This initiative aims to support local innovation approaches that are empowering and inclusive of African stakeholders.

Translation by Google Translate: Mpango huu unalenga kusaidia mbinu za uvumbuzi za ndani ambazo zinawezesha na kujumuisha wadau wa Kiafrika.

Human Translation: Mpango huu unakusudia kusaidia mbinu mpya za uvumbuzi wa kiasilia zinazowezesha na kuhusisha wadau wa Kiafrika.

This example shows that the software has failed to change the word form so that it aligns with the word ‘innovation’(uvumbuzi). The creation of these programs is reliant on language datasets which are available online and are composed of sentences sourced from interactions of Kiswahili speakers on online platforms. It is possible that many of us do not pay attention to these word forms in our use of the language.

The second is getting mixed up and losing the overall meaning particularly when the sentence being translated is quite long.

Example

Sentence in English: How race, nationality, and class biases limit the opportunities available to African innovators by foreign companies and investors.

Translation by Google Translate: Jinsi ubaguzi wa rangi, utaifa na tabaka unavyopunguza fursa zinazopatikana kwa wavumbuzi wa Kiafrika na makampuni na wawekezaji wa kigeni.

Human Translation: Jinsi upendeleo wa mbari, utaifa na wa kitabaka unavyowekea mipaka fursa zinazoletwa na makampuni na wawekezaji wa kigeni kwa wavumbuzi wa Kiafrika.

This example demonstrates how this sentence, which is long and complex, completely loses its meaning due to the use of many English conjunctions which require attentiveness in translation so as to retain the meaning.

It is clear that translations can differ by choosing between or among synonyms, which are words that have the same meaning, and this is another area that requires special attention so as to retain meaning.

In conclusion, I would like to emphasize the idea that these technologies should not be viewed as intended to replace humans because they cannot. In this context of translation, these technologies can ease the work of experts working in translation and advance their efforts, as well as ours as speakers of this language who would like to see it progress.