As Google Translate marks 6 years since Swahili was added to its scope of languages, very little is to be desired of one of the most interacted with AI program. Its formation was revolutionary, perhaps an indication that the world had really began interacting with machines fully. We could speak the same language, and understand each other. However, long after its formation, the reality has downed on us. Perhaps we overestimated the capability of machines learning human language.
I’m among the hopefuls who have always dreamt of a world where machines can help out with anything, not to replace humans, but be indispensable companions. With the addition of Google Translate Swahili version, it’s easier to envision a tool that could not only help people who are new to language, but also come in handy for school children, especially with very few Swahili resources available online.
I believe 6 years is quite a long time to grasp the basics of a language. An average human being will take a few months to understand and speak a new language and then one or two years to grasp the accent and fluency of it. If machines are to be smarter than humans, they should at least take shorter to learn.
I have been using Google Translate and as I found out, I have been getting most of the translations wrong. Thanks to Duolingo app for Android, I can confidently create sentences in a certain foreign language. In my analysis on how Google Translate decrypts sentences based on their context, I found that there isn’t much it does. In fact, all it does is to translate any word you feed in to English, before it translates it to your preferred language.
It took quite a huge amount of time and resources to feed the data into Google’s systems. It didn’t happen the way dictionaries or the Swahili Kamusi are created (by intellectuals in the language), Google decided to take it on the street. Using their incredibly huge databases across Swahili speaking countries, mostly Kenya and Tanzania, the company was able to collect millions of words and phrases which were then checked for accuracy.
Google Translate itself is powered by a Statistical Machine Translation algorithm that was created by mathematicians as early as the 19th century and perfected over time. It was used to tackle the distribution of words in languages and the author’s writing styles based on the frequency of words and the order in which they occur.
For example, by studying the nature of three-letter-words, four-letter-words, and five-letter-words and so on, it is practical to conclude certain facts about the writer as well as the language. The frequency and distribution could also tell if the language was fluent or not. This means by using this software, we are supposed to tell whether great writers like Shakespeare, Henry Miller, and Ernest Hemmingway actually mastered grammar.
Locally, the Swahili version of Google Translate is supposed to counter our authors too. We are supposed to raise doubts or be satisfied with works from Professor Ali Mazrui, Ken Walibora, and Wallah bin Wallah, among other greats. However, it’s the other way round. Those who want to learn are questioning Google Translates capability to help.
In my use of Google Translate to decode Swahili words, I got some weird expressions in English. Some were funny, but when the situation can’t allow humor, you may get angry at the first intelligent machine you are interacting with. It doesn’t help though coz it doesn’t have feelings.
First, I examined single Swahili word –it’s supposed to be accurate at this than phrases –and although it got most, there are still others that give it quite a hard time. Here is a screenshot of one of the words I tested.
For single words like bimdogo (young lady), machachari (excitement), or vibibi (pancakes), and many others, you will either get the same word on the English side or a misleading translation. When it comes to verbs, things get a little hectic for Google Translate Swahili to get the English version of the words.
It gets worse at phrases and even worse at sentences
(Swahili text sourced from a local daily)
If you didn’t know Swahili, you’d be utterly confused at what these words mean. Most of the times, they wouldn’t even make sense. I also used another approach to translate words to different languages before translating them back to English. I understand English is a medium that translate uses and if it got the context right, nothing should be lost. Instead, not a single phrase came back to its original state.
One of the developers of Statistical Machine Translation which powers Google Translate has long defended the approach to use single words that are combined mathematically during translation, but this approach fails to consider the context. It might be a very long time before its capable translating fluently. This means, the next breakthrough in Artificial Intelligence might not occur during our lifetime.
You can help Google Translate learn Swahili
You can help google translate learn Swahili faster in a few simple steps. Visit this link and set your languages to Swahili and the translation to English, type in a word and if google translate gives a wrong Swahili version, click on suggest an edit and give the accurate version.
Share your suggestions, comments, and opinions. I’d love to hear from you.