There are many assistants around, from Apple’s Siri and Microsoft’s Cortana to Amazon’s Alexa and Google’s latest Assistant. With other assistants catching up in advancing with algorithms, one of the differentiating factors right now is languages support.

Siri is amongst the oldest in the lineups and speaks 21 languages localised for 36 countries. Recently, Reuters reported that she will be learning Shanghainese as well, a special dialect of Wu Chinese spoken only around Shanghai.

Let’s put this in perspective. Microsoft Cortana speaks eight languages tailored for 13 countries. Google’s Assistant, recently rolled out to Android Nougat devices, speaks four. Amazon’s Alexa features only English and German.

Language is naturally difficult to program – even just in English. Local dialects and native tongues could lead to confusion. Take sports scores for example – football refers to the black white variant in Britain, while in US it is the almond shaped variant.

While Siri is well known for her quips, programming a sense of humour into the language takes local knowledge as well. For example, it takes a 29 strong editorial team to customise Cortana for local markets, such as Mexico, for distinguishing a Mexican’s way of life rather than a generically Spanish speaking country.

For Siri’s new language, Apple starts work by first working with humans to read passages in a range of accents, dialects and voices as well as sound range. This is then transcribed by hand, so the computer has an exact representation of the spoken text to learn from. Then, an acoustic model is built to attempt predicting words sequences.

To make it more accurate with less error, Apple uses a text-to-speech translator to capture a small percentage of audio recordings anonymously. These recording with background noises and mumbled words are transcribed again by humans, helping it cut error rates in half.

When enough data is collected, and a voice actor is assigned to record Siri’s voice, the assistant is then released with answers to estimates of the most common questions. Once live, Siri learns from real world usage and is updated every 2 weeks.

Google and Amazon said they plan to bring more languages to their assistants, but we have no updates on that.