Ko tōku reo taku ohooho, ko tōku reo taku māpihi maurea

by Kathryn Tyree

You might have met Douglas Bagnall, perhaps at NDF in 2014 or 2015. You might have stopped him in the halls because he is passionate about some of the same things as you.

Douglas wants to write software to create language models for Te Reo Māori and NZ English, so we can teach machines how to do voice and character recognition. Better.

Like you, he knows that in Aoteaora we’re not the same as everyone else, and we can’t expect to capture what is special about us, using the same language models as Australia, the US and the UK, or anywhere else for that matter.

At NDF2015 lots of people and organisations showed interest in building these models. Douglas has been testing some ideas, and some organisations are trying to get funding to start work. But it’s a big job, and New Zealand is too small to reinvent the wheel on this one, and still expect to get the best solution in the end. We need lots of content to teach the machines how to listen to us, and how to read us. It needs to use an open standard so we can all share and contribute content. Then we need to funding to pay for the development, and to publish it with an open license, so no one vendor can get control of all the hard work we put in.

Is this is a problem you are interested in? Is your organisation already doing some work in this area? We would love to hear from you.

If you’re at NDF and you’d like to talk to Douglas, he’ll be at NDF 2016 on Wednesday, and at the Catalyst exhibition stand during the breaks.

Contact:

Douglas Bagnall, Developer: douglas@catalyst.net.nz

Kathryn Tyree, Project/Account manager: kathryn@catalyst.net.nz

Further reading/viewing:

Kiwi PyCon 2012: Speech Recognition with pocketspinx

NDF2014 Spying on the past

NDF2015 Using machines to find hidden structure in text archives

Award for Maori language recognition system by Te Māngai Pāho