Stephen Downes

This article raises some interesting questions about early language learning (ELL) and AI. Specifically, it makes the point that while AI is already quite good for some things (such as speech to word transcription) it needs specialized training for others, such as  low-stakes oral reading fluency, and absolutely flounders at others, such as phoneme-level assessment. The model needs "a separate step that compares the recording with the expected sounds." And "Improvements on this narrower task also may not translate into better general transcription." So what does this tell us about phonics as a thing that is learned? Maybe (and I'm speculating here) we learn how words are pronounced first, and parts of words later. The pronunciation of 'cat' and transcription to the letters c-a-t happen as a single step, and are not composed of separate steps (pronunciation of 'k', pronunciation of 'at', transcription to correct letters). Something to think about.