Learning sounds you don't have

I long ago grew pretty tired of the pronunciation wars over Ancient Greek. I have opinions, have read a bunch of the literature, and have made my piece with how I currently pronounce Greek in different contexts. This doesn't mean I don't care about it, but I think there are lots of other things worth more time and attention that perfecting pronunciation.

Lately, however, I was thinking a little more about it, in conjunction with my ongoing reading of Felici's thesis. Old Irish has 66 phonemes. That's quite a lot. Scottish Gaelic does have 67 though, English around 44, I would put Ancient Greek at 25 consonants, 13 vowels, and a problematic number of diphthongs.

Of course, this depends upon your period and analysis of Greek pronunciation. Anyway, on the Old Irish side of the equation, this is one of the challenges for learners of Old Irish (including me): you need to learn a range of phonemes many of which are not distinguished or produced in your native language (L1), and you have not much in the way of audio. As I think about Felici's proposal for OI living language materials, it seems clear that extensive and high-quality audio is essential.

But lately I have been thinking a lot more about Ancient Greek as well. One of the problem learners face is typically φ/π/β and the parallel sets of χ/κ/γ and θ/τ/δ. It is much easier for most L1 English speakers to use fricatives, and then ignore aspiration on π,κ,τ, but if you want to adopt an attic restored pronunciation, then it's not that φ,χ,θ are difficult, it's that learning to de-aspirate π,κ,τ is difficult, especially in word-initial positions, and also to make sure that they remain distinct from β,γ,δ.

In the second half of 2025 as part of my Gaelic degree, we did a relatively (semi-technical but not all the way) deep dive into (Scottish) Gaelic phonology. This was done through the medium of Gaelic by the way. It was very helpful, partly because it also illuminated to me some features of Gaelic orthography.

Of course, learning how to produce particular sounds is not quite the same as learning to perceive particular sounds, and this has bearing on any L2, including Ancient Greek. So I got to wondering what sort of research and interventions exist for this.

I should have known better. There is a considerable amount of SLA work on these issues. From my relatively recent foray into that literature there are (naturally) competing theories about how these (perception and production in particular) relate, especially when learning an L2. This includes James Flege's Speech Learning Model, which if I understand it correctly posits that because an L2-learner doesn't come as a blank slate to an L2's phonemic system, how they distinguish L2 sounds depends in part upon distance/nearness to the phonemic categories in their L1. Learners need to develop new perceptual categories in order to also produce distinct sounds. Flege and Bohn's revised model argues that the interaction of L1 and L2 categories over time allows adults to form new phonetic categories.

Catherine Best's Perceptual Assimilation Model helps us because how an L2's sounds map to a learner's already existing categories, determines how difficult it is going to be to learn contrasts in the L2 sound system. E.g., because /p/ and /pʰ/ are perceived as /p/, it is going to remain hard for an English speaker to hear and/or produce them as distinct.

I don't have any experience in what I'm about to propose, but if you do I'd love to hear from you. One way to tackle this might be to apply High Variability Phonetic Training (HVPT). HVPT primarily tackles perception, not production, but it's built on the idea that exposure to multiple speakers producing 'the same' sound helps to train the learner to perceive the relevant distinctions. Because the multiple speakers have a natural degree of variation, the listener learns to focus on what is common, which is the contrasting sound they need to distinguish. There are some meta-analyses of HVPT which supports its efficacy for perception, and some research to suggest it might improve production as well (especially if you throw in some explicit articulatory training).

The problem for ancient Greek remains. You would need a number of speakers to record audio, each committed to realising the same pronunciation scheme, each with a pretty good ability to produce those distinctions, and then you'd need to organise that amount of sound into a discrete learning intervention. I don't think I have the time or energy to do that.

Which might be disappointing to hear, but perhaps it's a task someone will take up in the future. In the meantime, a little bit of attention to understanding the phonology of the language, learning to produce different sounds, and listening to whatever good audio recordings you can find, is probably the package of things I'd recommend you do.