Minimal Pairs for Greek Phonology

[Jonathan Robie]

I am looking for a set of minimal pairs that can be used to teach the Greek alphabet. In this context, I am looking for pairs of words that differ by a single sound. A few examples:

ἄν / ἐν
ἀββα / ἀλλά
γέ / τέ
δέ / γέ
τέ / ὅτε
ὅτε / πότε
ὅτε / ὥστε
πότε / τότε

Does anyone know of a list of minimal pairs that illustrates the sounds of each letter and diphthong in the Greek language?

[Stephen Carlson]

I’ve wondered about this myself, and thought that it ought to be fairly straightforward to program a search to find all the minimal pairs in a corpus (e.g. strings with an edit distance of one, as a rough cut approximation).

[dirkroorda]

I did it for Hebrew here: https://github.com/ETCBC/bhsa/blob/mast ... airs.ipynb

It should work for Greek as well: if you make a list of all words, this program can compute all minimal pairs of that list.

Written in Python (3), with a Levenshtein module. Follow the link, and you'll see how to install that module.

[daveburt]

Randall Buth's Living Koine Greek course does this to introduce the alphabet. It begins with a set of words which only have one sound, so every pair of them differs by a single sound. e.g. α, ου, αι, οι, η; and I think it does what you're asking for, but presumably it's copyright.

[Jonathan Robie]

I did it for Hebrew here: https://github.com/ETCBC/bhsa/blob/mast ... airs.ipynb

It should work for Greek as well: if you make a list of all words, this program can compute all minimal pairs of that list.

Written in Python (3), with a Levenshtein module. Follow the link, and you'll see how to install that module.

Excellent! Thanks, Dirk!

[Jonathan Robie]

Since Dirk's notebook provides the code but not the data, I decided to post the data separately. And that somehow turned into a blog post, which you can read here:

Minimal Pairs for Greek and Hebrew

Thanks again to Dirk for his solution!

[daveburt]

That is pretty cool.

These pairs seem wrong for different reasons:

ברא בוא
ζάω ζέω

The first pair differs by more than one sound ('bara'/'bore'/'bora' vs. 'bo' -- I think it's still a consistent list, just not phonological). The second pair don't seem to be real, inflected words at all, and the real words they correspond to differ by zero.

You could `s/[άέό]ω/ω` and add all the inflected forms of a single verb (perhaps λύω!) to get more results. Or perhaps skipping the lemmatization step would give better results, with real words as they appear in the texts?

[NathanSmith]

This has been a stimulating topic for me. Now I'm looking into how the python Levenshtein module handles diacritics.

A question arose for me: Are there any minimal pairs (in terms of Greek phonology) which have an edit distance of two or greater? Presumably a diphthong swapped with a single vowel would fit into this category (one edit and one additional = edit distance of 2). I feel certain that English would have many of these, but I'm not sure if Greek would.

[daveburt]

Yes, morphology doesn't match phonology perfectly. καινος and κενος have a Levenshtein distance of 2 but phonologically identical (in imperial/modern); then ἑνος is a 'minimal pair' with both of them phonologically, but 3 edits from καινος. To get the result Jonathan originally set out for, one would have to first convert the words into a phonetic representation (e.g. IPA). Should accents be ignored or represented and allowed for in some way?

Furthermore, if the goal is learning to write, showing the different ways to write the same sounds is another task not covered by the 'pairs of words that differ by a single sound.'

[Jonathan Robie]

Yes, morphology doesn't match phonology perfectly. καινος and κενος have a Levenshtein distance of 2 but phonologically identical (in imperial/modern); then ἑνος is a 'minimal pair' with both of them phonologically, but 3 edits from καινος.

Excellent examples.

To get the result Jonathan originally set out for, one would have to first convert the words into a phonetic representation (e.g. IPA). Should accents be ignored or represented and allowed for in some way?

Right - I originally asked for pairs of words that differ by a single sound. If you first converted to IPA, then computed Levinshtein distance, it would give you that. Dirk's result is pretty useful, though, converting to IPA would be better. Of course, to do that, we would have to agree on one pronunciation ;->

Does anyone have a Greek to IPA utility for any pronunciation? If so, it could be easily adapted. It would have to be sophisticated enough to understand the tricky cases of how letters are pronounced in context.

Furthermore, if the goal is learning to write, showing the different ways to write the same sounds is another task not covered by the 'pairs of words that differ by a single sound.'

Yes, for reading or writing, I think.