Improve IPA transcriber for OE

The current IPA transcriber for OE largely ignores the phonological environment, relying mostly on unconditioned mappings of letters to IPA symbols.

To give a basic example, currently c yields /k/, so that first person singular ic is mapped to /ik/. But according to this resource on OE phonology:

When followed by a front vowel (æ, e, i) or the diphthongs ea or eo, or when preceded by the letter i AND not followed by a back vowel, c is pronounced /ʧ/.

The proposal is to improve the IPA transcriber by encoding conditional phonological rules. This process is a nearly ideal use case for test-driven development. A basic set of tests will be put in place as a target for the initial implementation of the rules. More demanding tests may then be constructed as drivers of iterative improvement.

There are downstream uses for IPA transcription, so I’d like to address this issue sooner rather than later.

[free-variation]

@clemsciences you’re right, for simple rules, as in item-specific ones with simple specifications of environmental features, the mechanism is wordy. The more complex rules I think benefit from this format.

I have therefore added a simplified mechanism on top of the existing one, roughly similar to the rules in your system:

# /h/ is velarized after a back vowel
oe.add_rule(SimplePhonologicalRule(h, x, before = [Backness.back]))

# /h/ is palatized after a front vowel
oe.add_rule(SimplePhonologicalRule(h, ch, before = [Backness.front]))

Enabled by this:

def check_features(phoneme, feature_values):
	return reduce(lambda a, b: a or b, [phoneme[type(f)] == f for f in feature_values])

def SimplePhonologicalRule(target, replacement, before=None, after=None):
	if before is not None and after is None:
		cond = lambda b, t, _: t == target and check_features(b, before) 
	if before is not None and after is not None:
		cond = lambda b, t, a: t == target and check_features(b, before) and check_features(a, after)
	if before is None and after is not None:
		cond = lambda _, t, b: t == target and check_features(a, after) 
	if before is None and after is None:
		cond = lambda _, t, __: t == target

	return PhonologicalRule(cond, lambda _ : replacement)

[clemsciences]

Wow, it does a much better work, but it’s really verbose (higher complexity leads to more code?). I’m going to read your code tomorrow with more attention.