Module talk:grc-pronunciation

Definition from Wiktionary, the free dictionary
Jump to navigation Jump to search

Identical pronunciations[edit]

Sometimes the module outputs the same pronunciation twice in a row, like on ὄργανον. In this case, it should only output it once. —CodeCat 11:41, 20 May 2014 (UTC)

I fail to see why. This indicates that the word retained the same pronunciation for a period of time. I can't fathom what indelible harm is caused by this. -Atelaes λάλει ἐμοί 17:38, 20 May 2014 (UTC)
And I fail to see why showing the same pronunciation twice is somehow useful. If I didn't understand it, do you think any other users will? —CodeCat 19:07, 20 May 2014 (UTC)
Well, the clarification is in the expand. I remain steadfast in my opinion that it should stay. -Atelaes λάλει ἐμοί 05:43, 21 May 2014 (UTC)
Expand? —CodeCat 12:04, 21 May 2014 (UTC)
Yeah, there's a "more" button on the right. It expands the three inline pronunciations to five pronunciations with stated periods. -Atelaes λάλει ἐμοί 17:32, 21 May 2014 (UTC)
I didn't even see that button. —CodeCat 17:38, 21 May 2014 (UTC)
There is definitely room for improvement on the placement, but it hasn't struck me yet. -Atelaes λάλει ἐμοί 18:26, 21 May 2014 (UTC)

Long vowel with acute?[edit]

How do I get it to do an orthographically ambiguous but phonetically long vowel with an acute accent? At ἐάν I can't figure out how to get {{grc-pron}} to understand that the alpha is long. —Aɴɢʀ (talk) 17:56, 8 July 2014 (UTC)

Velar nasal and other issues[edit]

@Atelaes, ObsequiousNewt, CodeCat, Wyang (I'm just pinging anyone who's worked on this since Atelaes doesn't seem to be around) We have all this information concerning pronunciation, yet a lot is missing from this template. Most notably, the velar nasal in words like ἄγκος does not show up. ObsequiousNewt has suggested this be the standard template from now on, but we can't use it if it has such glaring flaws. —JohnC5 00:13, 17 May 2015 (UTC)

@JohnC5 I'll try and fix it. Can you find anything else that's wrong?ObsequiousNewt (εἴρηκα|πεποίηκα) 20:23, 18 May 2015 (UTC)
The velar nasal is not a phoneme in Ancient Greek, it's an allophone of /n/. So we should show /ng/ and not /ŋg/. —CodeCat 20:53, 18 May 2015 (UTC)
@CodeCat Thanks for that clarification! What would be the phonemic over phonetic approach for the compensatory lengthening of caused by nasal deletion before fricatives in Byzantine and sometimes Classical Greek (-γχ-, -μφ-, -νθ-, -νσ-? —JohnC5 21:02, 18 May 2015 (UTC)
While the vowel was merely nasalised, it was not phonemic. But when the nasalisation was lost, then this would have caused a change in the phonemic structure of words. I'd need a bit more information before I could tell for sure though. For instance, what happened to a sequence like εγχ? was the result eventually just εχ, or did it become ηχ? —CodeCat 21:20, 18 May 2015 (UTC)
If that's the case, why are we showing phonemic pronunciation and not phonetic pronunciation? (Also, it seems evident that /N/ is supposed to be there, but isn't working for some reason.) ObsequiousNewt (εἴρηκα|πεποίηκα) 14:23, 19 May 2015 (UTC)
It seems like the most comely solution would be to have the phonemic in the un-expanded form and both the phonetic and phonemic in the expanded form. —JohnC5 17:23, 19 May 2015 (UTC)
Sure, I can fix it. Do you have a citation for the sound change? If not I can probably find one. —ObsequiousNewt (εἴρηκα|πεποίηκα) 15:59, 14 October 2015 (UTC)

Contemporary pronunciation[edit]

I don’t know much about the Greek language, but would it be possible and useful to add the modern (20th/21st-century) pronunciation or is the sound change not systematic enough? Crissov (talk) 16:04, 1 May 2016 (UTC)

  • This page is only for the pronunciation of Ancient Greek words, not Modern Greek words. Or do you mean how modern Greek people pronounce Ancient Greek when they're reading it? —Aɴɢʀ (talk) 16:07, 1 May 2016 (UTC)
I think Modern Greek pronunciation is unpredictable in some ways, like whether certain letters and digraphs represent [i] or [j], and there isn't a thoroughly worked out analysis of MG phonology yet, so it would be difficult or impossible to make the pronunciation automatically generate as the Ancient Greek pronunciations do. — Eru·tuon 17:47, 1 May 2016 (UTC)

/s/ voicing assimilation[edit]

The module currently transcribes Attic ζ (z) as /zd/, σβ (sb) as /zb/, and σμ (sm) as /zm/, as in Ζεύς (Zeús), σβέννῡμῐ (sbénnūmi), σμῑκρός (smīkrós). This is incorrect, I think; they should be /sd sb sm/. [z] is not a separate phoneme in Attic Greek, but an allophone of the coronal sibilant /s/ through anticipatory voicing assimilation. [zd zb zm] should only be shown in the phonetic transcription, when it is added.

/z/ became a phoneme when /sd/ (or more likely /dz/) simplified to /z/ in Koine, so /zd zb zm/ is correct for that stage of the language. — Eru·tuon 21:23, 13 October 2016 (UTC)

Layout change[edit]

@Erutuon, while I like the idea the idea of having the more button not on the far right, your change seems to have messed up pages like Ἀριστοφάνης. —JohnC5 05:11, 19 October 2016 (UTC)

@JohnC5: Oh, I see that. Bizarre. The POS header has vanished, even though it's there in the HTML. I'm not sure what's going on, so I'll just revert my change for now. — Eru·tuon 06:34, 19 October 2016 (UTC)
Yeah, I dunno. I really would like if we could get a better solution. Perhaps we could ask someone in the WT:GP? —JohnC5 14:07, 19 October 2016 (UTC)
I posted there, after putting my code in MOD:grc-pronunciation/sandbox, and asked if anyone had an idea what was going on, or an alternative solution. — Eru·tuon 16:36, 19 October 2016 (UTC)

I came up with the solution of counting the characters in the inline transcriptions and making the <div> element containing the "more" button be a certain number of em times that. It's funny, but the only way I can think to do it without a function for measuring the width of the characters in the font that the browser is currently using (which is probably impossible). So it looks better now at least. — Eru·tuon 23:48, 27 January 2017 (UTC)

Do you have to set a fixed width? Letting the browser layout figure it out is generally better. —CodeCat 00:05, 28 January 2017 (UTC)
The problem is, the browser puts the "more" button to be on the far right side of the content area. This has been criticized multiple times in the past. It makes it hard to find. If you can figure out a better solution, I'd be delighted, but until then a crappy one is better than nothing. — Eru·tuon 00:10, 28 January 2017 (UTC)
divs expand to fill all available space, by default. Putting style="display: inline-block;" may help. —CodeCat 02:51, 28 January 2017 (UTC)

Diphthongs[edit]

low mid front mid back
i ai̯
aːi̯

ɛːi̯
oi̯
ɔːi̯
u au̯
aːu̯
eu̯
ɛːu̯

ɔːu̯

The Attic diphthongs are inconsistent right now. Long diphthongs use a tie and a semivowel symbol – for instance, ᾍδης (Hā́idēs): /ha͜ːj́.dɛːs/ – while short diphthongs use a tie – οἷος (hoîos): /hó͜i.os/. I propose that both use a non-syllabic diacritic, as shown in the table to the right. A tie bar is nonstandard, I think.

I tried to implement this myself, and I got the module to produce the right symbols, but the syllabification broke and I couldn't figure out how to fix it. — Eru·tuon 03:40, 24 October 2016 (UTC)

I'd prefer the nonsyllabic diacritic with no tie bar, too. I don't know how to change it without breaking the module, though. —Aɴɢʀ (talk) 13:01, 24 October 2016 (UTC)

I made alpha-iota subscript be transcribed with /aːi̯/, but currently the syllabification puts a syllable break between the /aː/ and the /i̯/. I don't understand the syllabification function yet. I'll continue looking at it, but if you have an idea of how to fix it, JohnC5, I would be glad for your help. — Eru·tuon 07:36, 30 January 2017 (UTC)

@Erutuon: You beat me to the punch with this edit. I was about to do the same thing. —JohnC5 08:01, 30 January 2017 (UTC)
@JohnC5: I still don't really understand what cVowel and nVowel are, or why my edit worked. ^_^ — Eru·tuon 08:03, 30 January 2017 (UTC)
@Erutuon: cVowel is the position of the end of the current vowel (cluster) and nVowel is the beginning of the next vowel (cluster). Then the function finds sBreak , the final character of the syllable, then lops off characters [1:sBreak] before repeating until there are no characters left. —JohnC5 08:10, 30 January 2017 (UTC)
@Erutuon, Angr: Should we actually implement a phonemic-phonetic distinction based on w:Ancient Greek phonology#Accent? It wouldn't be too bad. Also, would (hṓi) be /ɔ́ɔi̯/, [ɔːî̯] and (hôi), /ɔɔí̯/, [ɔːǐ̯] or something different? Moreover, it seems that graves should not be transcribed at all. Is that correct? —JohnC5 21:29, 30 January 2017 (UTC)
The acute would have rising tone, and the circumflex falling, so the transcriptions sould be reversed. The grave is a thorny issue, but my personal theory is that on di- or trisyllabic words content words, it represented high pitch, while on monosyllabic words (and maybe some disyllabic words), it represented lack of high pitch (that is, enclisis or proclisis). Probably monosyllabic function words that are written with circumflex were actually unaccented too. — Eru·tuon 21:42, 30 January 2017 (UTC)
@Erutuon: Yeah, I got those accents exactly backwards by mistake. Sorry about that. But you haven't answered the questions: Where do the accents go on trimoraic vowels? Should we provide phonemic or phonetic transcriptions or both (but not the odd mix we're giving now)? —JohnC5 02:56, 31 January 2017 (UTC)
@JohnC5: I guess I would just simplify and use rising and falling accents on all long monophthongs and diphthongs. Probably we should have phonemic and phonetic transcriptions. They are already needed because the voicing of /s/ in Attic that is shown in the transcriptions now is actually phonetic.
I was hesitant to respond for two reasons. I'm uncertain now about the form the phonemic accent markings should take, and I dislike using two vowels in the cases of eta and omega, since the short vowel phonemes /*ɛ *ɔ/ invalid IPA characters (*) don't exist. However, high pitch on the first or second element of the diphthong would be the typical analysis, and perhaps it's fine to have a doubled vowel when no short vowel exists.
So anyway, your proposed phonemic and phonetic transcriptions look fine, except I would add a falling pitch mark on a long vowel that follows an acute: for instance, ἀλήθεια (alḗtheia) [a.lɛ̌ː.tʰeî̯.ja], δίδωμι (dídōmi) [dí.dɔ̂ː.mi]. — Eru·tuon 03:30, 31 January 2017 (UTC)
Did you mean to give those as phonemic or phonetic? I think while going from phonemic to phonetic would be fairly simple (Think a few regex's could do it), adding a falling tone after an acute would be perhaps more complex. And would this post-tonic falling tone be phonemic or phonetic? —JohnC5 03:37, 31 January 2017 (UTC)
Oops, they should be phonetic. Corrected. — Eru·tuon 03:48, 31 January 2017 (UTC)
@Erutuon: This also opens a Pandora's box of what's phonemic vs. phonetic. My idea would be to just generate the phonemic representation with current system, then generate the phonetic by regex. I'm also going to set up diphthong parsing soon and then simplify the conversion function. What do you think? —JohnC5 03:43, 31 January 2017 (UTC)
Yeah, that should work. I'm not sure if the falling tone is phonemic or not. It's probably the location of stress, though. — Eru·tuon 03:55, 31 January 2017 (UTC)

@Erutuon: I've finished with my current set of alterations to the module. Now, the code uses the data module for all conversions (macrons and breves vowels now have an entry as something like "α¯" and every diphthong combination that we currently recognized has an entry). Could you check over the examples to ensure they are all rendering correctly? Now, however, the vowel ambiguity notes are not created. Could I ask you to write a regex to catch the ambiguous vowels? Also, any cleanup you can make is much appreciated! Once that's done, we can consider splitting stuff into phonemic vs. phonetic. —JohnC5 06:42, 31 January 2017 (UTC)

@JohnC5: All the examples look good now! I'm pleased. I'll look into a way to check for ambiguous vowels. Ideally it would take accent into account: for instance, penultimate accent on Διονύσια (Dionúsia), or the circumflex on πρᾶγμα (prâgma), that indicate the final alpha is short. But that might be hard to implement. The simplest would be to check for a macron, breve, or circumflex on the vowel in question. — Eru·tuon 07:05, 31 January 2017 (UTC)
@Erutuon: I was imagining something like '[αάὰἀἄἂἁἅἃιίὶἰἴἲἱἵἳϊΐῒυύὺὐὔὒὑὕὓϋΰῢ][^¯˘υύὺῦὐὔὒὖὑὕὓὗιίὶῖἰἴἲἶἱἵἳἷ]?' , but this isn't quite right. —JohnC5 07:19, 31 January 2017 (UTC)
I created a complicated function, but it is resulting in false positives since the macrons and breves are spacing characters. Figuring out a way around that now. — Eru·tuon 07:47, 31 January 2017 (UTC)
Okay, I went the regex way, but without listing all the characters. Very pleased with the result. — Eru·tuon 08:14, 31 January 2017 (UTC)
Oh, have to take into account diphthongs. — Eru·tuon 08:15, 31 January 2017 (UTC)
Note to self: Tomorrow, I'll also need to rewrite the syllabify function handle multi-word inputs better. —JohnC5 08:34, 31 January 2017 (UTC)
@Erutuon: So any idea what's causing the current error? —JohnC5 19:38, 31 January 2017 (UTC)
@JohnC5: No, I'm mystified. The getTokens function works fine in Module:grc-translit, and I don't see why it wouldn't work in the pronunciation sandbox module. There must be something I'm overlooking. — Eru·tuon 19:41, 31 January 2017 (UTC)
Oh, bleh. I finally realized the problem. tokens[1] is nil, so # and ipairs() don't work. So yay! Now it's fixed! — Eru·tuon 20:22, 31 January 2017 (UTC)
Each token in the tokens table contains consonants, monophthongs, or diphthongs plus their diacritics. So I can just search for an ambiguous vowel without any of the diacritics that mark it as long or short. — Eru·tuon 20:25, 31 January 2017 (UTC)

@Erutuon: So are you fine with me going ahead with the following phonetic-phonemic relationship?

ω (ō) acute circumflex grave
plain ώ ()
/ɔ͜ɔ́/
[ɔ̌ː]
(ô)
/ɔ́͜ɔ/
[ɔ̂ː]
()
/ɔ͜ɔ̀/
[ɔ᷆ː]
iotated (ṓi)
/ɔ͜ɔí̯/
[ɔːǐ̯]
(ôi)
/ɔ́͜ɔi̯/
[ɔːî̯]
(ṑi)
/ɔ͜ɔì̯/
[ɔːi̯᷆]

JohnC5 03:38, 1 February 2017 (UTC)

Go ahead. I'm not happy with the transcriptions of the grave accent, but that can easily be changed and otherwise it looks fine. — Eru·tuon 03:45, 1 February 2017 (UTC)
Nor am I, but then again, I wouldn't put any accent for the grave at all. What representation would you use instead? —JohnC5 03:48, 1 February 2017 (UTC)
I would put a high pitch, if it's a content word, or no accent, if it's a short enclitic or proclitic, like δὲ (). (Same for circumflex-accented articles like τοῦ (toû).) I suppose it would be easiest for graves to be transcribed as high pitch by default, and then to manually enter unaccented forms into the template in entries where appropriate. — Eru·tuon 03:54, 1 February 2017 (UTC)
@CodeCat, Angr, ObsequiousNewt, Isomorphyc, I'm so meta even this acronym, do any of you have opinions on this matter? —JohnC5 04:30, 1 February 2017 (UTC)
I dislike using a tie bar for diphthongs and long vowels, and I dislike having the tone mark over the nonsyllabic part of the diphthong. I don't have an opinion on the best way of indicating grave accent, but of course it won't come up very often in dictionary entries anyway. —Aɴɢʀ (talk) 09:56, 1 February 2017 (UTC)
@Angr: To clarify, the tie bar would not be used for long vowels and not diphthongs. It's meant to represent the bimoraic character of the long vowels and that, as Erutuon points out, /ɔ/ does not exist properly outside of this formation. As for the accent, would you prefer it always above the vowel? —JohnC5 15:38, 1 February 2017 (UTC)
Yes, the accent should go above the vowel; and long vowels can be indicated with the long mark, as in the transcriptions above inside brackets instead of slashes. If we can use [ɔ̌ː], we can use /ɔ̌ː/. —Aɴɢʀ (talk) 15:59, 1 February 2017 (UTC)
That is not phonemically correct. They are not underlyingly rising or falling tones, they are bimoraic vowels with the accent on one of the morae, which is realized phonetically as a rising or falling tone. —JohnC5 16:58, 1 February 2017 (UTC)
That's a distinction without a difference. But even /ɔɔ́/ would be preferable to /ɔ͜ɔ́/, since the tie bar is supposed to represent coarticulation, and it makes no sense to use it with two copies of the same sound. —Aɴɢʀ (talk) 18:46, 1 February 2017 (UTC)
The primary purpose of the tie is to simplify the syllabification process, but I guess we could use the tie internally then remove it after the fact. Maybe we'll just scrap this moraic notation. I certainly would like to supply other phonetic distinctions where possible (like Classical /s/ to [z] / _[+voiced]}). Are there any other phonetic features we are missing? Specifically, what of Wiktionary:Ancient Greek transliteration should we add or move from phonemic to phonetic? —JohnC5 19:11, 1 February 2017 (UTC)
One thing is that the palatals should be moved to phonetic. Since they occur in complementary distribution with the velars under our system, it makes no sense to consider them separate phonemes. — Eru·tuon 02:07, 2 February 2017 (UTC)
Yes. We've had it for a long while, but I don't like the tied symbol for long vowels. There is a perception that Greek vowels were moraïc, but it's not a good perception. First and most prominently, the actual length of Greek long vowels was less than twice the length of short vowels (or at least, so I think, but I can't remember where I read this); secondly, there is no "monomoraïc" /ɛ/ or /ɔ/ (or, after a certain point, dimoraic /e/ or /o/); thirdly, while tones do tend to combine moraïcally when vowels are contracted, this is really just a side effect of existing rules, and any sensible analysis of Greek tones would put them at the word-level rather than the mora-level—and fourth, counting tones by mora doesn't really make any sense either, since tones can fall on any of the last four mora of a word. I would advocate just using the simple system of representing ε έ ει εί εῖ as /e é eː ěː êː/ (regarding the grave, I'm not sure I see enough evidence to say it marked a higher than normal tone, and anyway it is almost never going to come up in entries.) I also would point out that, while τοῦ etc. may have been unstressed in casual speech, we are (or should be) giving phonemic-level pronunciations here, so it should be represented as /tôː/. In fact, I believe that we should not speculate on phonetic-level pronunciations at all. ObſequiousNewtGeſpꝛaͤchBeÿtraͤge 01:50, 2 February 2017 (UTC)
There are many phonetic or possibly phonetic processes that are not mere speculation: devoicing of r at the beginning of a word, voicing of s, gemination of the last element of diphthongs before a vowel, nasalization of g before m, assimilation of final n to a following consonant, phonation assimilation of stops in clusters. Why would we exclude these? — Eru·tuon 02:07, 2 February 2017 (UTC)
I am not familiar with the evidence for destressing τοῦ, so if you could provide it that would be appreciated. I also would contend that the /ŋm/ hypothesis is speculation, and should not be written. ObſequiousNewtGeſpꝛaͤchBeÿtraͤge 03:59, 2 February 2017 (UTC)
I would mention that Philomen Probert in A Companion to the Ancient Greek Language does support γμ (gm)[ŋm]. —JohnC5 04:52, 2 February 2017 (UTC)
@ObsequiousNewt: I think I read the thing about circumflexed articles being unaccented in The Prosody of Greek Speech by Devine and Stephens. I unfortunately don't own the book and it's been a while since I tried to read it, so I can't say specifically what the reasoning was. (I didn't understand the parts concerned with metrical analysis of Ancient Greek poetry.) I will post more when my keybord stops not typing the first letter of the lphbet. — Eru·tuon 05:39, 2 February 2017 (UTC)

Words only attested in Koine or Medieval Greek[edit]

I just happened upon the word ἀρχιμανδρῑ́της (arkhimandrī́tēs). As can be inferred from its meaning, it was coined in the Koine Greek period, and it is anachronistic to show the Attic and Egyptian pronunciations. The template could use a parameter to hide these pronunciations. I'm not sure how this should work, though. — Eru·tuon 21:44, 1 November 2016 (UTC)

Syllabification error[edit]

@Erutuon, ObsequiousNewt: The syllabification in the Egyptian pronunciation of Αὖλος (Aûlos) (/ˈa.w.los/) is wrong. Could someone fix it? —JohnC5 06:05, 27 January 2017 (UTC)

@JohnC5 That's more than my ability to fix at the moment. I don't understand the core of the module very well. — Eru·tuon 19:25, 27 January 2017 (UTC)
@Erutuon: Thanks. I'll check it out then. —JohnC5 19:59, 27 January 2017 (UTC)

@Erutuon: In case you were curious about the distinction between loadData and require: mw:Extension:Scribunto/Lua_reference_manual#mw.loadData. —JohnC5 04:40, 29 January 2017 (UTC)

@JohnC5: I've been looking at that, but so far I can't figure out what capability loadData lacks that we require in the module. — Eru·tuon 04:43, 29 January 2017 (UTC)
@Erutuon: I believe it is the for loop in the check function. If it were converted to pairs, it might work. —JohnC5 04:57, 29 January 2017 (UTC)
@JohnC5: Ah, there it is. It tried to use # on a variable that is sometimes an element in the data tables. 04:59, 29 January 2017 (UTC)
I converted it to ipairs(), because all the keys are numbers rather than strings. And now loadData works! Thanks for figuring it out! — Eru·tuon 05:07, 29 January 2017 (UTC)
@Erutuon: I'm finally starting to get how this module works. Do you need the sandbox at all? I'm gonna try to factor out a few more things. —JohnC5 05:10, 29 January 2017 (UTC)
@JohnC5: I was just thinking of moving some functions around, but that's not very important, so you can use the sandbox module if you've got something more substantial to do. — Eru·tuon 05:18, 29 January 2017 (UTC)
@Erutuon: I'm just going to extract all the functions defined internally. This should give better clarity on what variables each function actually uses. It shouldn't take too long, and I'll tell you when it's done. —JohnC5 05:21, 29 January 2017 (UTC)
@Erutuon: My attempt failed. Do what you were going to do. —JohnC5 05:32, 29 January 2017 (UTC)
@JohnC5: The frustrating thing now is that the module errors don't tell where in the module sub is being given a bad argument. — Eru·tuon 06:00, 29 January 2017 (UTC)
@Erutuon: Yeah, these parsing functions are the worst. —JohnC5 06:07, 29 January 2017 (UTC)
@JohnC5: Module:UnitTests might give more information, but I'm not sure how we would use it in a module with such complex output. — Eru·tuon 06:24, 29 January 2017 (UTC)
@Erutuon: With this ingenious and masterful edit, I fixed the sandbox. :PJohnC5 06:42, 29 January 2017 (UTC)

──────────────────────────────────────────────────────────────────────────────────────────────────── What! How did I miss that? — Eru·tuon 06:43, 29 January 2017 (UTC)

I missed it as well, so you're good. —JohnC5 06:46, 29 January 2017 (UTC)

Decomposition[edit]

I can't help wondering if the module would be simpler with decomposition (mw.ustring.toNFD). Then there wouldn't need to be the tables with diacritic = true; you could just search for the diacritic. Or there could be a function that decomposes the current letter and returns true if it has the diacritic you're asking for. — Eru·tuon 07:25, 29 January 2017 (UTC)

That could work well. I, on the other hand, was wondering whether it would be easier to just do a straight mapping from letters to a base form and then uses regex's to apply all the phonological rules. That would be much easier to maintain and much more obvious what was going on. —JohnC5 07:30, 29 January 2017 (UTC)
I tried to use regex to determine if a sound belongs to a category, but it seems to cause scripts to time out. What on earth? — Eru·tuon 08:28, 29 January 2017 (UTC)
Well, find is faster I guess. — Eru·tuon 08:32, 29 January 2017 (UTC)
It's an interesting question. Table lookups are far more efficient (hash tables so ), whereas the find function is ). The cost of hashing the tables upfront is the expensive part, but I'm not sure how expensive. —JohnC5 08:37, 29 January 2017 (UTC)

@Erutuon: ᾠδῇ (ōidêi) should produce /ɔ͜͜͜ːj.dɛ́͜ːj/. I think something in the accent decomposition is removing the iota. —JohnC5 00:53, 30 January 2017 (UTC)

@JohnC5: Oops yes, that was my fault. I added subscript to the list of all diacritics but not to the diacritic ordering part of the functions. It should show up again now. — Eru·tuon 01:09, 30 January 2017 (UTC)
@Erutuon: Thank you sir. —JohnC5 01:17, 30 January 2017 (UTC)
@JohnC5: Do you happen to know if the iota subscript is supposed to be placed at a particular place in the sequence of diacritics? I put it last, but maybe that's wrong... — Eru·tuon 01:27, 30 January 2017 (UTC)
@Erutuon: Seems to be working at the moment, so...I guess at the end is correct? —JohnC5 01:35, 30 January 2017 (UTC)
The end makes sense, since ὠιδῆι (ōidêi) is an alternative spelling of ᾠδῇ (ōidêi), in other words iotas subscript aren't always written subscript. —Aɴɢʀ (talk) 07:19, 30 January 2017 (UTC)

κδ, ρρ[edit]

@JohnC5: I think the reason why these clusters aren't decoding properly is that the function decode refers to itself, before it is defined. We should probably break up that function. — Eru·tuon 09:40, 30 January 2017 (UTC)

Then again, recursion seems to work with another function, check, so maybe I am wrong. — Eru·tuon 09:46, 30 January 2017 (UTC)

I fixed it, but I'm a bit unclear about the reason why the other way doesn't work. —JohnC5 16:57, 30 January 2017 (UTC)
@JohnC5: Thanks! It seems to be due to the or after the mw.ustring.match. It must mess with the assignment of the variables somehow. — Eru·tuon 21:11, 30 January 2017 (UTC)

Consonant clusters between syllables[edit]

Our method of dividing syllables needs some tweaking. In particular, πατρίς (patrís) should probably be /pat.rís/ rather than /pa.trís/, because its first syllable is heavy in Homer. The voiced stop–liquid clusters are the tricky ones: sometimes they're split between syllables, making the previous syllable heavy, sometimes they're not. I don't remember the details on that.

@JohnC5 How do I change the syllabification of /tr/ and its compatriots? The 'clusters' list in Module:grc-pronunciation/sandbox/data doesn't seem to affect it. — Eru·tuon 04:28, 31 January 2017 (UTC)

The "clusters" functionality currently has no usage in the code. My issue here is that the poetic scansion doesn't necessarily represent actual syllabification (Latin's scansion rules certainly has many exceptions to the phonology to make it work). AG poetry even changes the length of vowels when necessary! —JohnC5 04:49, 31 January 2017 (UTC)
Well, some clusters are capricious in connection to syllable weight, but others are definitely divided one way and not the other. For instance, στ (st) is definitely divided /s.t invalid IPA characters (/) between vowels. I don't remember which of the voiced stop–sonorant or voiceless stop–sonorant was the capricious one. I suppose scansion is removed more or less from actual pronunciation, but it is based on the phonological patterns of the language, and has been used as evidence for vowel length and syllable division. — Eru·tuon 23:41, 10 May 2017 (UTC)

Selecting periods to transcribe[edit]

Some words began to be used after the Classical period. For those, we need to have a way to prevent earlier periods from being transcribed. I've added an example, ἀρχιμανδρίτης (arkhimandrítēs), to Module:grc-pronunciation/sandbox/documentation, with the parameter |period=koi2. Not sure how to make the module recognize it, though. @JohnC5, can you help? — Eru·tuon 17:38, 6 February 2017 (UTC)

@Erutuon: This should not be very difficult. I'll work on this later tonight. —JohnC5 17:46, 6 February 2017 (UTC)

Invalid characters[edit]

@Erutuon, JohnC5 Some pages are producing invalid IPA characters: ὀρίνειν, ὀρίνειν, ὀνομαστή. DTLHS (talk) 17:41, 24 February 2017 (UTC)

Whoops, just realized that these entries don't use the module- sorry. DTLHS (talk) 19:40, 24 February 2017 (UTC)
@DTLHS: Yes. Many issues would be solve by our switching all the instances of {{grc-ipa-rows}} to {{grc-IPA}}. Indeed, the old template gives the wrong syllable count as well, as seen in θρόνους (thrónous). If @Isomorphyc could get is bot to convert all the instances of the old template to the new one, that would be much appreciated. We can set up mod:grc-pronunciation to fill a category with entries that have ambiguous vowels. Also, I'm wondering how @ObsequiousNewt's investigation into AG phonology is coming. —JohnC5 19:47, 24 February 2017 (UTC)
I've done the extremely easy ones, but I think 80% of the transclusions which remain are only moderately easy to fix, meaning there are about a dozen groups which each may take a couple of hours of testing to replace accurately. When I return more actively in the next week or two, I have a few other templates and modules on my to-do list, but I'll prioritise this just after the very important items. I'll probably post a new to-do list to remind myself what I'm working on. Thanks for bringing this again to my attention; hope you've all been well! Isomorphyc (talk) 20:49, 24 February 2017 (UTC)

Moving this into production[edit]

@Erutuon: Can we move this into production? Also, can we add a category to mark anything that has an ambiguous vowel? —JohnC5 19:14, 29 March 2017 (UTC)

@JohnC5: I think there are issues with nasal-stop sequences brought up by @Angr that should be resolved. Sure, we could add a category. I was thinking of making the headword templates do that, but so far haven't made that happen. (It might be easiest to do it with Module:grc-headword, which isn't functional yet. There are other things that a headword module could do to make generating headwords simpler for editors.) In any case, we should track ambiguous vowels in the various templates separately, since sometimes macrons or breves are marked in one but not another. We just need to come up with category names.
It might be useful to have a function that would add length marks based on the accent rules: so that, for instance, μοῖρα and χώρα would be changed to μοῖρᾰ and χώρᾱ before generating the IPA and checking for ambiguous vowels. That might cause problems with oddly accented words, though. — Eru·tuon 21:02, 29 March 2017 (UTC)
Okay, an idea: a general category for templates containing vowels of ambiguous length, with subpages for each template. (There could also be subpages for individual vowels, but that wouldn't strictly be useful; it would be interesting, though, to see which vowel is most frequently ambiguous.)
If the idea of using the accent rules were implemented, then there could be separate categories for ambiguous vowels and ambiguous vowels whose length cannot be inferred using the accent rules. — Eru·tuon 21:30, 29 March 2017 (UTC)
Using the accent rules would be cool. If you want, I can work on Mod:grc-headword for you. I've done a few of these guys already. —JohnC5 21:54, 29 March 2017 (UTC)
@Erutuon: I thought there was some debate about this nasal-stop business, which is why we didn't do it initially. —JohnC5 21:56, 29 March 2017 (UTC)
After further thought, I have no objection to the new code being moved to the main module. There are a bunch of issues I'd love to see resolved, but there's no reason that has to be done now and not later. — Eru·tuon 20:57, 25 April 2017 (UTC)
A question that bothers me is whether diphthongs before a vowel – as in Ἀθηναῖος (Athēnaîos), στοιχεῖον (stoikheîon), οἷος (hoîos) – are not really a vowel plus doubled semivowel (since their origin often suggests they were originally doubled semivowels), and whether diphthongs themselves do not include a semivowel phoneme. Other thorny questions relate to the pitch accent: what part of the pitch accent is actually phonemic, and to what degree are the written pitch accents (their type and position) really accurate for Classical Attic, rather than early Koine Greek, at the time when the pitch accent symbols were invented. It is, at the very least, clearly anachronistic to project the written Koine-era accents back onto fifth-century Attic; to what degree anachronistic is the question (perhaps ultimately insoluble). And I would also like to see a phonetic transcription added, at the very least to show voicing of /s/ before a voiced consonant. All these issues are somewhat difficult, and since they do not impact the functioning of our sandbox code, I would simply go ahead with implementing sandbox code for now. — Eru·tuon 00:42, 26 April 2017 (UTC)
@Erutuon: Shall we then. I'd mention that we should keep the cleanup categories. —JohnC5 03:28, 26 April 2017 (UTC)

I went ahead and copied the sandbox code into the main module. There shouldn't be any module errors. But the cleanup categories have to be re-added. — Eru·tuon 20:00, 10 May 2017 (UTC)

Is there anything different about using {{grc-IPA}}? —Aɴɢʀ (talk) 20:19, 10 May 2017 (UTC)
As far as I'm aware, there are no differences in the usage of the template, now that its module code has changed. However, the preview-only message about ambiguous vowels is different. — Eru·tuon 20:23, 10 May 2017 (UTC)
@Angr: The only new external feature is that you can specify the |period= at which you would like it to start. See ἀρχιμανδρίτης (arkhimandrítēs) for an example. Everything else is internal. —JohnC5 20:28, 10 May 2017 (UTC)
@JohnC5: That's a cool feature. Could you edit Template:grc-IPA/documentation to reflect this, and list the abbreviations required for each period? And I see that vowel quantity marks are no longer required as of Egyptian Koine, since vowel length distinctions had been lost by then. That's good! It's very frustrating trying to clear out CAT:Ancient Greek terms with incomplete pronunciation when it's full of Koine words that never appeared in poetry. —Aɴɢʀ (talk) 20:38, 10 May 2017 (UTC)
@Angr: I've added a description of the parameters to the template documentation page. — Eru·tuon 20:52, 10 May 2017 (UTC)
@JohnC5, Erutuon: period= is causing module errors at ἀρχιμανδρίτης and ἁρματοδρομία. —Aɴɢʀ (talk) 14:06, 11 May 2017 (UTC)
@Angr: Odd that there were no module errors in the sandbox module. Fixed. — Eru·tuon 16:34, 11 May 2017 (UTC)

Cleanup category[edit]

@Erutuon, JohnC5: The module already makes the template show an error message (in edit mode) when α, ι, or υ isn't marked as long or short. Would it be possible for it also to put the word into a (hidden) cleanup category, say CAT:Ancient Greek terms with incomplete pronunciation? —Aɴɢʀ (talk) 10:50, 15 April 2017 (UTC)

@Angr: We were discussing this in the preceding conversation. It is certainly an forthcoming feature. I'm waiting on a phonological survey from @ObsequiousNewt before progressing too far, but I can add this category now, I think. It may become more elaborate in the future. —JohnC5 23:25, 15 April 2017 (UTC)
@Angr: I've created the cleanup category, but remember that anything using {{grc-ipa-rows}} will not be included. —JohnC5 23:44, 15 April 2017 (UTC)
@JohnC5: that's fine: Special:WhatLinksHere/Template:grc-ipa-rows itself functions as a cleanup category. ObNewt hasn't been around in over two months, so I wouldn't wait on him for anything. —Aɴɢʀ (talk) 14:13, 16 April 2017 (UTC)
@Angr: Yeah, I know both of those things, but hope springs eternal in the human breast. Also, @Erutuon is hesitant to put the new code into production. —JohnC5 16:22, 16 April 2017 (UTC)
It isn't already? It's certainly widely used, and I already regularly replace {{grc-ipa-rows}} with {{grc-IPA}} wherever I encounter it. —Aɴɢʀ (talk) 17:48, 16 April 2017 (UTC)
@Angr: Ah, let me clarify: the “new code” is Erutuon and my massive refactoring found in Module:grc-pronunciation/sandbox. It is more phonologically accurate and full-functioned as well as more maintainable and efficient. You should continue to replace {{grc-ipa-rows}} with {{grc-IPA}}. I was trying to get someone to do it with a bot, but CodeCat won't help unless we remove the collapsing functionality (of which I could be convinced, mind you). —JohnC5 18:15, 16 April 2017 (UTC)

@Angr, Erutuon: I've pretty much cleaned out Category:Ancient Greek terms with incomplete pronunciation. I was hoping y'all could take a look at the remaining members and make a decision. Where there is no way of knowing, I have been choosing short then leaving a note in markup. —JohnC5 04:38, 5 June 2017 (UTC)

@Angr: Thanks for your help. I had been using Osthoff as a heuristic as well. Anyway, what's next? Should we make a concerted effort to get a bot to convert {{grc-ipa-rows}} to {{grc-IPA}}? I know he's not on very often, but @Isomorphyc would be very helpful in this respect. —JohnC5 15:40, 5 June 2017 (UTC)

Palatalized gamma error[edit]

At γε (ge), for example, the 10th- and 15th-century pronunciations are given as /ɟe/ with a voiced palatal stop. Shouldn't that be /ʝe/ with a voiced palatal fricative? —Aɴɢʀ (talk) 15:34, 19 May 2017 (UTC)

@Angr: Yes, it should. The condition was correctly programmed in Module:grc-pronunciation/data ({ '-1=γ', 'ɟ', },), but it was being misapplied by the function. (It was looking for a letter at position 0 in the word, and getting instead the letter at position 1: hence, the function thought the gamma was preceded by another gamma and gave it its pre-gamma pronunciation.) — Eru·tuon 16:07, 19 May 2017 (UTC)

Syllable after a circumflex[edit]

Could the module be made clever enough to know that α, ι, υ in a syllable after a circumflex must be short, and thus no longer requires explicit marking as such? CAT:Ancient Greek terms with incomplete pronunciation would probably be a lot smaller if unambiguous forms like βοῶπις weren't in it. —Aɴɢʀ (talk) 16:28, 23 May 2017 (UTC)

Ypsilon diphthongs[edit]

@Erutuon, can figure out why the υι (ui) in ὄργυιᾰ (órguia) is not getting picked up as a diphthong? —JohnC5 01:02, 4 June 2017 (UTC)

@JohnC5: Strange. υἱός (huiós) is also going crazy. I'll take a look at it. — Eru·tuon 01:29, 4 June 2017 (UTC)
@Erutuon: Oh, I'd not seen that. This makes me think it is almost certainly the tokenization going wrong. —JohnC5 01:34, 4 June 2017 (UTC)
@JohnC5: Strangely, it's not: I just added two examples to Module:grc-utilities/documentation § Tokenization and they're fine. — Eru·tuon 01:42, 4 June 2017 (UTC)
@Erutuon: Actually, the tokenization isn't used in operation of the pronunciation module outside of the ambiguous vowel finding, right? —JohnC5 01:48, 4 June 2017 (UTC)
@JohnC5: No, I don't think it is. The relevant thing here is the "iDiphth" list in Module:grc-pronunciation/data. I'm puzzled why it isn't working (there's an upsilon there). — Eru·tuon 01:58, 4 June 2017 (UTC)
Actually, the "iDiphth" doesn't seem to be used at all. Huh. — Eru·tuon 02:05, 4 June 2017 (UTC)
@Erutuon: What makes you say that or in what sense? —JohnC5 02:13, 4 June 2017 (UTC)
@JohnC5: Well, I searched both the main module and the data module, and couldn't find anywhere where the iDiphth thing was used. For instance, isIDiphth doesn't use it. Or else my brain is overheating and malfunctioning. — Eru·tuon 02:16, 4 June 2017 (UTC)
@Erutuon: Oh, you mean the thing at the veeeery bottom? Yeah, I think we don't use that. In any case, I fixed it with this. This issue was that υ (u), which uniquely among vowels may be both the first and second member of a diphthong, was having its pre function overwritten. So, we've not been parsing the υι (ui) diphthong correctly for a while now. My bad. Also, is the representation given in Wiktionary:Ancient_Greek_transliteration really correct for this diphthong? That seems very weird to me. —JohnC5 02:29, 4 June 2017 (UTC)
@JohnC5: Which part of it? It being a long vowel? I'm not sure if that is accurate or not, but I don't remember exactly what Allen says about this diphthong. — Eru·tuon 02:35, 4 June 2017 (UTC)
@Erutuon: Allen seems to agreee. —JohnC5 02:42, 4 June 2017 (UTC)

Nasals in consonant clusters[edit]

See also Template talk:grc-IPA § Possible fixes

Currently, the nasals in consonant clusters are not handled correctly, as they were in the old template (see the documentation of the old template for detailed explanations; also there were some deficiencies in the old template as well). The two clusters μβ and νδ are currently partially handled but still not entirely correctly, the others are not handled at all. --WikiTiki89 20:17, 14 June 2017 (UTC)

@Erutuon, JohnC5, Angr: Does anyone want to comment on this? --WikiTiki89 15:45, 19 June 2017 (UTC)
Not me; I don't know what the correct treatment is after the 5th century BC. —Aɴɢʀ (talk) 15:47, 19 June 2017 (UTC)
@Angr commented on a related topic in Template talk:grc-IPA § Possible fixes. I don't know much about this either. I know the endpoint in Modern Greek, but not all that much about what happened between then and Koine. — Eru·tuon 16:29, 19 June 2017 (UTC)
Other than certain minor inconsistencies, shouldn't we have just assumed the old template was correct? --WikiTiki89 17:15, 19 June 2017 (UTC)
This is about nasalization of vowels, right? —JohnC5 17:25, 19 June 2017 (UTC)
The nasalization or non-nasalization the preceding vowel, the dropping or non-dropping of the nasal itself, and the realization of the following consonant are all part of this. --WikiTiki89 17:27, 19 June 2017 (UTC)
Well, we certainly can implement that... —JohnC5 17:38, 19 June 2017 (UTC)
Here's a list of all the cases (at least that I've thought of; they're all made-up words):
Test cases
 
  • (5th BCE Attic) IPA(key): /óm.bos/
  • (1st CE Egyptian) IPA(key): /ˈom.bos/
  • (4th CE Koine) IPA(key): /ˈom.bos/
  • (10th CE Byzantine) IPA(key): /ˈom.bos/
  • (15th CE Constantinopolitan) IPA(key): /ˈom.bos/
  •  
  • (5th BCE Attic) IPA(key): /óm.pos/
  • (1st CE Egyptian) IPA(key): /ˈom.pos/
  • (4th CE Koine) IPA(key): /ˈom.pos/
  • (10th CE Byzantine) IPA(key): /ˈom.pos/
  • (15th CE Constantinopolitan) IPA(key): /ˈom.bos/
  •  
  • (5th BCE Attic) IPA(key): /óm.pʰos/
  • (1st CE Egyptian) IPA(key): /ˈom.pʰos/
  • (4th CE Koine) IPA(key): /ˈom.ɸos/
  • (10th CE Byzantine) IPA(key): /ˈom.fos/
  • (15th CE Constantinopolitan) IPA(key): /ˈom.fos/

  •  
  • (5th BCE Attic) IPA(key): /ón.dos/
  • (1st CE Egyptian) IPA(key): /ˈon.dos/
  • (4th CE Koine) IPA(key): /ˈon.dos/
  • (10th CE Byzantine) IPA(key): /ˈon.dos/
  • (15th CE Constantinopolitan) IPA(key): /ˈon.dos/
  •  
  • (5th BCE Attic) IPA(key): /ón.tos/
  • (1st CE Egyptian) IPA(key): /ˈon.tos/
  • (4th CE Koine) IPA(key): /ˈon.tos/
  • (10th CE Byzantine) IPA(key): /ˈon.tos/
  • (15th CE Constantinopolitan) IPA(key): /ˈon.dos/
  •  
  • (5th BCE Attic) IPA(key): /ón.tʰos/
  • (1st CE Egyptian) IPA(key): /ˈon.tʰos/
  • (4th CE Koine) IPA(key): /ˈon.θos/
  • (10th CE Byzantine) IPA(key): /ˈon.θos/
  • (15th CE Constantinopolitan) IPA(key): /ˈon.θos/

  •  
  • (5th BCE Attic) IPA(key): /óŋ.ɡoi̯/
  • (1st CE Egyptian) IPA(key): /ˈoŋ.ɡy/
  • (4th CE Koine) IPA(key): /ˈoŋ.ɡy/
  • (10th CE Byzantine) IPA(key): /ˈoɲ.ɟy/
  • (15th CE Constantinopolitan) IPA(key): /ˈoɲ.ɟi/
  •  
  • (5th BCE Attic) IPA(key): /óŋ.koi̯/
  • (1st CE Egyptian) IPA(key): /ˈoŋ.ky/
  • (4th CE Koine) IPA(key): /ˈoŋ.ky/
  • (10th CE Byzantine) IPA(key): /ˈoɲ.ɟy/
  • (15th CE Constantinopolitan) IPA(key): /ˈoɲ.ɟi/
  •  
  • (5th BCE Attic) IPA(key): /óŋ.kʰoi̯/
  • (1st CE Egyptian) IPA(key): /ˈoŋ.kʰy/
  • (4th CE Koine) IPA(key): /ˈoŋ.xy/
  • (10th CE Byzantine) IPA(key): /ˈoɲ.çy/
  • (15th CE Constantinopolitan) IPA(key): /ˈoɲ.çi/

  •  
  • (5th BCE Attic) IPA(key): /óŋ.ɡos/
  • (1st CE Egyptian) IPA(key): /ˈoŋ.ɡos/
  • (4th CE Koine) IPA(key): /ˈoŋ.ɡos/
  • (10th CE Byzantine) IPA(key): /ˈoŋ.ɡos/
  • (15th CE Constantinopolitan) IPA(key): /ˈoŋ.ɡos/
  •  
  • (5th BCE Attic) IPA(key): /óŋ.kos/
  • (1st CE Egyptian) IPA(key): /ˈoŋ.kos/
  • (4th CE Koine) IPA(key): /ˈoŋ.kos/
  • (10th CE Byzantine) IPA(key): /ˈoŋ.ɡos/
  • (15th CE Constantinopolitan) IPA(key): /ˈoŋ.ɡos/
  •  
  • (5th BCE Attic) IPA(key): /óŋ.kʰos/
  • (1st CE Egyptian) IPA(key): /ˈoŋ.kʰos/
  • (4th CE Koine) IPA(key): /ˈoŋ.xos/
  • (10th CE Byzantine) IPA(key): /ˈoŋ.xos/
  • (15th CE Constantinopolitan) IPA(key): /ˈoŋ.xos/

  •  
  • (5th BCE Attic) IPA(key): /ómp.sos/
  • (1st CE Egyptian) IPA(key): /ˈomp.sos/
  • (4th CE Koine) IPA(key): /ˈomp.sos/
  • (10th CE Byzantine) IPA(key): /ˈomp.sos/
  • (15th CE Constantinopolitan) IPA(key): /ˈomp.sos/
  •  
  • (5th BCE Attic) IPA(key): /ón.sos/
  • (1st CE Egyptian) IPA(key): /ˈon.sos/
  • (4th CE Koine) IPA(key): /ˈon.sos/
  • (10th CE Byzantine) IPA(key): /ˈon.sos/
  • (15th CE Constantinopolitan) IPA(key): /ˈon.sos/
  •  
  • (5th BCE Attic) IPA(key): /óŋk.sos/
  • (1st CE Egyptian) IPA(key): /ˈoŋk.sos/
  • (4th CE Koine) IPA(key): /ˈoŋk.sos/
  • (10th CE Byzantine) IPA(key): /ˈoŋk.sos/
  • (15th CE Constantinopolitan) IPA(key): /ˈoŋk.sos/
  • --WikiTiki89 17:45, 19 June 2017 (UTC)
    Could you provide what they should look like, somehow? —JohnC5 18:00, 19 June 2017 (UTC)
    I don't know the details myself. The old template's documentation talks about it in detail, but I'm not sure whether it's 100% accurate about everything. Here are the same examples above entered the way I think they should be in the old template:
    Old template for the above examples
















    --WikiTiki89 18:18, 19 June 2017 (UTC)
    @Wikitiki89: Could you remind me of the sources for all the claims in Template:grc-ipa-rows/documentation? —JohnC5 21:00, 7 July 2017 (UTC)
    @Wikitiki89: I remembered that @CodeCat: mentioned in this discussion that the nasalization of the vowel is not phonemic, which is true. Unless we split the phonemic representation into phonemic and phonetic, then we should not show the nasalization. Showing the nasalization as phonemic is actively false. —JohnC5 02:29, 8 July 2017 (UTC)
    Is 15th-century Constantinopolitan that different from Modern Greek in terms of nasal clusters? It seems surprising and unlikely to me that intervocalic νδ would be either /d/ after an oral vowel (new template) or /d/ after a nasal vowel (old template) when AFAIK it's /nð/ in Modern Greek, isn't it? —Aɴɢʀ (talk) 08:06, 8 July 2017 (UTC)
    @Angr: Thanks for this comment. The question remains: what are the sources for Wiktionary:Ancient Greek transliteration and Template:grc-ipa-rows/documentation (which differ on the trestment of nd)? Also, as Sihler points out in §228, the absolute prohibition on ns only seems to be in Attic Greek. —JohnC5 13:50, 8 July 2017 (UTC)
    @JohnC5: Whenever *ns became long vowel + s in Attic, it was reflected in the spelling. It's not as if Attic authors wrote "φέρονσι" and "πάνσα" but pronounced them "φέρουσι" and "πᾶσα". And in §228.2a, Sihler points out that the prohibition on ns wasn't absolute, as late formations like ὕφανσις (húphansis) keep the ns even in Attic. So we definitely should not be telling the module (as the old template does) to treat written ονσος as if it were ουσος for 5th c. Attic. As for the sources, I don't know. —Aɴɢʀ (talk) 15:42, 8 July 2017 (UTC)
    So that leaves the questions about the later dialects, for which I again would like sources. @Wikitiki89: any help on this would be spectacular. Otherwise, I think we should continue the conversion of templates. —JohnC5 16:58, 8 July 2017 (UTC)
    I don't have sources. I barely know anything about Greek, I just started learning it a few months ago. But it does seem like we are missing some crucial information and until we find sources, it would be misleading to continue with the template conversion. Even if the old template was simply wrong, which I doubt, we should find sources to back that up. Otherwise we should disable all our post-Koine pronunciations. --WikiTiki89 03:41, 9 July 2017 (UTC)
    So I think I've found the source of all this. Horrocks 2010 (§11.2.5) says:
    The deletion of nasals before fricatives, a process already in place in classical Greek before [s] and [z] (cf. σύ-στημα [sý-stɛːma] ‘system’, < *σύν-στημα [sýn-stɛːma]), was given greater scope with the shift of the voiceless aspirated plosives to fricatives. Voiced plosives, however, which in general also became fricatives, were retained after nasals in popular Greek (cf. 6.4 (19)), as sometimes reflected subsequently in the orthography, e.g. ἄντρας [ˈandras] for original ἄνδρας ‘man’. Thus the renewed onset of nasal deletion was effectively restricted to the context of a following voiceless fricative: e.g. νύφη [ˈnifi] ‘bride’, < νύμφη [ˈnimfi], etc.
    This says nothing about nasalization of the vowel, which again would be a phonetic difference, if extant. The scope is limited to deletion before voiceless fricatives without lengthening or nasalization. —JohnC5 05:15, 9 July 2017 (UTC)
    Well, I'm not sure why nasalization can't have been phonemic. Old Norse briefly had nasalized vowels after the loss or assimilation of nasals in certain environments (for instance, in Þórr and æsir according to this blog post), and at least some of them were phonemic before they merged with oral vowels in most North Germanic dialects. The same could have been true in Koine or Medieval Greek. But whether we have evidence for this, I don't know. — Eru·tuon 06:02, 9 July 2017 (UTC)
    @Angr: I think /nð/ was reintroduced in modern Greek by influence of the spelling. If you look at common words, their spelling has changed to accomodate the /(n)d/ pronunciation; e.g. άντρας (ántras), δέντρο (déntro). Similar case for other nasal clusters; e.g. νύφη (nýfi). --WikiTiki89 03:35, 9 July 2017 (UTC)

    @Wikitiki89, Angr, Erutuon: sorry to bother. We are approaching the completion of the conversion of the templates, and thus the issue has to be closed. Either someone implements this change (contingent on scholarly proof in favor of the change), or we throw it at as unsubstantiated. I've been hesitant to add the change because I am still unclear what the exact parameters of the change are, and I do not consider copying Wiktionary:Ancient Greek transliteration wholesale to be a viable option since it mixes phonemic and phonetic transcription freely and is not sourced. I have now looked through these books where I thought the relevant information would reside:

    • (Allen 1968) Vox Graeca-A Guide to the Pronunciation of Classical Greek (I actually used the find function to read ever passage using the word "nasal" in this entire book.)
    • (Bakker 2010) A Companion to the Ancient Greek Language
    • (Browning 1983) Medieval and Modern Greek
    • (Horrocks 2010) Greek-A History of the Language and its Speakers (again, I searched for every instance of "nasal".)

    Aside from the mention in Horrocks cited early, I've found nothing supporting the claim of deletion and nasalization in Byzantine Greek. There does appear to be agreement for -γμ- and -γν- resulting in /-ŋm-/ and /-ŋn-/, respectively, which Probert (in Bakker) says is phonemic. At this point, either we have to prove this claim and implement it, or ignore it. I'll also point out that, if we decide not to implement this now but later find evidence for the change, it's just a matter of changing the code.
    As for my changing σῠμφωνῐ́ᾱ (sumphōníā) back, Wiki, I hadn't noticed you'd restored the old template but instead thought I'd just failed to update it to the new one while editing the page. I was not intending to war with you. —JohnC5 01:57, 25 September 2017 (UTC)

    @JohnC5: Thanks for doing the research! If you've found nothing supporting the claim of deletion and nasalization in Byzantine, then I'd say our course of action is clear. —Aɴɢʀ (talk) 09:19, 25 September 2017 (UTC)
    @JohnC5, Angr, Erutuon: You guys are neglecting two important points:
    1. This isn't only about nasalization or deletion of nasals, but also about the effect of the nasal on the following consonant. Namely, the fact that voice stops did not become fricatives, and unvoiced unaspirated stops became voiced (merging with the voiced stops).
    2. Even regarding nasalization and/or deletion of nasals, we know for sure that this happened (based on the situation in Modern Greek, including Pontic Greek (Wikipedia mentions Pontic Greek νύφε (núphe) < Ancient Greek νύμφη (númphē)). The only question is when it happened. Horrocks 2010 seems to imply that this occurred at around the same time that the aspirated stops became fricatives.
    So there is enough evidence to suggest that our current implementation is wrong, but not enough evidence to enable us to fix it. The lack of scholarly material on this makes me question the entire rest of our reconstructions of post-classical Greek. Do we actually know when all these vowel shifts occurred, or are we again simply relying on the old template (which as John said is not sourced)? If we don't find evidence, I would then say we are not in a position to reconstruct post-classical Greek pronunciation and we should remove it entirely. --WikiTiki89 15:04, 25 September 2017 (UTC)
    @Wikitiki89: You make some very good points. I think we can say with certainty that nasals were deleted in the Byzantine period. I think we definitely can show the deletion of nasals and the changes to the ensuing stops. My question is about the nasalization of the preceding vowel can be shown. As for the vowels, Horrocks has very explicit description of the vowel shifts, if someone would like to read through that (I'm not super in the mood at the moment). But if we showed the deletion of the nasals, effects on the following consonants, and checked over the vowels, would that be sufficient? I'm not exactly sure what to say about the nasalization of the vowels at the moment.
    Also, does anyone know whether Gilgamesh responds to email? A great deal could be clarified if we knew what sources were used. —JohnC5 21:51, 25 September 2017 (UTC)
    @Wikitiki89, Angr, Erutuon, JohnC5: The entries for the inflected forms of ἀνήρ (anḗr) show the dropping of the nasal in Constantinopolitan Greek when followed by a voiced stop - a feature still present in Modern Greek - but what about the impact of nasal consonants on subsequent voiceless stops? Given the existence of a parallel phenomenon involving nasals and stops, it would seem that the cluster -ντ- in Κωνσταντινούπολις (Kōnstantinoúpolis), for example, should be voiced, since this is the pattern in the Modern language. I say this because a primary conclusion of Horrocks is that by the time of the Turkish conquest, Greek had phonologically already assumed its Modern form, more or less. As a lay observer, it would seem that if a voiced consonant cluster can first resist fricativization and then denasalize by the fifteenth century, an unvoiced nasal consonant cluster can become voiced, if only because it requires one less step. Indeed, I recall that this was how the module behaved until recently. Is the module not showing it voiced because the voicing was not phonemic? Thanks in advance for your answer and for all your hard work as well. Tectosax (talk) 06:27, 16 December 2017 (UTC)
    That's exactly what this thread is about. We're trying to fix the module to display everything, but it seems we are lacking enough scholarly sources that break up this phenomenon by period. Do you happen to have any? --WikiTiki89 18:36, 18 December 2017 (UTC)

    Nonspacing breve and macron[edit]

    The module used to accept the nonspacing breve and macron signs, but it doesn't anymore, resulting in thousands of pages in Category:Ancient Greek terms with incomplete pronunciation. Can the previous behavior please be restored? —Aɴɢʀ (talk) 12:01, 3 October 2017 (UTC)

    Fixed, I think. I had changed the tokenization function so that it did not automatically convert spacing characters to combining, because that was causing a coronis to disappear in transliteration. But that indirectly affected this module. — Eru·tuon 19:48, 3 October 2017 (UTC)

    Voiceless stops being aspirated before /s/[edit]

    @JohnC5, Erutuon Any verification on this? mellohi! (僕の乖離) 22:45, 12 January 2018 (UTC)

    @Hillcrest98: What's an example of this? I don't see any reason we would know unless there were evidence in manuscripts of alternations. —*i̯óh₁nC[5] 00:09, 13 January 2018 (UTC)
    ψῑλός --Per utramque cavernam (talk) 00:12, 13 January 2018 (UTC)
    @JohnC5: θρίξ (thríx), φρίξ (phríx), ξύω (xúō), basically every single word that contains a /ks/ and /ps/ cluster. mellohi! (僕の乖離) 00:20, 13 January 2018 (UTC)
    Oh, of course. Here's what Allen (Vox Graeca, p.57) says:
    In the pre-Eucleidean Attic alphabet they were written as χσ and φσ respectively (e.g. εδοχσεν, φσεφισμα), i.e. with aspirated first members; and this aspiration survives when, as occasionally, there is metathesis of the sounds (e.g. ε]υσχαμενοσ, σφυχ[ε).[1] It seems unlikely, however, that full aspiration was involved; in forms like γράψω, ἕξω from γραφ-, ἐχ- + -σω the grammarians in fact speak of loss of aspiration; and this is supported by the operation of Grassmann’s Law (see p. 13: e.g. original ἕξωἔχω, but not ἕξωἔξω) . Certainly there is no contrast between aspirate and non-aspirate in this position,[2] and any degree of aspiration that may have existed here can be ignored by the modern reader without any danger of confusion.
    This seem to argue against it generally. —*i̯óh₁nC[5] 00:48, 13 January 2018 (UTC)
    So we reach a question - should we let the module proper continue to aspirate /ks/ and /ps/ clusters or not? mellohi! (僕の乖離) 04:02, 13 January 2018 (UTC)
    @Hillcrest98: I don't think so. Showing them as aspirates is definitely the more extreme decision, and if we lack the evidence to support it, then we should avoid it. @Erutuon, do you agree? —*i̯óh₁nC[5] 19:14, 13 January 2018 (UTC)
    @JohnC5: I'm in favor of /ps ks/. There was no contrast in phonation of stops before /s/, and /p k/ seem the least marked of the three possible alternatives. — Eru·tuon 01:50, 14 January 2018 (UTC)
    @Erutuon: Great. How does this module work again? —*i̯óh₁nC[5] 02:35, 14 January 2018 (UTC)
    I'm also in favor of /ps ks/. —Mahāgaja (formerly Angr) · talk 17:20, 14 January 2018 (UTC)
    @JohnC5: All the rules are in Module:grc-pronunciation/data, and the consonant stuff uses tables to describe simple phonological rules. I'll look at changing it. — Eru·tuon 22:53, 14 January 2018 (UTC)
    @Erutuon: FYI, that was a joke. I just didn't want to do it. —*i̯óh₁nC[5] 02:57, 15 January 2018 (UTC)
    @JohnC5: Heh. It is somewhat tedious. — Eru·tuon 03:53, 15 January 2018 (UTC)

    Footnotes[edit]

    1. ^ They are also generally rendered in Armenian by kʿs, pʿs.
    2. ^ On e.g. ἐκ-σώζω, cf. L. Lupaş, SC, 8 (1966), p.9.

    missing pronunciation of ç[edit]

    The key for 15th CE Constantinopolitan pronunciation of ὀρχήστρα, for example, leads to https://en.wikipedia.org/wiki/Ancient_Greek_phonology, where there is of course no explanation of ç. It should instead be linked to https://en.wikipedia.org/wiki/Help:IPA/Greek. --Espoo (talk) 09:48, 20 February 2018 (UTC)

    We should create our own key instead of linking to Wikipedia. --WikiTiki89 19:27, 10 April 2018 (UTC)

    Related discussions[edit]

    I'm (temporarily) chaning Attic ζ from /sd/ to /zd/[edit]

    As long as we show assimilation of σδ (sd) as /zd/, it doesn't make sense to give ζ (z) as /sd/. --WikiTiki89 14:25, 12 April 2018 (UTC)