If natlangs were conlangs

[bradrn]

Ive seen a few people here question the analysis of Marshallese, particularly the vowels. Consider, for example, unless Im mistaken, that the standard analysis makes it impossible to ever have a single vowel .... it MUST be bimoraic and must also have either a consonant or a glide at each end. so e.g. there is no */ma/, it has to be /mʲaɰaʷ/ or whatever.

edit: OK not bimoraic, but Im leaving that part in because it seems that bimoraics occur much more often than one would expect.

I don’t think there’s anything wrong with the standard analysis. It makes good sense considering there are words like [nʲæ͡ɑɑ͡ætʲ], which is most easily analysed as having an underlying glide /nʲæɰætʲ/. And a restriction that every vowel must have a consonant before and after it is no crazier than any of the other phonotactic constraints out there (e.g. Arrernte, which has an underlying VC(C) syllable structure, but where phrases are often ended with /ə/, or Sakao with its undefined syllable structure).

[Nortaneous]

I can't claim to have an argument against the VVS analysis of Marshallese, but I don't understand why it's not analyzed as just having neutralizing [±front] and [±round] in certain contexts. Without detailed phonetic analysis I'm not convinced [nʲæ͡ɑɑ͡ætʲ] isn't just [nʲɑɑtʲ] -- of course palatalization affects vowels! (Cf. Austroasiatic, where final palatals are frequently realized with a [j] onglide, e.g. /ac/ [ajc].)

Diachronically, there are a lot of semivowels that apparently come out of nowhere, which is not very convincing. But speaking of Austroasiatic, there are some languages where stressed open syllables are prohibited. (English is almost such a language -- since the only exceptions are /ɑ o/, /ɑ/ is uncommon and was historically adjusted to /o/ sometimes, and there are places where /o/ is still diphthongized enough that you could maybe make a case for calling it /ɑw/, there are probably dialects that count.)

[KathTheDragon]

e.g. Arrernte, which has an underlying VC(C) syllable structure, but where phrases are often ended with /ə/

Supposedly. I don't find the argumentation required for this analysis very compelling.

[Travis B.]

(English is almost such a language -- since the only exceptions are /ɑ o/, /ɑ/ is uncommon and was historically adjusted to /o/ sometimes, and there are places where /o/ is still diphthongized enough that you could maybe make a case for calling it /ɑw/, there are probably dialects that count.)

That's only in your analysis where English lacks tense monophthongs other than /ɑ/ or /o/ (but in many dialects you can sufficiently argue that /o/ really is a diphthong, almost like my [ɑɔ] but more central and close, like that out in Maryland where I lived). Contrast that with the English here, whose only unambiguous diphthongs are /aɪ aʊ ɔɪ/, with even [eɪ oʊ] being mere optional allophones of /e o/, found mostly finally or prevocalically.

[bradrn]

I can't claim to have an argument against the VVS analysis of Marshallese, but I don't understand why it's not analyzed as just having neutralizing [±front] and [±round] in certain contexts. Without detailed phonetic analysis I'm not convinced [nʲæ͡ɑɑ͡ætʲ] isn't just [nʲɑɑtʲ] -- of course palatalization affects vowels! (Cf. Austroasiatic, where final palatals are frequently realized with a [j] onglide, e.g. /ac/ [ajc].) Diachronically, there are a lot of semivowels that apparently come out of nowhere, which is not very convincing.

Actually, I think I quite like this analysis! But it does have problems: for instance, what is the underlying vowel of a word like [tʲɛ͡ɔkʷ]? In your analysis, it could be either /ɛ/ or /ɔ/, but the VVS analysis simply states that it’s an open-mid vowel, the exact realisation of which depends on the context. As for the diachronics, I had thought that part of the reasoning behind the VVS analysis is that it works diachronically; I find it interesting that there’s no evidence for that!

e.g. Arrernte, which has an underlying VC(C) syllable structure, but where phrases are often ended with /ə/

Supposedly. I don't find the argumentation required for this analysis very compelling.

Why not? Quoting Wikipedia’s explanation:

The underlying syllable structure of Eastern/Central Arrernte is argued to be VC(C), with obligatory codas and no onsets. Underlying phrase-initial /ə/ is realised as zero, except before a rounded consonant where, by a rounding process of general applicability, it is realised as [ʊ]. It is also common for phrases to carry a final [ə] corresponding to no underlying segment.

Among the evidence for this analysis is that some suffixes have suppletive variants for monosyllabic and bisyllabic bases. Stems that appear monosyllabic and begin with a consonant in fact select the bisyllabic variant. Stress falls on the first nucleus preceded by a consonant, which by this analysis can be stated more uniformly as the second underlying syllable. And the frequentative is formed by reduplicating the final VC syllable of the verb stem; it does not include the final [ə].

I find this to be a pretty convincing argument — I’d be curious to know any issues you have with it.

[Xwtek]

Ive seen a few people here question the analysis of Marshallese

Ah, a language with emoji as the vowel.

... Hale (2000) paper that analyzed Marshallese with four vowel phonemes //, //, /☎/, /☯/?

[bradrn]

Ive seen a few people here question the analysis of Marshallese

Ah, a language with emoji as the vowel.

... Hale (2000) paper that analyzed Marshallese with four vowel phonemes //, //, /☎/, /☯/?

Yep, that’s the one! I personally prefer using the central vowel symbols /ɐ ɜ ɘ ɨ/, but with vowels as nebulous as those of Marshallese there’s really no good phonemic representation — hence the emoji.

[KathTheDragon]

Why not? Quoting Wikipedia’s explanation:

The underlying syllable structure of Eastern/Central Arrernte is argued to be VC(C), with obligatory codas and no onsets. Underlying phrase-initial /ə/ is realised as zero, except before a rounded consonant where, by a rounding process of general applicability, it is realised as [ʊ]. It is also common for phrases to carry a final [ə] corresponding to no underlying segment.

Among the evidence for this analysis is that some suffixes have suppletive variants for monosyllabic and bisyllabic bases. Stems that appear monosyllabic and begin with a consonant in fact select the bisyllabic variant. Stress falls on the first nucleus preceded by a consonant, which by this analysis can be stated more uniformly as the second underlying syllable. And the frequentative is formed by reduplicating the final VC syllable of the verb stem; it does not include the final [ə].

I find this to be a pretty convincing argument — I’d be curious to know any issues you have with it.

First and foremost, because spoken Arrernte actually has a normal syllable structure with onsets. All the other "arguments" in favour can be dismissed with alternative analyses, which I can't remember off the top of my head, so forgive me for that. If I find the paper that goes into detail on this, I'll share it.

[akam chinjir]

First and foremost, because spoken Arrernte actually has a normal syllable structure with onsets. All the other "arguments" in favour can be dismissed with alternative analyses, which I can't remember off the top of my head, so forgive me for that. If I find the paper that goes into detail on this, I'll share it.

I don't know if it's the paper you're thinking of, but section 2.1 of Paul Kiparsky, Formal and empirical issues in phonological typology gives a bunch of arguments against the VC analysis of Arrernte.

[Kuchigakatai]

Arrernte probably doesn't, but Yucatec Maya does have the restriction that every vowel must have a consonant before or after. This is easily seen in established Spanish borrowings, which get a final /ʔ/ or /h/ if they end in a vowel in Spanish: banda 'close-knit group of people' > Maya baandah /ˈɓàandah/ 'neighborhood'. And, as in German phonetically, word-initial vowels must be preceded by a glottal stop. Only short grammatical morphemes may end in a vowel, and these always precede a content word anyway.

Here is a sample from a Bible in Yucatec Maya:

11 Dios túuneʼ tu yaʼalajtiʼ: «¿Máax aʼaltech wa minaʼan a nookʼ? ¿Tsʼoʼok wa a jaantik u yich le cheʼ tin waʼalajtech maʼ a jaantikeʼ?». 12 Le máakoʼ tu núukaj: «Le koʼolel ta tsʼáajtenoʼ letiʼ tsʼáaten u yich le cheʼoʼ, le oʼolal tin jaantaj». 13 Jéeoba Dios túuneʼ tu yaʼalaj tiʼ le koʼoleloʼ: «¿Baʼax le tsʼoʼok a beetik beyaʼ?». Le koʼoleloʼ tu núukaj: «Letiʼe kaan tusenoʼ, le oʼolal tin jaantaj». (Genesis 3:11-3)

Two audio samples:
https://www.youtube.com/watch?v=z8eTwjQCCuA
Turn on close-captioning for a transcription from 0:10 onward, although I think it has some problems. I think he pronounces Maya as [ˈmàːjaʔ], but the transcription has "maaya". Amusingly, at 0:53 he pronounces Spanish actor /akˈtoɾ/ as [ˈʔaktoʔ] and at 1:19 video [biˈðeo] as [biðeˈóːʔ], with primary stress in the "wrong" location (from the point of view of Spanish). (Yucatec Maya doesn't have much of /r/ in native words except in dissimilations of /l/ from a nearby /l/, and otherwise just has it in borrowings from Spanish.)

https://www.youtube.com/watch?v=uunbH8NP1Xw
A video with transcriptions of recordings. Regarding the transcrition of the basic numbers, note that Victoria Bricker's dictionary transcribes the numbers 2, 3 and 7 as ka’ah [ˈkáʔah ~ ká̰ːh ~ káːh], ’óox [ʔóːʃ] and wúuk [wúːk].

[Nortaneous]

I can't claim to have an argument against the VVS analysis of Marshallese, but I don't understand why it's not analyzed as just having neutralizing [±front] and [±round] in certain contexts. Without detailed phonetic analysis I'm not convinced [nʲæ͡ɑɑ͡ætʲ] isn't just [nʲɑɑtʲ] -- of course palatalization affects vowels! (Cf. Austroasiatic, where final palatals are frequently realized with a [j] onglide, e.g. /ac/ [ajc].) Diachronically, there are a lot of semivowels that apparently come out of nowhere, which is not very convincing.

Actually, I think I quite like this analysis! But it does have problems: for instance, what is the underlying vowel of a word like [tʲɛ͡ɔkʷ]? In your analysis, it could be either /ɛ/ or /ɔ/, but the VVS analysis simply states that it’s an open-mid vowel, the exact realisation of which depends on the context. As for the diachronics, I had thought that part of the reasoning behind the VVS analysis is that it works diachronically; I find it interesting that there’s no evidence for that!

Right, it's neutralized. Wikipedia at least used to say Irish Gaelic had something similar in the short vowels. But when you have a word like pọpo /pʲawpʲɘw/, why shouldn't this just be /pʲɒpʲo/?

For another example, Marshallese /tʲ-aɰtʲɘj/ 'not know' < Proto-Micronesian *ata-i- 'know' (with tʲ- negative prefix), cf. Pohnpeian ɛsɛ 'know' and sɛ-ɛsɛ 'not know' as well as Kiribati ata-i 'to know', Kosraean etʌ 'know, understand', etc. Where are these not-even-semivowels coming from? Diachronically, nowhere, at least in this case. (Sometimes they do have diachronic sources; bizarrely, *f > j.)

Another issue is the distinction in the Marshallese dictionary between initial yiy-, yi'y-, and 'yiy-. At one point I think I saw the passage in one of Bender's papers where he explains what this means, but I can't find it now. Compare:
- /yiyakalep/, /yi'yakiyiw/
- /yiyewet/, /yi'yawat/
- /yi'yakʷey/, /yiyawiwwaw/
- /yi'yew/, /yiyẹw/
- /yi'yey/, /yiyẹy/, /'yiyey/

/'yiy/ seems to be spelled <ii>, so probably represents a full vowel... but then what are the other two? Could there be an [i- j-] distinction? (/yi'yakʷey/ is yokwe, so maybe that's [j], but then what's /yiy-/? There's also /yiyyV-/; how is this to be distinguished from /'yiy-/?)

[bradrn]

First and foremost, because spoken Arrernte actually has a normal syllable structure with onsets. All the other "arguments" in favour can be dismissed with alternative analyses, which I can't remember off the top of my head, so forgive me for that. If I find the paper that goes into detail on this, I'll share it.

I don't know if it's the paper you're thinking of, but section 2.1 of Paul Kiparsky, Formal and empirical issues in phonological typology gives a bunch of arguments against the VC analysis of Arrernte.

Interesting! I’m not sure I find those explanations entirely convincing — in particular, the VC(C) analysis solves all those issues with Arrernte with one explanation, whereas Kiparsky has to invoke several different principles to explain each individual issue in turn — but it certainly has convinced me that this is not at all a settled issue.

I can't claim to have an argument against the VVS analysis of Marshallese, but I don't understand why it's not analyzed as just having neutralizing [±front] and [±round] in certain contexts. Without detailed phonetic analysis I'm not convinced [nʲæ͡ɑɑ͡ætʲ] isn't just [nʲɑɑtʲ] -- of course palatalization affects vowels! (Cf. Austroasiatic, where final palatals are frequently realized with a [j] onglide, e.g. /ac/ [ajc].) Diachronically, there are a lot of semivowels that apparently come out of nowhere, which is not very convincing.

Actually, I think I quite like this analysis! But it does have problems: for instance, what is the underlying vowel of a word like [tʲɛ͡ɔkʷ]? In your analysis, it could be either /ɛ/ or /ɔ/, but the VVS analysis simply states that it’s an open-mid vowel, the exact realisation of which depends on the context. As for the diachronics, I had thought that part of the reasoning behind the VVS analysis is that it works diachronically; I find it interesting that there’s no evidence for that!

Right, it's neutralized. Wikipedia at least used to say Irish Gaelic had something similar in the short vowels. But when you have a word like pọpo /pʲawpʲɘw/, why shouldn't this just be /pʲɒpʲo/?

For another example, Marshallese /tʲ-aɰtʲɘj/ 'not know' < Proto-Micronesian *ata-i- 'know' (with tʲ- negative prefix), cf. Pohnpeian ɛsɛ 'know' and sɛ-ɛsɛ 'not know' as well as Kiribati ata-i 'to know', Kosraean etʌ 'know, understand', etc. Where are these not-even-semivowels coming from? Diachronically, nowhere, at least in this case. (Sometimes they do have diachronic sources; bizarrely, *f > j.)

I agree that those examples work very well with your analysis — but you never actually answered my question. In a word like [tʲɛ͡ɔkʷ], how would you know whether it is phonemically /tʲɔkʷ/, /tʲɛkʷ/ or /tʲɤkʷ/, or is there simply no way of knowing the underlying phonemic representation for this word?

Another issue is the distinction in the Marshallese dictionary between initial yiy-, yi'y-, and 'yiy-. At one point I think I saw the passage in one of Bender's papers where he explains what this means, but I can't find it now. Compare:
- /yiyakalep/, /yi'yakiyiw/
- /yiyewet/, /yi'yawat/
- /yi'yakʷey/, /yiyawiwwaw/
- /yi'yew/, /yiyẹw/
- /yi'yey/, /yiyẹy/, /'yiyey/

/'yiy/ seems to be spelled <ii>, so probably represents a full vowel... but then what are the other two? Could there be an [i- j-] distinction? (/yi'yakʷey/ is yokwe, so maybe that's [j], but then what's /yiy-/? There's also /yiyyV-/; how is this to be distinguished from /'yiy-/?)

Not sure what dictionary you’re using, but the MOD doesn’t use this notation. (It’s actually a fantastic resource for phonemic investigation as well, as it gives both the orthographic transcription and the VVS phonemic interpretation for every word.)

[Nortaneous]

Not sure what dictionary you’re using, but the MOD doesn’t use this notation.

it does

ie {yiyẹy}. Also ewi {yẹwiy}. where?; how much?; demonstrative, interrogative singular.
ie {yi'yey}. there (resumptive locative pronoun).
iie₁ {'yiyẹy}. needle.

[bradrn]

Not sure what dictionary you’re using, but the MOD doesn’t use this notation.

it does

ie {yiyẹy}. Also ewi {yẹwiy}. where?; how much?; demonstrative, interrogative singular.
ie {yi'yey}. there (resumptive locative pronoun).
iie₁ {'yiyẹy}. needle.

Oh, I hadn’t noticed that — sorry!

[bradrn]

Not sure what dictionary you’re using, but the MOD doesn’t use this notation.

it does

ie {yiyẹy}. Also ewi {yẹwiy}. where?; how much?; demonstrative, interrogative singular.
ie {yi'yey}. there (resumptive locative pronoun).
iie₁ {'yiyẹy}. needle.

Actually, you may be interested to know that I finally found an explanation of these sequences (and some other confusing ones) on Wikipedia:

• {yi'y} is for a "passing over lightly" version of the vowel allophone i that occurs at the beginning of certain words, phonetically pronounced [i̯] and existing on the phonemic level as /ji̯j/. For example, {yi'yakʷey} is equivalent to io̧kwe [i̯ɒɡʷɛ] /ji̯jækʷɛj/ "hello; goodbye; love".
• {'yiy} is for a "dwelling upon" version of i that occurs at the beginning of certain words, now generally written ii in the "new" orthography, phonetically pronounced [iː] and existing on the phonemic level as /jijj/, effectively making it identical to {yiyy}. An example is {'yiyayiyȩw}, which is equivalent to iiāio [iːæio] /jijjæjijew/ "reunion".
• {yiy} at the beginning of a word, without apostrophes, indicates a version of i whose reflex differs between the two dialects. In the Rālik dialect, this assumes the "dwelling upon" pronunciation, equivalent to {'yiy}. In the Ratak dialect, it instead assumes the "passing over lightly" pronunciation, equivalent to {yi'y}. An example is {yiyaļ}, equivalent to iaļ "road":
• …

[Nortaneous]

Actually, you may be interested to know that I finally found an explanation of these sequences (and some other confusing ones) on Wikipedia:

• {yi'y} is for a "passing over lightly" version of the vowel allophone i that occurs at the beginning of certain words, phonetically pronounced [i̯] and existing on the phonemic level as /ji̯j/. For example, {yi'yakʷey} is equivalent to io̧kwe [i̯ɒɡʷɛ] /ji̯jækʷɛj/ "hello; goodbye; love".
• {'yiy} is for a "dwelling upon" version of i that occurs at the beginning of certain words, now generally written ii in the "new" orthography, phonetically pronounced [iː] and existing on the phonemic level as /jijj/, effectively making it identical to {yiyy}. An example is {'yiyayiyȩw}, which is equivalent to iiāio [iːæio] /jijjæjijew/ "reunion".
• {yiy} at the beginning of a word, without apostrophes, indicates a version of i whose reflex differs between the two dialects. In the Rālik dialect, this assumes the "dwelling upon" pronunciation, equivalent to {'yiy}. In the Ratak dialect, it instead assumes the "passing over lightly" pronunciation, equivalent to {yi'y}. An example is {yiyaļ}, equivalent to iaļ "road":
• …

Ah, thanks, that sounds like what I remember.

If /yiyy/ is pronounced [i:], how is it distinguished from /yiyiy/? (cf. iaat /yahat/, aa- /haha-/, uuk /wiwik/) Does short i contrast with non-syllabic i word-initially?

Another bizarre thing in the dictionary is that uwaan kakūtōtō is given as /wiwahan kakitehteh/. Why not /kakitete/? /t/ is velarized anyway, so is there any way to tell the difference between /kakitehteh/ and /kakitete/?

[KathTheDragon]

I don't know if it's the paper you're thinking of, but section 2.1 of Paul Kiparsky, Formal and empirical issues in phonological typology gives a bunch of arguments against the VC analysis of Arrernte.

Ah, that's the one. Thanks.

Interesting! I’m not sure I find those explanations entirely convincing — in particular, the VC(C) analysis solves all those issues with Arrernte with one explanation, whereas Kiparsky has to invoke several different principles to explain each individual issue in turn — but it certainly has convinced me that this is not at all a settled issue.

Is it just one explanation, though? "It's underlyingly VC(C)" doesn't include the numerous ways of deriving the decidedly *not* VC(C) surface forms, which are largely independent hypotheses.

[bradrn]

If /yiyy/ is pronounced [i:], how is it distinguished from /yiyiy/?

I don’t think they are. Where do you see such a distinction?

(cf. iaat /yahat/, aa- /haha-/, uuk /wiwik/)

I think these are pronounced [æ͡ɑɑtˠ], [ɑː-], [uu͡ɯkˠ].

Does short i contrast with non-syllabic i word-initially?

The MOD has both /yi'yah/ and /'yiyah/, which seem to make a minimal pair.

Another bizarre thing in the dictionary is that uwaan kakūtōtō is given as /wiwahan kakitehteh/. Why not /kakitete/? /t/ is velarized anyway, so is there any way to tell the difference between /kakitehteh/ and /kakitete/?

I don’t know, actually. Maybe the /h/s were inserted to make all syllables conform to a CVC structure (which is something I don’t see as strictly necessary, as long as all vowels have a consonant of the correct secondary articulation before and after them).

Interesting! I’m not sure I find those explanations entirely convincing — in particular, the VC(C) analysis solves all those issues with Arrernte with one explanation, whereas Kiparsky has to invoke several different principles to explain each individual issue in turn — but it certainly has convinced me that this is not at all a settled issue.

Is it just one explanation, though? "It's underlyingly VC(C)" doesn't include the numerous ways of deriving the decidedly *not* VC(C) surface forms, which are largely independent hypotheses.

What numerous ways? As far as I’m aware, there’s only two rules:

• Initial /ə/ gets deleted (or turned into /ʊ/ before a rounded consonant)
• Phrases can get an extra /ə/ at the end corresponding to no underlying segment

My point is that those two rules, plus the VC(C) analysis, are sufficient to explain many of the difficulties with Arrernte at once, whereas an analysis like Kiparsky’s needs one explanation for stress, a separate one for morphophonemic alteration, yet another for play language etc.

(Of course, none of this precludes that Kiparsky could well be right — but I see the VC(C) analysis as more convincing for now.)

[Nortaneous]

If /yiyy/ is pronounced [i:], how is it distinguished from /yiyiy/?

I don’t think they are. Where do you see such a distinction?

There's no distinction, but Bender preserved the high-mid vowels, which only come from umlaut processes, because getting rid of them would make phonemicization of certain words not one-to-one. (e.g. [kɤk] could be either /kɨkɜ/ or /kɜkɨ/, or something like that, I forget the exact rules) If [i:] can be either /yiyy/ or a hypothetical /yiyiy/, the same objection applies. And if it's only /yiyy/ because /yiyiy/ is prohibited, why is /yiyiy/ prohibited? (There are examples of /yiyiC/ where C isn't a semivowel, e.g. iij /yiyij/.)

Does short i contrast with non-syllabic i word-initially?

The MOD has both /yi'yah/ and /'yiyah/, which seem to make a minimal pair.

Right, but it's orthographic long i. So if there's just a rule that word-initial prevocalic /yiy/ is realized as [j], then there wouldn't be a need for a true phonemic semivowel. But if, say, [jæ͡ɑ i.æ͡ɑ iː.æ͡ɑ] all contrast, you need /i̯/.

Another bizarre thing in the dictionary is that uwaan kakūtōtō is given as /wiwahan kakitehteh/. Why not /kakitete/? /t/ is velarized anyway, so is there any way to tell the difference between /kakitehteh/ and /kakitete/?

I don’t know, actually. Maybe the /h/s were inserted to make all syllables conform to a CVC structure (which is something I don’t see as strictly necessary, as long as all vowels have a consonant of the correct secondary articulation before and after them).

Could be that all Marshallese words have to be C-final and Bender (or whoever) interpreted it as having reduplication. But there's still AFAIK no principled way to tell between /kakiteteh/ and /kakitehteh/.

[bradrn]

If /yiyy/ is pronounced [i:], how is it distinguished from /yiyiy/?

I don’t think they are. Where do you see such a distinction?

There's no distinction, but Bender preserved the high-mid vowels, which only come from umlaut processes, because getting rid of them would make phonemicization of certain words not one-to-one. (e.g. [kɤk] could be either /kɨkɜ/ or /kɜkɨ/, or something like that, I forget the exact rules)

I didn’t know that! I had thought that /ɜ/ and /ɘ/ were completely separate vowels on the phonological level, so thanks for explaining!

If [i:] can be either /yiyy/ or a hypothetical /yiyiy/, the same objection applies. And if it's only /yiyy/ because /yiyiy/ is prohibited, why is /yiyiy/ prohibited? (There are examples of /yiyiC/ where C isn't a semivowel, e.g. iij /yiyij/.)

I’m not sure I understand this. Where do you see /yiyy/? And where did you see that /yiyiy/ is prohibited?

Does short i contrast with non-syllabic i word-initially?

The MOD has both /yi'yah/ and /'yiyah/, which seem to make a minimal pair.

Right, but it's orthographic long i. So if there's just a rule that word-initial prevocalic /yiy/ is realized as [j], then there wouldn't be a need for a true phonemic semivowel. But if, say, [jæ͡ɑ i.æ͡ɑ iː.æ͡ɑ] all contrast, you need /i̯/.

I don’t understand this either. What do you mean by ‘orthographic long i’?

Another bizarre thing in the dictionary is that uwaan kakūtōtō is given as /wiwahan kakitehteh/. Why not /kakitete/? /t/ is velarized anyway, so is there any way to tell the difference between /kakitehteh/ and /kakitete/?

I don’t know, actually. Maybe the /h/s were inserted to make all syllables conform to a CVC structure (which is something I don’t see as strictly necessary, as long as all vowels have a consonant of the correct secondary articulation before and after them).

Could be that all Marshallese words have to be C-final and Bender (or whoever) interpreted it as having reduplication. But there's still AFAIK no principled way to tell between /kakiteteh/ and /kakitehteh/.

I do agree with you here. I think that similarly to how your twelve-vowel analysis is ambiguous with words like [tʲɛ͡ɔkʷ], the VVS analysis is ambiguous with words like [kˠɑkˠɯtˠɤtˠɤ].