Why isn't intervocalic /ŋ/ analyzed as an onset in English?

Question

I think that sɪ.ŋɪŋ does not seem too unreasonable as a syllabification of the word singing, so I'm a bit puzzled why that option for the syllabification of intervocalic /ŋ/ seems to be dismissed in all of the literature about English syllabification that I have encountered so far. To me, the analogy with sitting sɪ.tɪŋ or driving draɪ.vɪŋ seems fairly compelling: I don't feel like the identity of the middle consonant affects how I divide the two syllables of words like this. An alternative position that seems reasonable to me is ambisyllabicity, or Wellsian syllabification based on morphology and coda capture by stressed syllables. (Or more abstractly, there are analyses where [ŋ] in words like sing(ing) is treated as derived from underlying /ng/, but right now I'm not asking about those.) But many analyses of English syllabification that I have seen do not accept ambisyllabicity or coda capture, but instead use the Maximal Onset Principle to derive syllabifications like sɪ.tɪŋ or draɪ.vɪŋ. I don't see why ng in words like singing should be an exception to the MOP.*

It's true that /ŋ/ does not occur word-initially, or after other consonants. But the phoneme /ʒ/ is also marginal in word-initial position, and I think that advocates of the syllabifications sitting sɪ.tɪŋ and driving draɪ.vɪŋ would accept the existence of onset /ʒ/ in intervocalic position in words like invasion ɪnˈveɪ.ʒən.

Also, I can think of a number of examples from languages other than English that seem to support the idea that restrictions on word-initial consonants may be stricter than restrictions on syllable-initial consonants in intervocalic position. In native vocabulary, German lacks word-initial /s/ and /x~ç/, Korean /l~ɾ/, Hebrew /f/, French /ɲ/. But my understanding is that in intervocalic position, these consonants occur and are analyzed as onsets rather than as codas in these languages. (Well, I've seen a proposal to analyze German intervocalic [s] as a "virtual geminate" that in some way occupies both a coda and onset position, but that seems a similar level of abstraction to analyzing [ŋ] as /ng/, so I'd set it aside for the purposes of this question.)

Admittedly, some of the "restrictions" I mentioned in the previous paragraph seem to be fairly violable: e.g., in German, word-initial [ç] seems to be well established in a few non-native words, and word-initial /s/ in some loans from English; in Korean, /l~ɾ/ seems to be possible word-initially in loans; and likewise in Hebrew, word-initial /f/ seems to occur in loans.

But I'm not sure that the English restriction against word-initial /ŋ/ is actually much stronger: it seems to me that instead, there just isn't a lot of opportunity for English speakers to have pressure to produce word-initial /ŋ/ in loanwords. The sound has no unambiguous spelling in English, and it doesn't occur at the start of words from the languages that English speakers are most likely to have heard spoken aloud.

Has my argument here been made before by anyone else? Have I missed some strong argument against /ŋ/ being an onset rather than a coda in intervocalic position?

---

*I don't find the Maximal Onset Principle completely compelling as a major principle of English syllabification, but it seems that many linguists do. A modified principle of avoiding empty onsets specifically seems a bit more plausible to me.

Answer 1

The first reason for [sɪŋ.ɪŋ] is the premise that [ŋ] only appears in the coda. The main argument for that conclusion is the analogy between word position and syllable position. Steriade has some discussion (reference not available at the moment) questioning syllables (attributing speaker behavioral such as intuitions to analogy to word-positions), and I agree that there are word-position limits on segments that don't reduce to syllable position. The restriction "does not appear after a consonant" is less obvious from an rule-observational perspective (i.e. I don't know of any convincing case of a limit saying "consonant X cannot appear after a consonant" which is not equally or better reduced to a consonant being coda-only). There is no formal problem with stating such a restriction, unless you have a specific theory of restrictions, for example "syllable structure is the only way to state distributional restrictions". In other words, the theoretical deck is stacked in a way that favors reducing distributional limits to syllable position.

The counterargument that [ʒ] is similarly limited is weak because [ʒ] is not sufficiently analogous: [ʒ] as a phoneme is (mostly) even more restricted, but the restrictions are also weaker. Restrictions on [ʒ] are weaker in that word-initial [ʒ] is rare and not non-existent. In contrast, [ŋ] is completely out and mostly un-utterable by English speakers. So we arrive at another theoretical debate: are rules of grammar responsible for statistical patterns, or only absolute patterns?

There is sufficient auditory input (in certain English-speaking areas) where names like Nguyen, Ngo are not rare, but speakers still adapt initial ŋ. Based on that and assuming (as I don't, necessarily) that absolute gaps have to be encoded in the grammar, I conclude that [ŋ] is in fact absolutely banned as an onset in English. Final ʒ (in contrast to final ŋ) is quite rare (barrage, garage, mirage, triage, camouflage, collage – the French [aʒ] words) and a respectable percentage of people utter these with [dʒ]. I think the correct generalization is that [ʒ] is an even more marginal phoneme, statistically peaking in intervocalic foot-medial position. In other words, distribution of [ʒ] is not grammatically restricted.

You present an interesting clash between two approaches. I do not see any hope for resolving this clash based on introspection, since coda-enthusiasts intuit just as strongly that ŋ in "singing" is in the coda. The only problem with having both of the two driving principles ("stick ŋ in the coda!" vs. "Gotta have an onset!") going on is if you insist that one of these has to be absolutely never disobeyed. Only UG-addicted rule-and-constraint theory has a problem with this. OT and regular rule-based phonology use basically the same mechanism (dominance, ordering) to generate the desired outcome.

There is no fact-based argument of English that argues for [sɪŋ.ɪŋ] vs. [sɪ.nɪŋ], or for [sɪ.ŋɪŋ] and [sɪ.nɪŋ]. I don't believe that there is such a thing as a purely factual argument for deciding between two essentially theoretical analyses. If we accept the existence of "coda-only (onset-only) consonant" and accept MOP, where there is an ordering of these two devices in a grammar, then [sɪŋ.ɪŋ] [sɪ.nɪŋ] is perfectly possible under one order, and [sɪ.ŋɪŋ] [sɪ.nɪŋ] is perfectly possible under the other order. Within a theory you might develop an analysis where one outcome is simpler to formalize, but in lieu of a high-level stipulation regarding interaction between coda-only vs. MOP (or similar) mechanisms, I conclude that both analyses are possible, and nothing decides between them.

Answer 2

You seem to be asking about what convention about syllable boundary would it make most sense to adopt. I am very uncomfortable with that approach, since it is more a question about linguists than it is about language. Can't we find a way of dealing with the truth of the matter rather than what convention is best?

You write, "To me, the analogy with sitting /sɪ.tɪŋ/ or driving /draɪ.vɪŋ/ seems fairly compelling." Analogy? Why use analogy? Why do you use slashes for your examples rather than square brackets? Is this a way of saying the actual facts don't count -- it is a question pitting convention against convention? What connection does this Maximal Onset Principle have with the facts of pronunciation?

As I understand the pronunciation facts, your syllabification [sɪ.tɪŋ], taken to be phonetic rather than phonemic, is just wrong for familiar dialects of American English. If that were correct, the [t] would be aspirated (strengthened) rather than flapped (weakened). Actually, it's [sɪt.ɪŋ]. But if you write slashes, I suppose that I can't make this criticism, because you're not trying to talk about facts. It's about something else. I don't know what.

Answer 3

This isn't really an answer, just a post to include some additional information that I find relevant I've come across since asking the question.

I brought up /ʒ/ in my original post, but the analogy seems to be worse than I initially thought. In fact, /ʒ/ does occur foot-initially for many speakers in word-medial context in the word luxurious, and potentially in caesura. So there is clear evidence for syllable-initial /ʒ/ of a kind that we don't have for syllable-initial /ŋ/.

/ŋ/ never occurs after another consonant in any position; I think it's not clear whether this is solely a matter of syllable structure, but explaining this in terms of a rule excluding /ŋ/ from the onset of any syllable does seem plausible. That said, Cŋ sequences are not only forbidden word-medially and word-initially, but also word-finally, even though English has coda clusters rm, rn, lm, and less commonly ln. This gap would not be explained by a syllable-based rule unless we assume that the final consonants in these sequences are the onset of a degenerate syllable (but that analysis raises the question of how to analyze the syllabification of the final /ŋ/ in words like sing, thing, building).

Evidence that the restricted distribution of /ŋ/ is not accidental

The restricted distribution of /ŋ/ is fairly clearly explained by its diachronic development from assimilation of /n/ to a following /g/ or /k/. This diachronic explanation neither excludes nor supports the possibility that synchronically, the restricted distribution is somewhat accidental. However, there does seem to be some evidence from studies that shows that English speakers are inclined to perceive the n-ŋ contrast better in the context #V_# than in the context #_V#, which indicates that the gap in ŋ-initial words (or ŋ-initial feet) is not just accidental ("Perception of synthetic nasal consonants in initial and final syllable position", by Larkey, Wald and Strange, 1978).

I found a comment on Reddit that says that an unnamed (possibly unpublished) study conducted by the author in 2011 found that monolingual English speakers front ŋ (presumably to /n/) in "foot-initial" position. On the other hand, a comment below that one says that some English speakers from New Zealand may use word-initial [ŋ] in placenames from Maori.