I’ve been reading Lynda Boudreault’s A Grammar of Sierra Popoluca, and the behaviour of glottal stops is both interesting and at first a little confusing. I think this is not helped by certain orthographic conventions and the organisation of the phonology chapter, but having gone through it and taken notes I think I’ve now got it straight. What’s still unclear to me, though, is to the extent to which glottal stops are normal consonants, compared to the alternative that they are a combination of epenthetic and suprasegmental features.

What follows is a summary of my understanding followed by some armchair speculation. I’m definitely not an expert on Sierra Popoluca.

The Basic Distributional Facts

Glottal stops are very common in Sierra Popoluca, and in the example texts probably the majority of words contain a glottal stop somewhere. They are subject to the following basic distributional constraints:

• They may only occur word-initially and directly after a vowel
• They may follow any vowel quality, and both long and short vowels.
• When they follow long vowels, the pronunciation often involves a broken vowel with glottal stricture in the middle. Such glottalised long vowels are written VʔV by convention, even though they are the nucleus of a single syllable and not two separate syllables.
• When they occur after a vowel, they may precede almost any other valid coda consonant sequence, with the exception of codas beginning with ts, s, and ŋ. They do not occur after another consonant, so they can only be the first member of an intervocalic cluster.
• A phonetic glottal stop occurs at the end of underlying vowel-final words, although these must be distinguished from words with underlying final glottal stops because of different phonological behaviour

The realisation of glottal stops may be by glottalisation of one or more vowels without complete glottal closure. For example, /pɔːpɔ=ʔaj/ may be pronounced as [pɔ̰ːba̰j̥] after application of certain rules.

Syllable and Word Shapes

There is some uncertainty about whether words may begin with vowels or not. It seems to be true that (almost?) all words which do not begin with another consonant begin with the glottal stop, although there is some uncertainty about whether these are phonemic or not. Boudreault analyses initial glottal stops as phonemic, although she mentions that Elson analyses glottal stop initial words as underlyingly vowel-initial.

Within the word, vowels do not occur in hiatus. Every syllable begins with at least one consonant, and almost always exactly one consonant (onset clusters are rare and extremely restricted). Syllables may be closed with codas of up to three consonants, but all codas of three consonants have the glottal stop as their first member. In general, any valid coda consonant sequence may be preceded by a glottal stop to make a longer valid coda, with the exception of codas starting with ts, s, and ŋ.

Both bound and free morphemes also almost always start with a consonant (assuming glottal stops are phonemic). Only four suffixes begin with a vowel; the rest begin with a consonant, with glottal stop being a fairly frequent option.

Resolution of Impermissible Glottal Clusters

Impermissible sequences are common because morphemes, including roots, prefixes, suffixes and clitics commonly begin with glottal stops and glottal stops may not follow another consonant. These impermissible clusters are resolved via a mixture of merger and metathesis. The basic rules are:

• If a glottal stop follows another stop, then it voices it and then normally deletes.

Underlying voiced stops are uncommon enough that Boudreault variably calls them restricted phonemes and denies that they are phonemes at all, but on the surface they are common because of this rule. An example would be /ta-tɔp-ʔaʔj-pa+ʔun/ [ta.tɔ.baʔj.pa.ʔun], where /pʔ/ becomes [b].

In the case where the stop is immediately preceded by a stressed vowel, then in addition to voicing the stop, the glottal resurfaces before it, as in /kəːpi-ʔaH-pa/ [kəːʔ.baːpʰ] (Note that in this example I have adjusted slightly the transcription of the long glottalised vowel to make the process clearer. The reason for the loss of the vowel /i/ is, however, not clear to me).

• If a glottal stop follows a nasal, then they are swapped so that the glottal stop becomes post-vocal.

For example, /ʔan-ʔaːpa/ [ʔaʔ.naː.paˀ].

• If a glottal precedes a fricative (which is not allowed), an echo vowel is normally inserted between them

It is unclear to me from the phonology section what happens when glottal stops follow longer clusters, or consonants which are neither stops, nasals or fricatives. The resolution of clusters such as /tsʔ/ and /jʔ/ is not described.

Other Processes (Not) Involving Glottal Stops

The following additional processes are worth noting:

• In the proclitics cluster before the verb, the rule ʔV → ∅ / CV _ C applies, i.e. ʔV deletes between a vowel and a consonant.
• There is a rule which lengthens short vowels in open stressed syllables, and a coda glottal stop is the only consonant which is invisible to this process. I.e. a syllable with only a glottal stop for a coda is treated as open for the purposes of stressed vowel lengthening.
• Similarly, primary stress is assigned partly by syllable weight, but glottal stops are invisible to this process. In this case, however, other coda consonants are also invisible (only nasals contribute weight).
• A /p/ following a stressed syllable ending in a glottal stop is voiced.

Thoughts

Glottal stops occur in three basic positions:

• Word initially
• Intervocalically
• As a coda / associated with the syllable nucleus

In word initial position they may be epenthetic or suprasegmental. The only explicit evidence I can see that Boudreault raises to prove that they are phonemic is that the same stop voicing rule applies across word boundaries, e.g. in /ʔan+na+ʔitʲ-W ʔidʲək/ [ʔaɾaʔidʲidʲək]. And yet… voicing of plosives between vowels is far from an uncommon process. Could a lot of these glottal stops be eliminated from the representation?

Hypothesis:

• The glottal stop is not a phoneme
• glottal stops are epenthetically inserted utterance initially and whenever vowel hiatus would result
• Separately, glottalisation is a suprasegmental feature of vowels / syllables

If this is true, then the stop voicing rule simply becomes that in most cases, a morpheme final stop is voiced if the following morpheme begins with a vowel, including when the morphemes belong to separate words.

The tight integration of glottal stops with the nucleus (they cannot be separated from their vowel by another consonant) and their floating nature (they are perceived as occurring mid-vowel to create interrupted / broken vowels when their vowel is long) is also explained by this. And it explains why glottal stops do not contribute weight for vowel lengthening and stress rules, and the tendency of the glottal stops to become spreading glottalisation without closure.

Some rules are problematic under this hypothesis though:

• Insertion of glottal stop at the end of vowel final words

A final vowel must be glottalised.

• Insertion of echo vowels before ts, s

Is it possible that for ts and s, this is actually a rule that glottalised vowels lengthen before fricatives and affricates? Remember that long glottalised vowels are perceived and written as broken, i.e. as two identical vowels with a glottal stop between them. English lengthened many vowels before fricatives.

• Deletion of ʔV within the verbal proclitic cluster

This becomes a rule that when vowels occur in hiatus within the proclitic cluster, the second vowel deletes.

• Metathesis of oral and glottal stops occurs after a stressed syllable

This one is hard to explain. If such glottal stops are not phonemic, I think it amounts to a rule that a monosyllabic root ending in a stop is glottalised if a vowel-initial suffix is added to it. There’s no obvious phonological explanation for this.

• Metathesis of nasals and glottal stops

This is another tricky one. It would have to become a rule that if a morpheme ending in a nasal is followed by a vowel-initial suffix, then the final vowel of the morpheme is glottalised.

There may be other difficult cases or arguments for the suprasegmental account I’m missing here, because of the lack of clarity about what happens in some more complex clusters involving glottal stops (in Boudreault’s analysis).

So this hypothesis may not quite work, but it seems clear that the glottal stop has a special status in Sierra Popoluca. It is frequent, mobile, and more rules seem to refer to it than any other phoneme.

Is This So Unusual?

Glottal stops seem extremely prone to this kind of ambiguous realisation and positioning. They often merge with vowels and/or consonants, producing outcomes such as gemination, glottalisation and laryngeal modification (both consonants and vowels), tone, etc. They also often move due to the extended cues of anticipatory and lagging glottalisation of surrounding vowels and consonants. According to Blevin’s Evolutionary Phonology:

“A common example of perceptual metathesis as sound change are historical inversions of h and ʔ adjacent to vowels. In many languages, Vh or hV sequences are produced with ambient aspiration or breathiness, while Vʔ and ʔV sequences can show ambient laryngealisation or creak. The problem of feature localisation arises in the course of language acquisition. If a learner attributes the long-domain feature of a laryngeal segment in a non-historical position, laryngeal metathesis has taken place. Laryngeal metathesis has occurred in a wide range of genetically unrelated languages including Arbore (Cushitic), Cayuga (Iroquoian), Bisayan (Austronesian), and Thompson River Salish (Salishan).”