Zompist Bboard Again

I'm feeling productive today, so here's a scratchpad. I've managed to avoid posting any of my conlangs over the past four years (discounting the relay) so it's probably about time I shared something – and no, there isn't something I've been working on for four years, there's just something I've been working on for precisely one day.

Contents

Pchekeho
Phonology

Mitsiefa, an overly minimalist language

• Some fella yapping on about universals
• Typological sketch
• Phonology - part 1
• Phonology - part 2

Ignore this, this is now a Mitsiefa thread. Pchekeho is relegated to the deepest recesses of my brain (i.e. not very deep, there's not very much room in there)

Overly thorough minimalism

As you may have divined from the phonology of Pchekeho, I like small phoneme inventories; the smaller the better. This got me wondering, what's the smallest I could go? In natlangs we seem to have a lower bound of ten phonemes, or just possibly nine. But is that a hard limit? It's very difficult to say. In this post I'm doing a fairly thorough analysis of phonological universals to hopefully produce the true holy grail of minimalism. Since I'm not bound by worries about correct citation or academic integrity, I can throw out previously-proposed universals much more easily than a real linguist, and I shall do exactly that (similarly unprofessionally, on the offchance that it irritates some passing academics, I'll be referring to all my sources on a first-name basis). Once I've produced for myself a set of absolute phonological universals, I will try to use them to produce a working, naturalistic language. In the spirit of this thing (whatever it is), I'll also give it the least amount of morphology possible, and probably the fewest grammatical categories possible, if that's sensible. In other words I'm making an anti-kitchen sink language. A toilet cistern language, if you will.

It's notoriously difficult to propose phonological universals, and it's only getting more so as new Papuan and South American (and occasionally African) languages. For this thought experiment I want to try and determine what the smallest possible phonemic inventory is that doesn't violate any phonological universal of natural languages. I'll be using mostly descriptive universals rather than "analytic" ones – if I say for instance, "all languages contrast consonants for place of articulation", I'm merely describing a feature of all known languages, not claiming that it's impossible for such a language to ever arise. However, to qualify they have to have absolute universals; it's no good saying that "all languages have coronal phonemes" since Northwest Mekeo doesn't have any, even though it's true for 99.99% of language. I'll also make the caveat that I can't refer to any numbers other than "multiple" – it's no good saying "all languages have at least 9 phonemes" because that would defeat the whole point of this, but it is acceptable to say that "all languages have multiple consonant phonemes". (NB I end up ignoring this rule.)

For my starting place I'll be using Larry Hyman's 2008 article Universals in Phonology which is still pretty definitive, even though languages like the Mekeos, Ontena Gadsup and Biritai have disproven some of its universals since. To supplement this, I'll be taking all uncontested phonemic inventories I know of in good faith – for instance I'll accept Jones' /p k g m ŋ w j β/ inventory for Northwest Mekeo even though it hasn't been independently verified. On the other hand, I won't accept Kabardian's zero-vowel analysis since there are newer analyses which are preferable. It's also worth noting that I'm only talking about natural spoken languages, even though most conlangs conform to these universals so they wouldn't have a great deal of effect anyway. Sign languages presumably have their own set of phonological universals too, but I don't know anywhere near enough about it (I don't know if anyone does to be honest).

It's obviously impossible to prove a universal, either descriptive or analytic; the former because we don't know all the languages which have ever existed, and the latter because it's impossible to prove anything in science. For a "universal" to qualify I will arbitrarily say that it must have no counter-examples in any known languages or dialects. I'll also give myself artistic license in vetoing universals which I think are stupid, with or without explanation.


1. General remarks

Firstly I'll discuss some very general, non-specific universals which are trivial but worth pointing out.

​ ​ ​ ​ ​ 1a. All spoken languages have phonemes

This is worth mentioning at least. All languages draw from groups of sounds (allophones) that can be defined as finite, mostly discrete sets (phonemes). I say "mostly discrete" here because there can be accoustic overlap especially in the vowel space. Also note that I mention only spoken languages, even though I believe sign languages still have something parallel to phonemes. It's hard to imagine what a language which didn't have phonemes would be like. Presumably there would be no limitations on word form or even manner of production whatsoever. I'm sure it would be possible to construct a language without any phonemes that uses only suprasegmentals to express lexical and grammatical ideas, but it certainly isn't reasonable for human languages.

An interesting one which I don't think has been discussed before is:

​ ​ ​ ​ ​ 1b. There are never fewer phones than phonemes in a phonological system

I'll define a phone as a segment which occurs frequently (i.e. in a conditioned environment or in non-nonce free variation) which is marked by a phonological difference that could form the basis of a phonemic distinction in another language. This means that English /t/ → [t tʰ] counts as two phones, because /t tʰ/ are different enough to be separate phonemes in e.g. Hindi, but /i/ → [i ɪ̝̟] still counts as one phone. It isn't impossible to produce a phonological system where the underlying form has an additional contrast, for instance:

/p t ʈ k/
/m n/
/s/

/i u a/

/ʈ/ → [p] / _u
/ʈ/ → [t] / _[a,#]
/ʈ/ → [k] /_i

However I'm certain that this never occurs in a language to the exclusion of any other allophony. I could go on and define probably dozens more inane universals like this but they won't provide very much interest so I won't.


2. Consonants

First off:

​ ​ ​ ​ ​ 2a. All spoken languages have multiple consonant phonemes

While some languages have been analysed without vowels, no language has ever been analysed without consonants. Most languages have between eight and a hundred consonants; it's obviously not going to be worthwhile looking at languages with enormous consonant inventories since they will inevitably have more complexity than small consonant inventories and won't provide possible counter-universal examples, so I'll start with the smallest ones. A reasonable number of languages have seven consonants; some of these include Pirahã (under Everett's slightly dubious analysis), Buin, Puinave, Sikaritai and West Mekeo. Six consonants is also attested, famously with Rotokas, but also for numerous Lakes Plain languages such as Iau, Kirikiri and Obokuitai, and even more obscurely for North Mekeo (and possibly also for East Mekeo which is in the process of losing /ʔ/). Five consonants has been proposed for proto-Lakes Plain which most likely had */p t k b d/; but this is a reconstructed protolanguage so it must be taken with a grain of salt. There are basically two questions as to the validity of this inventory; firstly whether */ɾ/ was a phoneme, and secondly whether */s/ was a phoneme; in both cases it seems likely that they were allophones (of */d/ and */t/ respectively), but it's really impossible to say without a whole lot more research. Finally there's the most obscure of all, which is the 5-consonant system proposed for Biritai in a talk by Mark Donohue. This would be a normal Lakes Plain inventory save for the lack of */k/, and complete unconditional loss of /k/ is attested, so I'll accept this inventory as true even though there is no thorough analysis of it anywhere. There's also a chance that some other Lakes Plain language lacks, for instance, a phonemic sibilant, and has an inventory of /b t d k ɸ/, so I don't think excluding five-consonant inventories is wise.
It would technically be possible for me to propose that ?"all languages have at least five consonant phonemes", but to begin with, that defeats the whole point of this exercise, and furthermore I'm not entirely sure that there hasn't ever been a language with four consonant phonemes. Consonant inventory size seems to be pretty normally distributed, and there must have been at least 100,000 languages spoken in human history, so the chances that one of them once ended up having four consonants may not be infinitessimal.
Larry Hyman came up with four universals surrounding consonant systems, three of which have been disproved in the past 16 years:

Larry wrote:Consonant Universal #1: Every phonological system has oral stops.
Consonant Universal #2: Every phonological system contrasts phonemes that are [–cont] (= stops) with phonemes that are specified with a different feature.
Consonant Universal #3: Every phonological system contrasts phonemes for place of articulation.
Consonant Universal #4: Every phonological system has coronal phonemes

#1 is disproven by Ontena Gadsup, which has the following absolutely nuts inventory:
The phonemes /ɸ s ɾ x/ surface as [p t d k] word-internally following /ʔ/, and /ns/ surfaces as [nt]. This can be contrasted with Akuna Gadsup (also shown above) in which lenition is incomplete and thus stops can be treated as underlying. No other series gets as close to stops as being universal; sonorants are absent from numerous languages and nasals from a fair handful, while fricatives are only really an areal feature of Afro-Eurasia and the Americas.
#2 is automatically countered as well, although it can be reformatted. Subjectively speaking, all consonant inventories have at least two lines:
Note that for Rotokas I'm taking /b d g/ as archi-phonemes; traditionally they're given as /β ɾ g/ but for our purposes it really doesn't matter. And Maxakalí has one element of one line offset slightly, so you could say that it needs three lines.
What this actually means is that all languages have some kind of "MOA" (manner/mode of articulation) contrast. This can almost be described as "all languages have an MOA contrast at multiple POAs", save for Obokuitai:
Here MOA is only contrasted at the coronal POA (unless you get into "central" vs. "peripheral" which IMO is too abstract an approach for this purpose). This means that systems like ?/p t k ʔ s n l/ or even ?/p t k s/ can't be ruled out at this stage, although I don't know of any languages other than Obokuitai (and probably other East Tariku languages) which have this restriction. As such we arrive at our first proper universal:

​ ​ ​ ​ ​ 2b. All consonant inventories have multiple degrees of sonority including multiple obstruents

This is very similar to Larry's #2, but it allows for Ontena Gadsup – for instance /ɾ/ is more sonorous than /s/. To my knowledge no language has only a single obstruent. Note that I'm working on a standard sonority hierarchy of "voiceless plosive < voiced plosive < voiceless fricative < voiced fricative < nasal < lateral < flap < approximant < high/close vowel < mid vowel < low/open vowel"; it's important to distinguish voicing in plosives since some languages limit their sonority differentiation to this. In fact, we can produce another universal stemming from #2b, this time with a bit more specification:

​ ​ ​ ​ ​ 2c. There are always multiple consonant phonemes which are more sonorous than the least sonorous series of phonemes

This covers all languages with sonority contrasts at multiple POAs, but also covers languages like Obokuitai, since all of /b d s h/ are more sonorous than /t k/. The same goes for Ontena Gadsup; /β j m n ɾ/ are all more sonorous than /ɸ s x/ (I'm not sure where /ʔ/ falls on the sonority hierarchy but it doesn't make a difference wherever it goes). #2b is very much a descriptive universal rather than a bottom-up one, but I feel like it will inevitably be true – I'm fairly sure we can rule out inventories like /p t ʈ c k kʷ q qʷ ʔ s/.

Larry's universal #3 still holds up pretty well, although I'll word it slightly differently:

​ ​ ​ ​ ​ 2d. There are always multiple contrastive places of articulation in a consonant inventory (?)

This means that we can rule out an inventory like /k b ɸ m l j/, but we can't automatically rule out /p k s m ʁ/. It might be possible to say that there are always multiple contrastive places of articulation at multiple levels of sonority, since for instance Obokuitai has /t : k/, /b : d/, /s : h/. However, a traditional analysis of Rotokas does stump this, as does North Mekeo described previously:
Regardless of whether this is the best analysis, I can't really reject it for any reason other than "it's not very neat", so I will leave #2d as it is. There's also a fair few near-misses like Karajá:
/h/ is not technically a fricative, so it could be separated from /θ/'s line, and /w/ is a semivowel rather than a tap, so you could tabulate it as (2), in which case it would only contrast POA at the sonority level of "voiced stop". Note how Karajá shows that POA does not always contrast at the least sonorous MOA (although it does tend to have the most degrees of contrast there) – we can't exclude ?/k m n ŋ/ on this basis. Take also North Mekeo or Ontena Gadsup (if /ʔ/ is taken to be least sonorous, which it probably is) above; in the former POA only contrasts in nasals.
I'm not entirely certain it's impossible for a consonant inventory to be entirely vertical. Say Karajá had /b t kʰ/ for its stops; you'd end up with just /kʰ t b ɗ θ h l ɾ w/ under a very strict analysis, which would violate this universal without looking overly bizarre. Likewise, a Mekeo dialect with /k b β m l w/ wouldn't seem wholly out of place. In this case the best you could do is say that "there will be multiple places of articulation in a consonant inventory which contrast either in the same sonority rank, or one differing only by phonation", which is far less pithy but possibly more absolute. In the end I don't think this has a great deal of influence on the form of an inventory.

Finally, Larry's #4 is blocked by Northwest Mekeo.
Interestingly, it's blocked only by NW Mekeo; to my knowledge there isn't a single other coronal-free consonant inventory. If it wasn't for this one Austronesian dialect, coronal phonemes could indeed be considered universal. In fact the reason they're missing from this language might be because of the basically arbitrary stigma against coronals as being infantile; Mekeo people replacing /k ŋ/ with [t n] will be told to stop acting like children (except for /ŋ/ → [n] / i_, i_).
Labials are fairly frequently missing, especially from North American languages, while a lack of velars is not unheard of. However, there aren't any languages which lack any two of these series, so we can modify this universal to state that:

​ ​ ​ ​ ​ 2e. Consonant phonemes will always occur at at least two out of labial, coronal and velar POAs.

Note that as per the discussion about Karajá and POAs above, I don't mention a contrast here. I don't think it would be impossible for a language to have an inventory like /b k ʔ n ɸ h l w j/, but there are always either {labial and coronal} or {labial and velar} or {coronal and velar} phonemes, so an inventory like /t tɕ ʔ s ɕ h n ɲ l j/ can be excluded. I don't think that this is just a random conflation of probabilities either, but rather a combination of a) a natural tendency for phonemes to spread out to fill the phonemic space and b) the lack of common sound changes which can remove these POAs without moving them into another POA. I guess you could have something like */p t k/ → */k t tɕ/ → /ʔ t tɕ/ but it seems vanishingly unlikely, more so than just say the 1% chance of not having bilabials combined with the 0.1% chance of not having velars. Even then we would only expect one language in human history to have lacked both series, so I am confident following 2e. (Cue me being proven wrong by South-central Klastafak from the Mamberamo river basin in a few years' time)

I do not believe anything more can be said universally about consonants, in part because of languages like Rotokas and Biritai which have so few phonemes. Any more universals would end up being rephrasings of the previous, or just uselessly conditional. You could technically say "all languages have either oral stops or two labial fricatives that contrast in voicing alone" because of Ontena Gadsup, but that's just petty and wouldn't add anything to our understanding of phonetics.


3. Vowels, or perhaps Nuclei

Vowels are at the same time a very fruitful field for phonological universals, yet also extremely frustrating. I'll be limiting myself to universals surrounding the minimal end of vowel inventories to save time and money (well, mostly just time). Vowel systems are fraught with competing analyses, with some languages being analysed as having one vowel by one linguist, and seven by another. Vowel systems also interplay a lot with consonant systems; it's naturalistic to have a vowel system of /ɨ a/, but not when combined with a consonant inventory like /p t k b d g/.

I'll start off by making a not wholly uncontroversial claim:

​ ​ ​ ​ ​ 3a. All languages have at least one vowel phoneme.

Several languages have been analysed to have zero vowels. Most famous of these is Kabardian, which Aert Kuipers claimed had no vowels whatsoever. Traditionally Kabardian can be thought of as having three vowels /ɨ ə a/ like some other Northwest Caucasian languages. However, the vowel [a] has some unusual distribution and stress properties, such that it can be non-trivially analysed as /hə/ (and [aː] as /əh/). Following this, /ɨ/ can be analysed as an epenthetic vowel, since (surface) CV syllables predominate. At this point the analysis is not untenable. However, it falters at the next point of Aert's analysis. For no particular reason, he then ascribes the height distinction between [ɨ] and [ə] onto consonants. This means that Kabardian would have to be analysed as having 97 consonants, all of which fell into [±high] pairs (other than /h/ I believe). For this reason I won't accept a zero vowel analysis, although I might accept one vowel (I do for other languages anyway). Zero vowels has also been claimed for a few other languages like pre-proto-Indo-European or even Mandarin. However, in both these cases there are syllabic resonants "/j̩ w̩/" which I do not believe are sensibly different from /i u/. I could rephrase the claim as "all languages have at least one vowel or syllabic resonant phoneme", but functionally speaking they are roughly the same.

Following on from this, Larry lists six vowel system universals:

Larry wrote:Vocalic Universal #1: Every phonological system contrasts at least two degrees of aperture.
Vocalic Universal #2: Every phonological system has at least one front vowel or the palatal glide /y/. [i.e. IPA /j/]
Vocalic Universal #3: Every phonological system has at least one unrounded vowel
Vocalic Universal #4: Every phonological system has at least one back vowel.
Vocalic Universal #5: A vowel system may be contrastive only for aperture only if its vowels acquire vowel color from neighboring consonants.
Vocalic Universal #6: A vowel system can be contrastive for nasality only if there are output nasal consonants.

I will have to reject his first universal on the basis of languages such as Kabardian in which a one-vowel analysis is acceptable. Another language which has been reasonably analysed as having only one vowel is the Chadic language Moloko. The only phonemic vowel is /a/; /ə/ is produced from epenthesis, while there is word-level prosody (notated as ᵒ and ᵉ) which determines the roundedness of all vowels in a word; hence

/mnzar/ → [mənzar] "see! (sg.)"
/mnzar-amᵒ/ → [mʊnzɔrɔm] "see! (pl.)"
/mnzar-aᵉ/ → [mɪmɪnʒɛrɛ] "seeing"

The problem with this "universal" is that occasionally vowel features are transferred onto consonants or entire words (the reverse can happen too, but not to this extent). It is possible to produce a fairly irritating but nevertheless valid universal:

​ ​ ​ ​ ​ 3b. All languages have multiple vowel phonemes unless consonants or words have markedness for F2.

Larry's second universal on the other hand does to my knowledge hold true, although I will be a bit more conservative in producing my 3c:

​ ​ ​ ​ ​ 3c. All languages have at least one [+front] vowel, or a [+front] consonant such as /j/.

His third universal is somewhat problematic to me. I don't exactly believe that a vertical vowel system of /ɨ ə a/ can be considered markedly [–round] (or [+unrounded]); they're simply unspecified for roundedness. His fourth universal falls down for the same reason; backness is equally unspecified. The fifth universal seems more promising, and is roughly equivalent to my 3b, but there's a problem when you run into Eastern Arrernte, which has /ə a/ where /ə/ is [ɪ~ə~ʊ] in pretty much free variation. Consonants have no [+front] feature, and while there is [+round] marking on numerous consonants, it doesn't actually affect the realisation of /ə/ except in very limited environments (specifically #_Cʷ). While this does technically conform to his universal, I'll stick with my wording of 3b above (which doesn't affect Eastern Arrernte anyway). I'll ignore his #6 because, as Larry himself notes, not all languages have nasality as a feature at all.

Worthy of mention is a fourth universal:

​ ​ ​ ​ ​ 3d. No language distinguishes frontness (F2) without also distinguishing height (F1).

This is needed to prevent systems like /i ɨ/ which would otherwise be permitted. Like Larry, I'm not treating /i u a/ systems as horizontal because I think that's stupid. Only one horizontal vowel system has been seriously proposed, which is proto-Indo-European's */e o/. First off, this is a protolanguage so immediately a poor counter-example. Additionally, there are several instances of */a/ which aren't satisfactorily explained through */h₂/. And finally, I believe, as do many other Indoeuropeanists that, */o/ was lower than */e/ and possibly longer also. In the end I'm conducting this train and I choose to ignore PIE, so there.

There's one other universal I can consider including, which would be something along the lines of "for a language to have fewer than three vowels it must have a large number of consonants". On the one hand, I believe this is true, and would produce a more naturalistic inventory, but on the other hand, it isn't worded very absolutely and uses numbers which I said I wouldn't. However, in the interests of producing something naturalistic, I will begrudgingly permit this one to remain:

​ ​ ​ ​ ​ 3e. No language has fewer than three vowel phonemes which does not also have more than ten consonant phonemes.

Ten is probably a very low limit here; twenty would probably work just as well, but I'm covering my bases here. It's the universal I'm least happy with but I can't say it's not true.


4. Anything else

Larry discusses a few more universal proposals. Some of them are too theoretical to be of much use - "No vowel can be [+high, +low]" is true, but it tells us more about the features [high] and [low] than it does about phonemes. He also acknowledges that there is little use or validity to the claim that "all languages have accent".
The idea of the syllable is of interest. There are three syllable universals that Larry discusses, as well as one he mentions in another paper, which I paraphrase below:

Paraphrasing Larry, I wrote:#1: All languages have syllables
2#: All languages have CV syllables
3#: Consonants and vowels always belong to a syllable
4#: Syllabification is always predictable

Conveniently, all of these rest on the validity of #1. If a language can be found with no syllables, then it obviously won't have CV syllables, and none of its consonants or vowels will belong to a syllable. On the other hand, a language without syllables would have predictable syllabification in that all of its phonemes would be not syllabified, but that universal is fairly tangential to my aim anyway.
To determine if a language has syllables, we first have to work out what a syllable actually is and what a language without them wouldn't have. Larry actually comes back with another paper to discuss this very thing; Does Gokana really have no syllables? Or: What's so great about being universal?. He posits four features which indicate the existence of syllables:

1. Phonological rules conditioned by syllable structure
2. Morphological rules or allomorphy conditioned by syllable structure
3. Prosodies or word-stress targeting the syllable as a feature-bearing unit
4. Prosodic grouping of syllables into higher order constituents, e.g. feet

These don't all need to co-occur; some languages use moras rather than syllables for prosody determination, and feet don't exist in monosyllabic languages. In fact the first two don't have to exist in a language even if it did have syllables. An isolating obligatory-onset CV language wouldn't follow the first two, and wouldn't necessarily have to follow the third or the fourth (in fact in such a case I'd have to wonder if there was any reason to posit sub-syllabic phonemes other than just convention).
The case of Gokana suggests does go against all of these tendencies to the point that syllables are fairly useless. The fundamental problem with this universal is that syllables don't really *exist* in the way phonemes exist. You can hear phonemes in a language, but you can't hear "." (hence why you can't distinguish [ˈsɪt.ɪŋ] from [ˈsɪ.tɪŋ] without introducing some acoustic feature like aspiration, glottalisation, vowel length, pause, etc.). Syllables are an abstract concept that basically just exist as a tool to explain the four things Larry describes above (and probably a few more things too). The universal "consonants and vowels always belong to a syllable" is really entirely nonsensical, since you can divide anything up into syllables.
Returning to the point at hand, the best we can say is that syllables aren't helpful in a given language. This is the case in Gokana. There's no point in trying to syllabify the sequence /kɛ̃̄ɛ̃̀ɛ̃̀ɛ̃̀ɛ̃̄ɛ̃́/ (wake-CAUS-LOG-3SG-FOC). You can do it (/kɛ̃̄ɛ̃̀ɛ̃̀ɛ̃̀ɛ̃̄ɛ̃́/ or /kɛ̃̄ɛ̃̀ɛ̃̀.ɛ̃̀ɛ̃̄ɛ̃́/ or /kɛ̃̄ɛ̃̀.ɛ̃̀ɛ̃̀.ɛ̃̄ɛ̃́/ or /kɛ̃̄.ɛ̃̀.ɛ̃̀.ɛ̃̀.ɛ̃̄.ɛ̃́/) but it doesn't add anything to the understanding of Gokana.

Gokana does however have moras and systematic root shapes. This suggests that you could say "all languages form phonemes into strings on the basis of sonority", but then you encounter difficulties with languages like Nuxálk and various Berber languages where obstruent sequences can occur practically ad infinitum.

Ultimately I don't think there's much more that can usefully be added. In the next post I'll provide the maximally minimal inventory and start working on grammar. Before I go, here's a list of the 12 universals I've come up with, with the actually useful ones in bold:

​ ​ ​ ​ ​ 1a. All spoken languages have phonemes
​ ​ ​ ​ ​ 1b. There are never fewer phones than phonemes in a phonological system
​ ​ ​ ​ ​ 2a. All spoken languages have multiple consonant phonemes
​ ​ ​ ​ ​ 2b. All consonant inventories have multiple degrees of sonority including multiple obstruents
​ ​ ​ ​ ​ 2c. There are always multiple consonant phonemes which are more sonorous than the least sonorous series of phonemes
​ ​ ​ ​ ​ 2d. There are always multiple contrastive places of articulation in a consonant inventory
​ ​ ​ ​ ​ 2e. Consonant phonemes will always occur at at least two out of labial, coronal and velar POAs.
​ ​ ​ ​ ​ 3a. All languages have at least one vowel phoneme.
​ ​ ​ ​ ​ 3b. All languages have multiple vowel phonemes unless consonants or words have markedness for F2.
​ ​ ​ ​ ​ 3c. All languages have at least one [+front] vowel, or a [+front] consonant such as /j/.
​ ​ ​ ​ ​ 3d. No language distinguishes frontness (F2) without also distinguishing height (F1).
​ ​ ​ ​ ​ 3e. No language has fewer than three vowel phonemes which does not also have more than ten consonant phonemes.

That was absolutely fascinating; thank you so much for writing it up! Though, I must admit, I was expecting to find some kind of hyper-minimal conlang at the end. (Maybe I’ll make one… I’m fond of such inventories too.)

Darren wrote: ↑Sun Jan 28, 2024 12:27 am ​ ​ ​ ​ ​ 3a. All languages have at least one vowel phoneme.

Several languages have been analysed to have zero vowels. Most famous of these is Kabardian, which Aert Kuipers claimed had no vowels whatsoever.

Mandara/Wandala has also been analysed with zero vowels. Nortaneous tracked the claim down to a German source, but I’m not sure anyone here actually got around to reading it to confirm.

(In general, Biu-Mandara is a great place to find small, weird vowel systems: you’ve already mentioned Moloko with its single underlying vowel. Wolff’s ‘Vocalogenesis’ essay is a good starting point. I did also find a book on it once, but frustratingly I’ve never been able to find it again.)

Darren wrote: ↑Sun Jan 28, 2024 12:27 am Larry actually comes back with another paper to discuss this very thing; Does Gokana really have no syllables? Or: What's so great about being universal?.

Let me link his postscript too, in case you haven’t seen it: http://linguistics.berkeley.edu/~hyman/ ... gokana.pdf

bradrn wrote: ↑Sun Jan 28, 2024 12:48 amI must admit, I was expecting to find some kind of hyper-minimal conlang at the end.

Oh, don't you worry, it's a-coming!

Darren wrote: ↑Sun Jan 28, 2024 12:27 am ​ ​ ​ ​ ​ 3a. All languages have at least one vowel phoneme.

Several languages have been analysed to have zero vowels. Most famous of these is Kabardian, which Aert Kuipers claimed had no vowels whatsoever.

Mandara/Wandala has also been analysed with zero vowels. Nortaneous tracked the claim down to a German source, but I’m not sure anyone here actually got around to reading it to confirm.

(In general, Biu-Mandara is a great place to find small, weird vowel systems: you’ve already mentioned Moloko with its single underlying vowel. Wolff’s ‘Vocalogenesis’ essay is a good starting point. I did also find a book on it once, but frustratingly I’ve never been able to find it again.)

I feel that zero-vowel systems might be possible, but I feel like they'd be better publicised if they were really convincing. I've seen Wolff's slideshow before – the only problem with it is that I couldn't access any of the relevant papers. For me to accept a zero-vowel system, there'd need to be some very good evidence that [i u a] are really /j w ʔ/ beyond just "oh look what I can do here" (the same goes for epenthesis arguments). Still, I'm not saying it's impossible. Regardless, for my purposes universals 3b and 3e (sadly the ugliest one) make up for it if I'm trying to get the smallest number of phonemes possible.

bradrn wrote: ↑Sun Jan 28, 2024 12:51 am

Darren wrote: ↑Sun Jan 28, 2024 12:27 am Larry actually comes back with another paper to discuss this very thing; Does Gokana really have no syllables? Or: What's so great about being universal?.

Let me link his postscript too, in case you haven’t seen it: http://linguistics.berkeley.edu/~hyman/ ... gokana.pdf

Ah thanks! I hadn't seen that before. It seems to just confirm that syllables are to an extent an artefact of analysis.

Darren wrote: ↑Sun Jan 28, 2024 1:08 am 3e (sadly the ugliest one)

Looking at that one again, I think you can merge it with 3b:

3b′. All languages have >2 vowel phonemes unless consonants or words have markedness for F2.

Because all the two-vowel systems I know of co-occur with secondary articulation. That being said, depending on how you analyse them, Ndu languages and Mandarin Chinese may be counterexamples.

bradrn wrote: ↑Sun Jan 28, 2024 1:34 am

Darren wrote: ↑Sun Jan 28, 2024 1:08 am 3e (sadly the ugliest one)

Looking at that one again, I think you can merge it with 3b:

3b′. All languages have >2 vowel phonemes unless consonants or words have markedness for F2.

Because all the two-vowel systems I know of co-occur with secondary articulation. That being said, depending on how you analyse them, Ndu languages and Mandarin Chinese may be counterexamples.

I think Mandarin is too strong of a counterexample there. I guess I could say

3b′′. All languages have >2 vowel phonemes unless a) consonants or words have markedness for F2, or b) resonant phonemes are permitted in nucleus position.

The only thing that would counter this would be a language with phonemic /i u/ and epenthetic [a] and I'm fairly sure that doesn't exist.

Darren wrote: ↑Sun Jan 28, 2024 1:41 am 3b′′. All languages have >2 vowel phonemes unless a) consonants or words have markedness for F2, or b) resonant phonemes are permitted in nucleus position.

The only thing that would counter this would be a language with phonemic /i u/ and epenthetic [a] and I'm fairly sure that doesn't exist.

Why that specific combination? Surely a language with a ‘normal‘ consonant inventory and a /ə a/ vowel system would also violate it?

bradrn wrote: ↑Sun Jan 28, 2024 1:49 am

Darren wrote: ↑Sun Jan 28, 2024 1:41 am 3b′′. All languages have >2 vowel phonemes unless a) consonants or words have markedness for F2, or b) resonant phonemes are permitted in nucleus position.

The only thing that would counter this would be a language with phonemic /i u/ and epenthetic [a] and I'm fairly sure that doesn't exist.

Why that specific combination? Surely a language with a ‘normal‘ consonant inventory and a /ə a/ vowel system would also violate it?

True, but I would be astonished if the latter existed and only fairly surprised if the former did.

Darren wrote: ↑Sun Jan 28, 2024 2:35 am

bradrn wrote: ↑Sun Jan 28, 2024 1:49 am

Darren wrote: ↑Sun Jan 28, 2024 1:41 am 3b′′. All languages have >2 vowel phonemes unless a) consonants or words have markedness for F2, or b) resonant phonemes are permitted in nucleus position.

The only thing that would counter this would be a language with phonemic /i u/ and epenthetic [a] and I'm fairly sure that doesn't exist.

Why that specific combination? Surely a language with a ‘normal‘ consonant inventory and a /ə a/ vowel system would also violate it?

True, but I would be astonished if the latter existed and only fairly surprised if the former did.

Fair enough!

I also wonder if the following universal is true:

All languages have multiple consonants which are less sonorous than the most sonorous series of consonants.

This would have major implications for inventory size. Currently something like
Is acceptable, but with this new universal you'd have to have at least four consonants such as
However, if 2d ("there are always multiple contrastive places of articulation in a consonant inventory") proves not to be true in place of 2d' ("there will be multiple places of articulation in a consonant inventory which contrast either in the same sonority rank, or one differing only by phonation"), then it would be entirely permissible to have
This would have both /b ɾ/ more sonorous than /t/, /b t/ less sonorous than /ɾ/, and /b t/ contrasting by POA. However, I don't believe there are any natlang violations of the original 2d, so I'm not super happy positing that. I think I'll have to include this one.

Darren wrote: ↑Sun Jan 28, 2024 12:27 am Finally there's the most obscure of all, which is the 5-consonant system proposed for Biritai in a talk by Mark Donohue. This would be a normal Lakes Plain inventory save for the lack of */k/, and complete unconditional loss of /k/ is attested, so I'll accept this inventory as true even though there is no thorough analysis of it anywhere.

Just thinking about this… I had a look at your Lakes Plains comparative wordlist, and I had no problem finding Biritai words like /kɛcia-ka/, /sudîaɣa/, /asaikɛdi/. Do these not contradict the given consonant inventory?

bradrn wrote: ↑Sun Jan 28, 2024 4:56 am

Darren wrote: ↑Sun Jan 28, 2024 12:27 am Finally there's the most obscure of all, which is the 5-consonant system proposed for Biritai in a talk by Mark Donohue. This would be a normal Lakes Plain inventory save for the lack of */k/, and complete unconditional loss of /k/ is attested, so I'll accept this inventory as true even though there is no thorough analysis of it anywhere.

Just thinking about this… I had a look at your Lakes Plains comparative wordlist, and I had no problem finding Biritai words like /kɛcia-ka/, /sudîaɣa/, /asaikɛdi/. Do these not contradict the given consonant inventory?

Yes it does, doesn't it. I don't know why Mark thinks there's no /k/. The wordlist though is old (ca. 1980s) and not very reliable (no associated fieldwork), and Mark was confident enough to publish his inventory (or at least put his name to it) twice, so I don't think the wordlist is sufficient counter-evidence. I mentioned it here a while ago when I first saw it, and Nort and I agreed it was kinda dubious but not enough to outright reject it. k → Ø is attested in PNG, and even within Lakes Plains (under some conditions I don't understand in Central Tariku and Iau especially). Biritai's also spoken right next to Iau, and there's the difference between <h> in the wordlist and /ɸ/ listed by Mark, which suggests maybe a different dialect was at play. Anyway, I don't think it actually influences any of the universals, so it's thankfully a moot point. Perhaps that's a good argument in favour of it being legit.

Putting it all together

The minimum size seems to be four consonants and three vowels. It's impossible to go any lower without violating "all languages have multiple consonants which are less sonorous than the most sonorous series of consonants". There's a very large (maybe infinite depending on your methods) number of acceptable 4C inventories, so I'm choosing one I happen to like:
I'll combine this with a Wichita-style three-vowel inventory:
For a total of seven phonemes. Theoretically this inventory has no features (other than just numeric smallness) which are not found in other natural languages. It's ended up looking quite a lot like Mekeo, which is not surprising; but it also feels vaguely Japanese from all the m's and g's and ts's and dz's. I may attempt to derive this diachronically – perhaps with some delectable Mekeo *l → /i/ shenanigans. Or even better, *d → *l → /i/.

---

Typological sketch

A lot of newer grammars start off with a short sketch grammar, which outlines the major features of the language without going into detail or providing examples. I'm going to do that here, so that when I later go through each aspect in more detail I won't have to keep explaining everything else. For now I'm calling the language Mitsiefa because it's got to be called something. Also when I drop out of in-world description into out-world description I'll use these helpful boxes.

Mitsiefa is an entirely isolating, verb-final, head-marking language which favours a topic-comment clause structure. Notable features include the small phoneme inventory of only seven segments (two fewer than the smallest previously posited), a lack of coronal phonemes, extensive use of serial verb compounds and lack of an overt distinction between subjects and objects.

PHONOLOGY
Mitsiefa has four consonant phonemes and three vowel phonemes. Only two places of articulation are contrasted for consonants, combined with a three-way sonority distinction (voiceless plosive, voiced plosive and voice-unspecified fricative). Vowels occur at three heights (high/close, mid and low/open), with two front vowels and one non-front vowel. These inventories are shown in tables 1 & 2 below:

Table 1: Mitsiefa consonants

Table 2: Mitsiefa vowels ​ ​ ​

	Bilabial	​ Velar​ ​ ​			​ ​ ​ Front​ ​ ​ ​	Non-front
Stop (-VC)		​ ​ ​ /k/		High	​ ​ ​ ​ ​ ​/i/
Stop (+VC)	​ ​​ ​ ​ /b/	​ ​ ​ /ɡ/		Mid	​ ​ ​ ​ ​ ​/e/
Fricative	​ ​​ ​ ​ /ɸ/			Low		​ ​ ​ ​ ​ ​/ɑ/

Notably absent are coronal and non-obstruent phonemes – both attested (in Northwest Mekeo and numerous Lakes Plain languages respectively) but the combination thereof is not. Rounded vowels are absent phonemically but present phonetically. Allophony is extensive as is typical of small consonant systems (cf. Pirahã, Kirikiri, Mekeo). Coronals [t͡s d͡ʑ] occur as allophones of velars following /i/ (/ɑiɡɑ/ → [ɔid͡ʑa]), and [s] is inserted into hiatus between /i/ and /ɑ/ (/⁠ɑɑɡebiɑi/ → [ɔaŋeβ̞isɑi]). Nasals are allophones of voiced stops in word-initial position or following word-initial /ɑ/ (/beiɑ/ → [meisɑ]), while /ɡi/ may optionally surface as [ni] rather than [ŋi]. /ɸ/ varies freely in voicing in all positions. /b k ɡ/ are lenided to [β̞ x~ħ ɡ~ɣ~ʕ~ʔ] following /e/ or /ɑ/. /ɑ/ is rounded to [ɔ] word-initially before a vowel or between two instances of /b/. /i/ varies freely with [u] and zero in initial position. Orthography is fairly phonetic; [t͡s d͡ʑ s] are written ⟨ts dz s⟩, [m n ŋ] are ⟨m ng ng⟩, [β̞ x~ħ ɡ~ɣ~ʕ~ʔ] are ⟨b h g⟩, [i u e ɑ ɔ] are ⟨i i e a o⟩ and all remaining /b k ɸ/ are ⟨b k f⟩.
Syllables are restricted to the form (C)V. There is no restriction on vowel hiatus; like vowels can appear in sequences of at least up to four such as /ebaaaa/ "fish sp." and unlike vowels appear in seemingly unlimited clusters (/ɑeɑɑiɑe/ "shutter"). There is no reliable method to syllabify these sequences other than that /iɑ/ must be heterosyllabic since it surfaces as [isɑ]. Arguably the concept of the syllable has little relevance for Mitsiefa; it is sufficient to say that words are composed of sequences of vowels and consonants, such that two consonants may not come into contact and a consonant may not occur at a right edge word boundary. There is no audible word-level stress or tone, although there is a sentence-level downdrift tendency.

WORD CLASSES
No Mitsiefa word classes show any concatenative or non-concatenative morphology; it is a perfectly isolating language. There are two closed word classes; pronouns, which contains three elements (mifaa "1SG", ea "1PL.EX" and meha "2 SG/PL/1PL.IN") and particles of which there are less than a dozen (e.g. mi "TOP"). The remaining word classes are open and consist of verbs, nouns and ideophones (exx. ebisafe "cook", mefae "proton", mebebebe "person sleeping"). Also of note are names of geographical locations, peoples, persons and languages. These are syntactically indistinguishable from the class of nouns. Verbs are distinguished by being invariably clause-final while nouns are able to head NPs. Ideophones are distinguished by being unable to appear in topic position. Some ideophones are also distinguished by being formally reduplicated or even retriplicated – note that the boundary between repeated elements is treated as a word edge, so lenition and coronalisation do not occur but nasalisation does, hence /gaki-gaki/ → [ŋaxiŋaxi] "joint popping noise" compared to /eɑkigɑ/ → [eɑxid͡ʑa] "be frosted, smoked (of glass)". Not all ideophones which have repeated syllables are reduplicated; e.g. mebebebe above cannot be split up into *mebe·bebe.

CLAUSE STRUCTURE
All Mitsiefa clauses contain at least a verb, unless isolated ideophones are considered clauses of their own. Verbs can occur either singly or multiply; in the latter case they produce various types of serial verb constructions. This can be optionally modified by a noun phrase and/or an adverbial phrase and/or an ideophone. This entire construction can be considered the "predicate", "verb phrase" or "comment". I refer to this as the comment, in opposition to the other optional element of a clause, the topic. The topic is the element of a sentence which is either the subject (in discourse rather than linguistic sense) of previous conversation, or an argument which is to be established as the future topic of conversation. The topic may consist of any part of speech or any phrase type, other than an ideophone – even particles can occur as topics, although this is a highly marked construction. Topics invariably appear clause-initially:

1.1	Ibea e amaai ifagafi ngefefa. ibea dog e AGT amaai man ifagafi own ngefefa bite "As for the dog, he bit his owner."

A topic may optionally be marked with the post-particle mi, which is always optional but very commonly used to mark topics which are not simple NPs. This also has the function of nominalising subordinate clauses:

1.2	Amaai meabeitse fitsa mi oaa e isa ka oeae. amaai man meabeitse battery fitsa eat mi TOP oaa poison fafibi lead.metal e AGT isa DEMONSTR ka ADV oeae give It was his eating that battery which gave him lead poisoning.

At this point I'm not quite sure how to cleft NPs with instrumental PP modifiers due to the INSTR/AGT syncretism; this sentence could also be interpreted as "it was his eating that battery which the lead-based poison gave to him".

Note that the concept of a topic is not related to the idea of subject or object. Any argument can occur in topic slot, including an oblique one. There is no overt marking of subject or object role, although there are two adpositions which mark agency; agentive/instrumental e and patientive ame, which tend to correspond to transitive subjects and objects respectively. They can be attached either to topics or non-topic NPs and tend only to be used when agency is not clear from context; for instance ibea amaai ifagafi ngefefa would normally be interpreted as "the dog bit his owner", while ibea amaai ifagafi e ngefefa would mean "as for the dog, he was bitten by his owner." However, the latter meaning would be interpreted even without e in the context of for instance a story about an insanse man who goes around biting things.

It is possible for a clause to lack a topic, consisting solely of a comment, i.e. a verb ± adverbial phrase(s) and/or ideophone(s). It is not possible however to have a comment NP without a topic (traditional intransitive "subjects" must be topicalised, or there must be an adverbial topic). Non-topic phrases are used for impersonal constructions such as description of the weather; a particularly common device is the use of an ideophone + the verb eeea "speak" (with or without an adverbial phrase) to describe a sound:

1.3	Fahahaha ibefa e eeea. fahahaha falling.object.crash ibefa noise e INSTR eeea speak "There was a loud crash."

Note that if there is an adverbial clause at the left edge of a clause (e.g. Ibefa e fahahaha e eeea), it is not possible to tell syntactically whether or not this is a topic, and any semantic distinction is slight; I do not see any use in arbitrarily analysing this as one or the other.

THE VERB COMPLEX
The comment is invariably headed, and ended, by one or multiple verbs in what can be referred to as a verb complex. If there are multiple verbs, then the complex is described as a serial verb construction (SVC), of which there are two types; headed and distributed.
Headed SVCs have one semantic head verb which is modified by one or multiple light verb roots which contribute various TAM information to the head verb. These constructions tend to be lexicalised, although in no cases does the light verb not also occur elsewhere as a single verb; for instance kiibe can be used as a light verb to mark gnomic aspect, but also independently means "to own". These senses can co-occur:

1.4	Feeie isa Eaebaa kiibe kiibe. feeie land isa DEMONSTR eaebaa PN kiibe own kiibe own "(It is a fact that) this land is owned by the Eaebaa people."

Note that light verbs may themselves be distributed SVCs for added emphasis, for instance megafa oeka (strike begin) "to jump-start" can be used to mark very rapid inchoative actions. However, the head of a headed SVC may not be a distributed SVC (or for that matter a headed SVC) itself.
Distributed SVCs consist of multiple verbs each with equal head status, in a theoretically unlimited sequence, although in practice SVCs of more than five verbs are very rare. These SVCs can describe sequential or simultaneous events, but can also be used as a derivational method as with megafa oeka above.
Headed and distributed SVCs are clearly distinguished semantically, but syntactically are identical in isolated clauses. However, in tail-head constructions (where two sentences are linked by the repetition of the head verb of the previous clause), only the head of a headed SVC is repeated, while the entire of a distributed SVC is repeated (supporting the analysis of the former as headed and the latter as pluriheaded). In 1.5 below, midzisa faba is a headed SVC with midzisa marking completive aspect, whereas efa ngafe is a distributed SVC.

1.5

Maaidzaba i angega isa oee iisafa mieea ii oama mae mifaa midzisa faba. Faba i, maaidzaba oaisaei, angiea amaai maidzaba angaiefa·angaiefa efa ngafe. Efa ngafe i, amaai eabeiea ii mifaa itsa e faba ngebagi. maaidzaba shop i LOC angega uncle isa DEMONSTR oee BEN iisafa weight.unit mieea flour ii GEN oama obtain mae PURP mifaa 1SG midzisa finish faba walk faba walk i T.H maaidzaba shop oaisaei be.closed angiea and amaai man maaidzaba shop angaiefa·angaiefa snoring.noise efa sleep ngafe snore efa sleep ngafe snore i T.H amaai man eabeiea never ii GEN mifaa 1SG itsa eye e INSTR faba walk ngebagi awaken "I had gone to the shops to get his uncle a pound of flour. (I went, and) the shop was closed, and the shopkeeper was sleeping and snoring . (He slept and snored, and) I saw he was not going to wake up."

On a tangential note, I think emojis are a fairly good way of translating ideophones. Well, not a good way since it makes me look like a brainless millenial blogger, but an effective way at least.

NOUN PHRASES
Noun phrases are fairly consistently head-initial. Modifiers of noun heads may be either other nouns, prepositional phrases, or relative clauses. Relative clauses require no explicit introduction; the relevant element can simply be gapped – no rearrangement is required, since the head noun is always the topic role of the relative clause:

1.6	Amaai meha isaba mi mifaa afi. amaai man meha 2 isaba discuss mi TOP mifaa 1SG afi see "I saw the man you spoke of."

Note that this sentence could also mean "I saw the man who spoke of you". For the former meaning, the particle e would be required for disambiguation (amaai meha e isaba mi mifaa afi); for the latter, gapping is still an option, leaving e "floating" at the start of the relative clause without a head:

1.7a	Amaai e meha isaba mi mifaa afi. amaai man ⌈(amaai) ⌊(man) e AGT meha 2 isaba ⌉ discuss⌋ mi TOP mifaa 1SG afi see "I saw the man who spoke of you."

There is some ambiguity here since this construction can also be parsed as a complement clause:

1.7b	Amaai e meha isaba mi mifaa afi. ⌈amaai ⌊man e AGT meha 2 isaba ⌉ discuss⌋ mi TOP mifaa 1SG afi see "I saw that the man spoke of you."

This can be disambiguated through pauses (1.7a amaai | e meha isaba | mi mifaa afi vs. 1.7b amaai e meha isaba mi | mifaa afi). Alternatively gapping may be foregone – either amaai can be repeated, or the demonstrative particle isa can be used (i.e. a gapping of amaai ({amaai isa} e meha isaba "I saw the man you spoke of this man"), or the second person pronoun meha can be used as a surrogate third person pronoun (1.8).

1.8	Amaai meha e meha isaba mi mifaa afi. amaai man ⌈meha ⌊2=3 e AGT meha 2 isaba ⌉ discuss⌋ mi TOP mifaa 1SG afi see "I saw that the man spoke to you."

POSSESSIVE PHRASES
There are two distinct types of possessive phrases; alienable and inalienable. The former are simple NPs with a PP modifier:

1.9	Maeeidzie mitsaboba ii. maeeidzie dirt mitsaboba worm ii GEN "The worm's dirt."

Inalienable possession describes body parts, kinship terms, and optionally some man-made objects such as pillows and houses and socks. These phrases appear to be head-final since they are of the form "possessor + possessee", but syntactically speaking it is the possessor which is the head, and thus the only element required for anaphor:

1.10	Mifaa ee kafi. Mifaa meha e faba ifeaisaga. mifaa 1SG ee brother kafi. exist. mifaa 1sg meha 2 e AGT faba walk ifeaisaga encounter "This is my brother. You will be seeing a lot of him." lit. "... You will be seeing a lot of 1SG."

These constructions are somewhat ambiguous and may require repetition of the syntactic modifier to prevent misparsing as "you will be seeing a lot of me."

I'm not sure if this is attested but I like it so I'm keeping it anyway. Come to think of it, it can also be used to describe relative clauses with third person meha; i.e. amaai meha e meha isaba mi mifaa afi is really amaai (meha amaai e meha isaba) mi mifaa afi "I saw the man that (you discussed your man)". If that makes any sense.

I like this!

TomHChappell wrote: ↑Sat Feb 03, 2024 2:27 pm I like this!

Anything in particular you like about it?

Darren wrote: ↑Sun Feb 04, 2024 2:14 am

TomHChappell wrote: ↑Sat Feb 03, 2024 2:27 pm I like this!

Anything in particular you like about it?

I tried to think how to answer; but that would have been so tl;dr, that, I think I had better answer just “Everything”!
It’s complete, understandable, and interesting.

2. Phonology (part 1)

Mitsiefa phonology has been made as minimal as possible based on my own understanding – necessarily limited – of the phonological universals which can be gleaned from natural languages. The most relevant ones of these are that: (a) all languages have a sonority contrast in their consonant inventory, (b) all languages have multiple consonants which are more sonorous than their least sonorous consonant series, (c) all languages have multiple consonants which are less sonorous than their most sonorous consonant series, and (d) all languages have multiple contrastive POAs. The combination of (a), (b) and (c) basically means that a consonant inventory must have at least two "rows", and if it has only two then there must be multiple consonants in both of them. An inventory like */p t k s/ does not seem to be possible. Statement (d) means that a consonant inventory must have at least two "columns" filled in one "row". Something like /k b s ɣ m w/ probably isn't possible, and /t d s z n ɹ/ definitely isn't. Of course, this could just be a statistical coincidence, but I do believe there to be some reason behind this – if a language has very few consonants, then a reasonably use of both sonority- and POA-contrast will be made. Small inventories thus tend to be fairly square-ish: Code: Select all N. Mekeo Obokuitai Pirahã Rotokas Buin Puinave 2*3 3*3 4*3 3*2 3*4 3*3 b k t k p t k ʔ p t k p t k p t k m ŋ b d b g β ɾ g g s h β ɫ s h s h m n m n r These constraints mean that Mitsiefa must have at least four consonants. Three vowels are also necessary, although by a much simpler rule – (e) all languages with consonant inventories of less than 10 segments must have three or more vowels. Seven phonemes thus seems to be a hard limit (coincidentally also the number of phonemes in Solresol). The actual values of these four consonants and three vowels are partly determined by a number of other universals I've gone over in previous post, but in the end it's largely down to personal taste. Mitsiefa has /b k g ɸ i e ɑ/; Equally permitted would be something like /t k s n e o a/ or /p t m n i u e/, any of which would give the language a completely different look. I've attempted to come up with a phonology that is at least somewhat reasonable and wouldn't look entirely out of place on Earth.

Mitsiefa has seven phonemes in total; four consonants and three vowels. This is the smallest inventory that has been posited for any language, with the previous record holders being the form of Pirahã (isolate, Amazon) spoken by women, which has been claimed to have nine phonemes (although this analysis has been disputed). Both consonant and vowel systems fall well within WALS's definition of "small" (14 for consonants and 4 for vowels), and the consonant inventory is also the smallest posited – previously Biritai (Lakes Plain, Papua) has been claimed to have five consonants, as has its reconstructed ancestor Proto-Lakes Plain. As expected from such a small inventory, allophony is rampant; there are 23 distinct phones encountered in normal speech, an average of 3.3 per phoneme. Allophonic processes produce nasals, coronals, fricatives and back vowels not found phonemically in the language. In this section I will also discuss the typological implications of Mitsiefa phonology (§2.3.6), the formation of words (§2.4), sentence-level intonation patterns (§2.5) and the social influences which may have caused the development of such a typologically marked language (§2.6).


2.1 Phone inventory

I will begin by showing the phones encountered in Mitsiefa in tables 2.1 and 2.2 below, alongside the orthography where it differs from the IPA. The (linguistic) orthography used only reflects broad allophonic processes which cause major changes in pronunciation, hence why numerous phones share the same grapheme.

Table 2.1: Mitsiefa consonant phones ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​​ ​ ​

	​ ​ Bilabial​ ​ ​	​ ​ Coronal​ ​	​ ​ ​ ​ ​Velar​ ​ ​ ​​ ​ ​​	Laryngeal
Stop (-VC)		​ ​ ​ ​ ​t͡s ​ ​ts	​ ​ ​ ​ ​ ​ ​ ​k	​ ​ ​ ​ ​ ʔ ​ ​g
Stop (+VC)	​ ​ ​ ​ ​ ​ ​ ​b	​ ​ ​d͡ʑ ​ ​dz	​ ​ ​ ​ ​ g ​ ​g
Fricative (-VC)	​ ​ ​ ​ ​ ɸ ​ ​f	​ ​ ​ ​ ​ ​ ​ ​s	​ ​ ​ ​ ​ ​x ​ ​h	​ ​ ​ ​ ​ ​ħ ​ ​h
Fricative (+VC)	​ ​ ​ ​ ​ β ​ ​f		​ ​ ​ ​ ​ ɣ ​ ​g
Nasal (+VC)	​ ​ ​ ​ ​ ​ ​ ​m	​ ​ ​ ​n ​ ​ng	​ ​ ​ ​ ​ŋ ​ ​ng
Approximant (+VC)	​ ​ ​ ​ ​ β̞ ​ ​b			​ ​ ​ ​ ​ ​ʕ ​ ​g

Table 2.2: Mitsiefa vowel phones ​

	​​ ​ Front​ ​ ​	​ Central​ ​	​ ​ Back​ ​ ​
High	​ ​ ​​ ​ ​ ​​ i		​ ​ ​ u ​ ​i
Mid	​ ​ ​ ​ ​ ​ e		​ ​ ​ ɔ ​ ​o
Low		​​ ​ ​ ​ ​ a	​ ​ ​ ɑ ​ ​a

Consonant phones appear at four places of articulation; bilabial, coronal, velar and laryngeal. The labels "bilabial" and "velar" are precise, while "coronal" phones include the alveolars [t͡s s n] and the palato-alveolar [d͡ʑ], and "laryngeals" include glottal [ʔ] and pharyngeal [ħ ʕ]. Six levels of sonority are distinguished, with the most sonorous being the approximants [β̞ ʕ], followed by the nasals [m n ŋ], the voiced [β ɣ] and voiceless [ɸ s x ħ] fricatives, and the voiced [b d͡ʑ g] and voiceless [t͡s k ʔ] stops. The only notable gap in this phonetic inventory is that of plain coronal plosives, with only affricates appearing in their place – this situation is also the case in Northwest Mekeo where the only non-marginal coronal obstruent phone is [d͡z]. Plain [t d] only occur in wholly unassimilated borrowings and in imitation of foreigners' speech (see §2.6 for further discussion).

Vowel phones display three degrees of frontness, although only two at any one height – arguably [a] could be grouped with [i e] as a "front" vowel, in which case there would only be two degrees. Three degrees of openness (also a feature of sonority) occur. There are no pairs differing by rounding only; the front and central vowels [i e a] are unrounded, as is low back [ɑ], while high and mid back vowels [u ɔ] are rounded. This follows the universal tendency that front vowels will occur unrounded, and back vowels rounded – other than [ɑ] which is at a height where frontness (roughly equivalent to the second formant F2) is considerably less contrastive.

2.2 Allophonic processes

Allophony is extensive in both consonant and vowel inventories, although more so in the former. There are five distinct processes which can be identified; coronalisation, nasalisation, lenition, rounding and fronting.

2.2.1 Coronalisation
All coronal phonemes appear only in the vicinity of [i]. [n] is a free variant of [ŋ] before [i], as will be discussed in §2.2.2. [t͡s d͡ʑ] are allophones of /k ɡ/ following said vowel:

/ibike/ → [ibit͡se]
/biɡeɑe/ → [mid͡ʑeae]

[s] appears only in the sequence [isɑ], while in contrast [iɑ] does not occur, despite all other vowel sequences being frequent. This suggests that surface [s] can be viewed as the result of a rule "∅ → [s] / i _ ɑ". This presumably reflects earlier *[ija] → *[iʒa] → *[iza] or some similar development (dialects of the Austronesian language Mekeo show a similar rule, inserting variously one of [j ʒ ʃ z d s] between /i/ and /a/).

/eɸiɑbɑe/ → [eɸisɑβ̞ɑe]
/ɸeɡiɑ/ → [ɸeɣisa]

These processes are somewhat obscured by the optional rounding (see §2.2.5) or loss of initial /i/, which can produce apparent phonemic word-initial [s t͡s d͡ʑ]:

/iɑɡɑe/ → [isɑɣae], [usɑɣae], [sɑɣae]
/ikɑkee/ → [it͡sɑxee], [ut͡sɑxee], [t͡sɑxee]
/iɡe/ → [id͡ʑe], [ud͡ʑe], [d͡ʑe]

However, since there is free variation with underlying [i] in these cases, it can be confidently stated that this is an allophonic process. Initial /i/ can be seen to elide before other (i.e. bilabial) consonants also, although it is not so common:

/iɸɑ/ → [iɸɑ], [uɸɑ], [ɸɑ]
/ibeɑɑɸi/ → [ibeaɑɸi], [ubeaɑɸi], [beaɑɸi]

2.2.2 Nasalisation
Nasals are clearly identifiable as allophones of voiced stops; they appear in entirely complementary distribution. [m ŋ] appear word-initially, or when separated from word-initial position only by a sequence of one or more of the phoneme /ɑ/; otherwise various oral allophones occur:

/beeɡe/ → [meeɣe]
/ɡiɸi/ → [ŋiɸi] ~ [niɸi]
/ɑbeɑkɑ/ → [ɑmeaxa]
/ɑɑɡe/ → [ɔaŋe]

[n] appears only before the vowel [i], and is always in free variation with [ŋ] in that position, as in [ŋiɸi] ~ [niɸi] above.


2.2.3 Lenition
All consonants display some form of lenition (i.e. allophonic increase in sonority). [ɸ] and [β] are in free variation in both initial and word-internal positions (when this specific process is not relevant I will just write [ɸ] in phonetic forms).

/ɸɑie/ → [ɸɑie], [βɑie]
/biɸɑe/ → [miɸɑe], [miβɑe]

Note that the voiced fricative displays marked frication and is clearly distinct from the approximant [β̞], which sounds more like [w] to English speakers. The stops all lenide to fricatives or approximants, although only in intervocalic position following the vowels /e/ or /ɑ/; /b/ becomes [β̞], /k/ becomes [x] or [ħ], and /ɡ/ varies between [ɡ], [ɣ], [ʕ] and even [ʔ]. The use of [x] or [ħ] and that of [ɡ], [ɣ], [ʕ] or [ʔ] seems to be largely determined by the care the speaker is taking; in careful speech velars [x ɡ ɣ] appear, while in more casual speech laryngeals [ħ ʕ ʔ] predominate. There is also a weak correlation between the laryngeal allophones and preceding /ɑ/; /ɑk ɑɡ/ are slightly more often lenided to [ɑħ ɑʕ~ɑʔ]. I will represent them by [x] and [ɣ] in most phonetic forms.

/eebɑe/ → [eeβ̞ɑe]
/ɡɑbei/ → [ŋaβ̞ei]
/ɡekɑ/ → [ŋexa] ~ [ŋeħa]
/ibeɑɑkɑ/ → [ibeaɑxa] ~ [ibeaɑħa]
/ɑeeɡiiɑ/ → [ɔeeɡiisɑ] ~ [ɔeeɣiisɑ] ~ [ɔeeʕiisɑ] ~ [ɔeeʔiisɑ]
/eɡi/ → [eɡi] ~ [eɣi] ~ [eʕi] ~ [eʔi]

Note that this rule applies after nasalisation (§2.2.2). Preceding /i/ appears not to be sonorous enough to cause lenition of stops:

/ɑibiee/ → [ɔibiee]
/ɑikeɸe/ → [ɔit͡seɸe]
/eiɡi/ → [eid͡ʑi]

2.2.4 Rounding
The two rounded vowels [u] and [ɔ] are allophones of /i/ and /ɑ/. [u] is seen only in word-initial position where it alternates with [i] and zero. The elided forms are more common in casual speech, but rounding (and backing) to [u] appears to be fairly random.

/iɑee/ → [isɑee] ~ [usɑee] ~ [sɑee]
/iɸɑ/ → [iɸɑ] ~ [uɸɑ] ~ [ɸɑ]

/ɑ/ is rounded to [ɔ] word-initially when the next segment is another vowel, including in the sequence /ɑɑ/:

/ɑikeɸe/ → [ɔit͡seɸe]
/ɑeɑ/ → [ɔea]
/ɑɑebɑɑe/ → [ɔaeβ̞ɑae]

It is also rounded between two instances of /b/:

/ebɑbe/ → [eβ̞ɔβ̞e]
/bɑbi/ → [mɔβ̞i]
/ɸibibɑbe/ → [ɸibibɔβ̞e]

2.2.5 Fronting
/ɑ/ has a fronted allophone [a] which occurs following velars or /e/ [ɔ].

/kɑbeɡeɑɡɑ/ → [kaβ̞eɣeaɣa]
/bekɑi/ → [mexai]
/ɡɑɸɑ/ → [ŋaɸɑ]
/ɑeɑ/ → [ɔea]

This is not triggered by coronals, so the sequences /ikɑ iɡɑ/ retain the back vowel, but it is triggered by all the lenided allophones of /k ɡ/.

/kiikɑɑɡe/ → [kiit͡sɑɑɣe]
/iɡɑebe/ → [id͡ʑɑeβ̞e]
/ɸekɑ/ → [ɸexa] ~ [ɸeħa]
/eɡɑ/ → [eɣa] ~ [eʕa] ~ [eʔa]

2.2.6 Summary of allophonic variation
Surface forms can be produced from underlying forms using the following ten rules in standard sound change notation. # represents a word boundary, ∅ zero, ~ unconditioned variation and … a sequence of any length of a given phone.

• k ɡ → t͡s d͡ʑ / i _
• ∅ → s / i _ ɑ
• b ɡ → m ŋ / # (ɑ…) _
• ŋ → ŋ~n / _ i
• ɸ → ɸ~β
• b k ɡ → β̞ x~ħ ɡ~ɣ~ʕ~ʔ / {e,a} _
• i → i~u~∅ / # _
• ɑ → ɔ / # _ {i,e,ɑ}
• ɑ → ɔ / {b,m,β̞} _ β̞
• ɑ → a / {k,x,ħ,ɡ,ŋ,ɣ,ʕ,ʔ}_

Table 2.3 below shows all Mitsiefa phonemes and their allophones with environments.

Table 2.3: Mitsiefa phonemes and allophones thereof ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​​ ​ ​

Phoneme	Allophone	Environment	Example
/k/	[t͡s] [x] [ħ] [k]	Following /i/ Following /e ɑ/ Following /e ɑ/ Other	/ibike/ → [ibit͡se] /ɡekɑ/ → [ŋexa] /ɡekɑ/ → [ŋeħa] /kiɸɑɸɑ/ → [kiɸɑɸɑ]
/b/	[m] [β̞] [b]	Word-initially Word-initially following /ɑ…/ Following /e ɑ/ Otherwise	/beeɡe/ → [meeɣe] /ɑbeɑkɑ/ → [ɑmeaxa] /eebɑe/ → [eeβ̞ɑe] /ibeɑɑkɑ/ → [ibeaɑxa]
/ɡ/	[d͡ʑ] [ŋ] [n] [ɡ] [ɣ] [ʕ] [ʔ]	Following /i/ Word-initially Word-initially following /ɑ…/ Optionally word-initially before /i/ Optionally word-initially following /ɑ…/ before /i/ Following /e ɑ/ Following /e ɑ/ Following /e ɑ/ Following /e ɑ/	/biɡeɑe/ → [mid͡ʑeae] /ɡiɸi/ → [ŋiɸi] /ɑɑɡe/ → [ɔaŋe] /ɡiɸi/ → [niɸi] /ɑɡiɑ/ → [ɑnisɑ] /ɑeeɡiiɑ/ → [ɔeeɡiisɑ] /ɑeeɡiiɑ/ → [ɔeeɣiisɑ] /ɑeeɡiiɑ/ → [ɔeeʕiisɑ] /ɑeeɡiiɑ/ → [ɔeeʔiisɑ]
/ɸ/	[ɸ] [β]	All positions All positions	/ɸɑie/ → [ɸɑie] /ɸɑie/ → [βɑie]
/i/	[u] ∅ [i]	Optionally word-initially Optionally word-initially All positions	/iɸɑ/ → [uɸɑ] /iɸɑ/ → [ɸɑ] /ɑɡi/ → [ɑɣi]
/e/	[e]	All positions	/ɑikeɸe/ → [ɔit͡seɸe] ​ ​
/ɑ/	[ɔ] [a] [ɑ]	Word-initially before /i e ɑ/ Between two instances of /b/ Following /k ɡ e/ or [ɔ] Otherwise	/ɑeɑ/ → [ɔea] /ebɑbe/ → [eβ̞ɔβ̞e] /ɡɑɸɑ/ → [ŋaɸɑ] /iɡɑebe/ → [id͡ʑɑeβ̞e]

2.3 Phonemic inventory

The phonemic inventory of Mitsiefa has been established as one of four consonants and three vowels.

2.3.1 Consonant inventory

Table 2.4 below shows Mitsiefa's phonemic inventory of consonants:

Table 2.4: Mitsiefa consonants

	​ ​ Bilabial​ ​	​​ ​ ​Velar​ ​ ​
Stop (-VC)		​ ​ ​ ​ ​ ​/k/
Stop (+VC)	​ ​ ​ ​​ ​ ​/b/	​ ​ ​​ ​ ​ /ɡ/
Fricative	​ ​​ ​​ ​ ​ /ɸ/

There are two phonemic places of articulation – bilabial /b ɸ/ and velar /k ɡ/ – combined with three degrees of sonority – voiceless stop /k/, voiced stop /b ɡ/ and voiceless fricative /ɸ/. Gaps are seen in two of the six possible slots, which if filled would contain */p/ and */x/. It would be possible to simplify the inventory to a two-row table of "voiceless" vs. "voiced" (table 2.5), which probably reflects the historical situation of Mitsiefa.

Table 2.5: Mitsiefa consonants

	​ ​ Bilabial​ ​	​​ ​ ​Velar​ ​ ​
​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ -VC	​ ​ ​ ​​ ​ ​/ɸ/	​ ​ ​ ​ ​ ​/k/
​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ +VC	​ ​ ​ ​​ ​ ​/b/	​ ​ ​​ ​ ​ /ɡ/

From a functional point of view this is, however, not very effective. /ɸ/ and /k/ do not pattern together – the former has free voicing allophones, while the latter is fricated without voicing in lenition environments (both described in §2.2.3). In contrast, /b/ and /ɡ/ form a very clear class, by nasalising word-initially (§2.2.2) and sonorising following non-high vowels (§2.2.3). /k/ and /ɡ/ also form a class in terms of POA; they both palatalise following /i/ (§2.2.1). Interestingly, /ɸ/ and /b/ do not form a strong class either, since /b/ can trigger rounding of /ɑ/ (§2.2.4) while /ɸ/ has no effect.
It seems prudent to describe /b/ as [+bilabial][+voice], /ɡ/ as [–bilabial][+voice] and /k/ as [–bilabial][–voice], but it does not necessarily follow that /ɸ/ is [+bilabial][–voice]. /ɸ/ is the only consonant to have both voiced and voiceless allophones (the occasional [ʔ] allophone of /ɡ/ cannot really be considered "voiceless" since it is physiologically impossible to voice a glottal stop), so it should be considered voice-unspecified. I will instead treat /ɸ/ as being marked simply [–stop].
Notably absent are several series which are found very frequently in spoken languages, and which the lack thereof is of typological note – there are no coronal phonemes, no nasals and no other sonorants. I will discuss this in more detail in §2.3.6.


2.3.2 Consonant minimal pairs

Minimal pairs are not difficult to find for any pairs of Mitsiefa consonants. Since there are just four consonants + zero, there are only 10 combinations which are necessary to demonstrate phonemicity for all oppositions, and I will provide examples of all of them below (it is fairly likely that there are also all 10 minimal triplets needed to demonstrate phonemicity, and perhaps although less likely even the 5 minimal quadruplets, but for these purposes minimal pairs will suffice). Two pairs are given for each combination, one in word-initial position and one in word-internal position.

Table 2.6: Minimal pairs demonstrating phonemicity of Mitsiefa consonants​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​ ​ ​ ​ ​ ​ ​​​ ​ ​

​ ​ ​ ​ ​ ​ ​	/k/	/b/	/ɡ/	/ɸ/
∅	/kɑiɸii/ "bug" vs. /ɑiɸii/ "name" /ɸɑkɑ/ "hostel" vs. /ɸɑɑ/ "shoes"	/beɑe/ "hurry" vs. /eɑe/ "dog" /ɑbɑɡi/ "day" vs. /ɑɑɡi/ "do often"	/ɡekɑ/ "lizard" vs. /ekɑ/ "fire" /beeɡi/ "now" vs. /beei/ "sparkling"	/ɸɑbɑ/ "walk" vs. /ɑbɑ/ "now" /ɡɑɸe/ "thunder" vs. /ɡɑe/ "sugar"
/ɸ/	/kiibe/ "possess" vs. /ɸiibe/ "bat" /eɑke/ "wing" vs. /eɑɸe/ "male"	/bekɑ/ "you" vs. /ɸekɑ/ "base" /beɑbɑ/ "lazy" vs. /beɑɸɑ/ "change"	/ɡiɑi/ "very big" vs. /ɸiɑi/ "season" /ɑɡi/ "k.o. fruit" vs. /ɑɸi/ "stick"
/ɡ/	/kɑbiɑ/ "beat" vs. /ɡɑbiɑ/ "method" /ɸikɑ/ "eat" vs. /ɸiɡɑ/ "clay"	/bɑie/ "pot" vs. /ɡɑie/ "numerous" /ɑɑbɑ/ "notice" vs. /ɑɑɡɑ/ "nail"
/b/	/kiɑi/ "step on" vs. /biɑi/ "snap" /ɡɑkɑi/ "woman" vs. /ɡɑbɑi/ "rain"

I have made no effort to choose pairs of the same word class since there does not appear to be any phonological distinction between word classes, barring the reduplication seen in ideophones which I will discuss in §2.4.3 and does not appear in any of these pairs.


2.3.3 Vowel inventory

Table 2.7 below shows Mitsiefa's phonemic inventory of vowels:

Table 2: Mitsiefa vowels ​ ​ ​ ​

	​ ​ ​ Front​ ​ ​ ​	Non-front
High	​ ​ ​ ​ ​ ​/i/
Mid	​ ​ ​ ​ ​ ​/e/
Low		​ ​ ​ ​ ​ ​/ɑ/

The only contrasting feature is height; /i/ is high, /e/ is mid and /ɑ/ is low. Backness is entirely secondary; there is no front counterpart to /ɑ/, and there is considerable frontness fluctuation in both /i/ [i ~ u] and /ɑ/ [a ~ ɑ ~ ɔ]. Roundedness is similarly secondary. The only height variation is /ɑ/ → [ɔ] which is still lower than any of the allophones of the next-lowest vowel /e/. This height contrast is also one of sonority; /i/ is the least sonorous vowel and thus the most prone to elision (see §2.2.4) and also the only vowel which displays non-syllabic features (§2.2.1). Also a distinct feature of /i/ is [+palatal], which causes progressive palatalisation of /k ɡ/ and occasional regressive palatalisation of [ŋ].


2.3.4 Vowel minimal pairs

There are numerous minimal triplets which demonstrate the phonemicity of all three vowels, and also of zero:

/bi/ "speech"
/be/ "house"
/bɑ/ "fingernail"

/kiɑ/ "increment"
/keɑ/ "k.o. parrot"
/kɑɑ/ IDEO "expression of pain"
/kɑ/ adverbial phrase particle, "vine"

As such it is not necessary to reproduce a table like table 2.6.


2.3.5 Distribution

All phonemes in Mitsiefa are necessarily fairly common, and all have a high functional load. The three vowels are the most common; /ɑ/ alone accounts for 27.1% of all phonemes, and 40.5% of vowel phonemes. Interestingly the voiced stops are the most common consonants, with /b/ making up 11.2% of phonemes and /ɡ/ 8.7%. Least common is the voiceless stop /k/, only 6.5% of all phonemes. This data is summarised in the chart below.

This data is taken from the words I've used as examples so far, so it's not really a very good sample size. I've been using zomp's word generator, but with some degree of personal selection. The raw output from gen gives the following frequencies: /ɑ/ - 30% /e/ - 23% /i/ - 15% /b/ - 13% /ɸ/ - 8% /k/ - 7% /ɡ/ - 5% Which suggests I favour /i/ and /ɡ/ and disfavour /ɑ/ and /ɸ/. The biggest disparity is /ɡ/, which I can't explain, and /i/, which I suspect I was favouring because there's lots of allophonic processes it takes part in that I needed examples for.

It is noteworthy that Mitsiefa is a fairly rare example of a language where words containing each one of the language's phonemes are not uncommon; examples include /kɑieɸebeɡi/ "tweezers" and /ɡɑbekɑɸɑbi/ "mortar". Due to the word structure limitations there is no word which contains only the seven phonemes, but there is at least one eight-phoneme word, /keɡibɑɸe/ "trip, voyage".


2.3.6 Typological implications

Mitsiefa is a very unusual language. As mentioned previously, its lack of numerous phoneme series – rounded vowels, coronals, nasals, resonants – makes is typologically highly marked, as does its simple numeric paucity of phonemes. In this section I will compare Mitsiefa to the list of phonological universals established in Larry Hyman's article Universals in Phonology (The Linguistic Review, April 2008. DOI: 10.1515/TLIR.2008.003) to compare this in some way to the linguistic tendencies of Earth languages.

• Consonant universal 1 - "Every phonological system has stops."

Mitsiefa does follow this rule with its three-stop inventory /b k ɡ/. Previously it was thought that "all languages have three out of /p t k ʔ/" was also a universal, but numerous languages have disproven this (e.g. Abau, Yidiny, Vanimo, Omurano, etc.). Some doubt has been cast on this one of Hyman's universals with an analysis of the Ontena dialect of Gadsup, which has /ʔ/ but no other phonemic oral stops (/ɸ s ɾ x/ surface as [p t d k] following /ʔ/). This specific stop inventory is not to my knowledge known in any other language, although all non-eastern dialects of Mekeo have (according to Jones' Lexicogrammar of Mekeo, 1998) similar sets of /p k ɡ/ (northwest), /p b k ɡ/ (west) and /b k/ (north). Some Papuan languages such as Abau and Central Sentani have the same places of articulation but without a voicing contrast (i.e. just /p k/).

• Consonant universal 2 - "Every phonological system contrasts phonemes which are [−cont] (= stops) with phonemes that are specified with a different feature."

With Ontena Gadsup in mind, this can be reformulated as "Every phonological system contrasts obstruent phonemes with phonemes that are more sonorous (without necessarily not being obstruents themselves)". This allows for the analysis of Rotokas with /p b t d k ɡ/, even if /b d ɡ/ are not considered [+cont]. In any case, Mitsiefa again follows this universal, as /ɸ/ is most certainly [+cont] and /b ɡ/ are more sonorous than /k/.
The presence of nasal consonants was also once thought to be universal, but numerous languages without nasals have been found in both of the Americas, Africa and New Guinea. Some languages go even further than Mitsiefa in lacking phonetic nasals also (notably the languages around Puget Sound, as well as the eastern Lakes Plain languages and Rotokas of New Guinea). Spontaneous allophonic word-initial nasalisation is also found in Pirahã, Wichita and some Lakes Plain languages such as Kirikiri.

• Consonant universal 3 - "Every phonological system contrasts phonemes for place of articulation."

Although Mitsiefa has only two places of articulation, they do contrast, including a contrast at the same degree of sonority (/b/ vs. /ɡ/). I know only of one other language which has only two places of articulation, which is the northwest dialect of Mekeo which has an inventory of /p k β ɡ m ŋ w j/. This does however require a somewhat lax interpretation of /p k β m w/ as "labial" and /k ɡ ŋ j/ as "dorsal".

• Consonant universal 4 - "Every phonological system has coronal phonemes."

At this point Mitsiefa does violate an apparent universal. However, northwest Mekeo as described above has already been shown to violate this, so it cannot be considered a true universal. It is however an extremely strong tendency, and I have no doubt it would be true to say that all languages have coronal phones ([dz n] appear predictably in northwest Mekeo). In section §2.6 I discuss whether this may be a partially intentional behaviour by Mitsiefa people, or at least a culturally conditioned one.

• Vowel universal 1 - "Every phonological system contrasts at least two degrees of aperture."
• Vowel universal 2 - "Every phonological system has at least one front vowel or the palatal glide /j/."
• Vowel universal 3 - "Every phonological system has at least one unrounded vowel."
• Vowel universal 4 - "Every phonological system has at least one back vowel." (?)
• Vowel universal 5 - "A vowel system may be contrastive only for aperture only if its vowels acquire vowel color from neighboring consonants."
• Vowel universal 6 - "A vowel system can be contrastive for nasality only if there are output nasal consonants."

All of Hyman's vowel universals are true in Mitsiefa. Hyman himself notes that #4 is only true given analysis of "vertical vowel system" languages as having more normal vowel systems. Mitsiefa can be somewhat characterised as having a vertical vowel system (a system contrasting "aperture", i.e. height, alone), although with a far lesser degree of consonant colouring than the prototypical vertical vowel system languages of the Caucasus. Nevertheless the default value of Mitsiefa's low vowel is back, so it follows the universal. Mitsiefa only barely qualifies for #5, since only /ɑ/ gains "colour" (i.e. frontness and/or roundedness) from neighbouring consonants; preceding velars front it to [a] and surrounding [b] backs and rounds it to [ɔ]. There is at least one other example of a language with an /i e a/ vowel system, Wichita, which has surface [o] arising from sequences of short vowels + /w/ ([e] may also be secondary, arising from vowel + /j/ sequences, but with less certainty).

While Mitsiefa does not violate any universals on its own, it is still a record-holder, with only seven total phonemes. This far surpasses Pirahã, an isolate from the Amazon which has been dubiously claimed to have only nine phonemes /p t ʔ b ɡ h i o a/, although normally /s/ appears as a distinct phoneme from /h/, and [k] must be analysed as the sequence /hi/ for reasons which are not entirely clear. More likely is an eleven-phoneme inventory of /p t k ʔ b ɡ s h i o a/. This number is shared by a few other languages – all of which, interestingly, have five consonants and six vowels – Rotokas (North Bouganville), with /p b t d k ɡ i u e o a/, Obokuitai (Lakes Plain) with /b t d k s h i u ɛ ɔ a/, North Mekeo (Austronesian) with /b k m ŋ β l i u e o a/, , Pawnee (Caddoan) with /p t t͡s k s h ɾ w i u a/ and the Northwest River Montagnais spoken by young people with /p t t͡ʃ k h m n i u e a/. . In addition to these segments, Rotokas , Pawnee and NWRM have contrastive vowel length and Obokuitai a three-way tone distinction. One fewer phoneme is posited for proto-Lakes Plain, the ancestor of Obokuitai, which likely had an inventory of */p b t d k i u ɛ ɔ a/, although as a reconstructed language this cannot be taken as precedent. Additionally there is some morphophonological evidence which suggests [p] may be secondary in Pawnee, which would drop it down to ten phonemes also.

Mitsiefa's consonant system is also the smallest known. Four languages listed above all have six consonants (although Pirahã probably has eight), as do numerous other Lakes Plain languages related to Obokuitai, and the Elema language Orokolo (this having /p t k m l h/). Proto-Lakes Plain seems to have had only five consonants, but I know only one currently-spoken language which has been claimed to have this number; in a single talk, Mark Donohue mentions in passing that Biritai (another Lakes Plain language) has only five consonants /b t d ɸ s/. Whatever the veracity of this claim, Mitsiefa is certainly a new benchmark.

2. Phonology (continued)

2.4 Word structure

In almost all instances the Mitsiefa morphological word and phonological word are identical, since there is no synthetic morphology – the one exception to this is some ideophones which appear to be fossilised reduplicants (§2.4.3). It is possible to analyse a structure intermediate between the phoneme and the word – the syllable – but the evidence for this is scanty and it seemingly has no effect on phonological processes.

2.4.1 Syllables

A traditional analysis would describe Mitsiefa as having two types of syllable, V and CV, with no restrictions as to the comibination thereof:

/e/ (agentive particle)
/ɸi/ "cotton"
/ɑ.e/ "leaf"
/ɑ.ɸi/ "stick"
/ɸɑ.ɑ/ "shoes"
/ɸi.kɑ/ "eat"
etc.

There is no sure way to syllabify vowel sequences, which are extremely common and have no apparent bounds. In the special case of /iɑ/, we can be fairly sure the two vowels are of separate syllables since an allophonic [s] is inserted between them, causing a marked trough in sonority normally indicative of a syllable edge:

/ɡi.ɑi/ → [ŋi.sɑ(.)i] "very big"

Other vowel sequences appear with alarming frequency. Not only are all 9 two-vowel sequences attested, but also are all 27 three-vowel sequences, most of the 81 four-vowel sequences, several five- and six-vowel sequences, the occasional seven-vowel sequence, and at least one eight-vowel sequence.

/kei/ "car"
/beiɑ/ "be bulbous"
/ɑɸɑeɑe/ IDEO. "going in a smooth zigzag or wavy pattern"
/iekeeiɑɑ/ "buttocks"
/bɑeɑeɑi/ "cook dinner"
/ɑeɑɑiɑe/ "shutter"
/eɑiɑ-eɑiɑ/ IDEO. "hissing"

Like vowels are frequently doubled or tripled, as well as the occasional sequence of four instances of /ɑ/ (/ebɑɑɑɑ/ "fish sp.", /ɑɑɑɑi/ "needle"); no four-/i/ or four-/e/ sequences are known. There is no clear indication as to whether nine-vowel sequences might theoretically be permitted, or for that matter four-vowel sequences of other vowels. /ɑ/ appearing in a sequence of four in three or four words may just reflect its high frequency in general. It is worth noting that there are no minimal pairs between /ɑɑɑ/ and /ɑɑɑɑ/, although again this may simply be coincidental. Also the majority of seven-vowel sequences contain /iɑ/, which may be significant in phonetically breaking up the sequence, but ⅔ of seven-vowel sequences would include this sequence by random chance and there are not enough examples to be sure whether there is a genuine correlation.

Another analysis of Mitsiefa simply states:

1. Words are constructed from a sequence of vowels and consonants, such that:
2. No two consonants may come together, and,
3. A consonant may not appear at the right edge of a word.

This is equally as effective as a syllabic analysis, and has the advantage of not requiring the insertion of somewhat arbitrary syllabification of vowel sequences.


2.4.2 Words

All words bar a small number of ideophones are constructed from a series of vowels each with or without a preceding consonant. There is no hard word-length boundary; the longest monomorphemic word attested so far is the ideophone /kiɑkɑɸikɑiɑkɑɸɑkɑ/ which (quite aptly) describes a rapid unintentional movement involving frequent impacts (for instance a small child being dropped into an industrial tumble-drier). This contains by a syllabic analysis ten syllables, or by a non-syllabic analysis ten vowels and seven consonants. A few reduplicative ideophones are 12 syllables such as /bebeeɸiee-bebeeɸiee/ "buzzing of insects".


2.4.3 Reduplicative ideophones

While the majority of ideophones are simply normal words, a few have what could be considered a morpheme boundary within them. These are always reduplicants or retriplicants, although there is never a corresponding un-reduplicated form.

/ɡɑkɑɑ-ɡɑkɑɑ/ "crow cawing" (*/ɡɑkɑɑ/ "?")
/iiɡeɑi-iiɡeɑi/ "slipping on mud"
/ɸiɑ-ɸiɑ-ɸiɑ/ "k.o. bird call" (/ɸiɑ/ exists as a word but means "seize")

The boundary between these acts as a word boundary, preventing lenition of /k b ɡ/, palatalisation of /i-k i-ɡ/, hiatus repair of /i-ɑ/ and fronting of /e-ɑ/, while also triggering nasalisation of /b ɡ/ and rounding of /i ɑ/.

/ɡɑkɑɑ-ɡɑkɑɑ/ → [ŋaxaɑŋaxaɑ]
/ɑɸii-ɑɸii/ → [ɑɸiiɑɸii] (*[ɑɸiisɑɸii])
/keibi-keibi/ → [keibikeibi] (*[keibit͡seibi])
/ɑɸeɸe-aɸeɸe/ → [ɑɸeɸeɑɸeɸe] (*[ɑɸeɸeaɸeɸe])
/ɑɑbɑbe-ɑɑbɑbe/ → [ɔaβ̞ɔβ̞eɔaβ̞ɔβ̞e]

This structure seems to contrast with "coincidental" reduplicants, where the same phonological form appears twice but normal word-internal processes take place, e.g. /ɡɑiɡɑi/ → [ŋaid͡ʑɑi] "hock".


2.5 Phrases and intonation

While there is no stress or tone in Mitsiefa, there is a strong tendency for phrase-level downdrift. A phrase is marked by a short pause at the left and right edges (marked with "|") and a fairly steady drift downwards in pitch from the first vowel of the first word to the last vowel of the last word (marked with "↘︎"), tending to accelerate towards the right edge. For this purpose there are five types of phrase – the topic, the comment, the adverbial phrase, the noun phrase and the ideophone. The topic is generally a single phrase (as is the comment), the topic being followed by a fairly long pause (notated "‖").

Amaai kagibiba mi ↘︎ ‖ oefe efieae ↘︎ amaai man kagibiba be.old mi TOP oefe people efieae shoot "As for the old guy, they shot him."

If there is a relative clause consisting of more than a single verb within a topic NP however, the comment of the relative clause starts a separate phrase from the noun:

Mabaa ↘︎ | mifaa isabifae i efa mi ↘︎ ‖ mifaa ngi idzia ahefaa ↘︎ mabaa infant mifaa 1SG isabifae preschool i LOC efa send mi TOP mifaa 1SG ngi NEG idzia remember ahefaa retrieve "The infant whom I took to preschool, I forgot to pick him up."

Comment phrasing is more variable. Generally speaking simple verb + NP or verb + AP or verb + ideophone comments are treated as a single phrase:

Mifaa ↘︎ ‖ ka·ka meaifa ↘︎ mifaa 1SG ka·ka glancing.blow.noise meaifa hit.on.head "As for me, well, bam! I whacked him on the head."

Whenever there are multiple of one type of comment modifier (multiple NPs, multiple APs or multiple ideophones), they are always separated into distinct phrases:

Mifaa↘︎ ‖ egaga idzidzi meaifa ↘︎ mifaa 1SG egaga punching idzidzi slapping meaifa hit.on.head "As for me, well, pow, whap! I punched him and slapped his face."

However, there is variability when it comes to multiple modifiers of different types, which may be phrased together with the comment, or partly or wholly broken up into individual phrases:

Mefe ↘︎ ‖ mifaa e ↘︎ | fahaaifa e mieitse i ↘︎ | fisa kagi ↘︎ mefe object mifaa 1SG e AGT fahaaifa long.distance e INSTR mieitse lake i LOC fisa hold kagi carry "I carried it all the way over to the lake."

Ideophones also appear as distinct utterances, in which case they are their own phrase:

Mebeefiee·mebeefiee ↘︎ mebeefiee·mebeefiee buzzing.insect.noise "Bzzzzz."

Negated comments and questions have distinct intonation types. The former (achieved with the particle ngi, or an adverbial phrase of a negative word such as ngaabaa "not true" + the adverbial particle ka, or both) is marked by a rising intonation from the beginning of the comment phrase up to the negative marker, followed by marked low pitch on the first syllable of the negative marker, then a return to normal down-drift phrases:

Ngakai isa mi ↘︎ ‖ mifaa ↗︎ ngiꜜ ngaaꜛbaba ka oa eidziabe oangi ↘︎ ngakai woman isa DEM mi TOP mifaa 1SG ngi NEG ngaagaba nothing.at.all ka ADV oa fuck eidziabe be.involved.with oangi do.often "I did not have sexual relations with that woman."

If the comment is broken up into multiple phrases, then the entire phrase preceding the negation has a rising intonation:

Mifaa e mehabi ea ↘︎ | kaabe i ↗︎ ngiꜜ faꜛba faba ↘︎ mifaa 1SG e AGT mehabi noise ea PRIV kaabe night i LOC ngi NEG faba walk faba walk "I will not go quietly into the night."

There are however a few negated verbs which are lexicalised and are not separated from the intonation of the comment:

Amaai mobitsi ↘︎ ‖ ngi ahaefa ↘︎ amaai man mobitsi police ngi NEG ahaefa prevent "The policeman stood idly by."

Questions are marked by a global rise on the topic – with any internalised relative clauses retaining their distinct downdrift, but with the last phrase of complement clauses taking the rise.

Meha angei moboba mifaa ii ↘︎ kitsidze fiha mi ↗︎ ‖ oaka? ↘︎ meha 2 angei rat moboba chubby.face.fat mifaa 1SG ii GEN kitsidze kill fiha eat mi TOP oaka be.true "(Is it true that) you killed and ate my hamster?"

2.6 Cultural influence on phonology

In some societies there is a higher degree of linguistic awareness and identity than in others. Whereas in modern western society language is more often seen as a tool, and language difference as an impedement to communication, in some cultures language differences are an important part of a group's identity. It is undoubtedly for this reason that the islands of Vanuatu were able to support at least 140 distinct languages in an area smaller than that of metropolitan Sydney, while the entirety of Europe boasts only a few dozen vibrant languages. This linguistic identity can lead to a deliberate, although unplanned, group effort to make one's own community's speech distinct from that of others. This process has been posited as a possible explanation for the adoption of seemingly bizarre sound changes in some Melanesian languages – although the changes themselves may have arisen naturally, the adoption thereof could have been aided by a sense of linguistic unity in opposition to the surrounding language groups. A similar situation may be the case in other areas of the world where language groups are small and isolated, such as the alpine areas of Francoprovençal speech which have seen far more radical sound changes than any other area of Romance.

Mitsiefa is an interesting case. Up until recent history it was only one of many languages spoken in its area, with a fairly closed speaker base of around 5,000 individuals. It was at this time that Mitsiefa lost its coronal phonemes and back vowels, possibly in part to distinguish themselves from surrounding language groups. Mitsiefa speakers will use these sounds with great frequency when imitating foreigners, regardless of whether these are particularly common sounds in the language they are attempting to imitate – lateral [l] is particularly popular in this regard, perhaps because it does not exist even phonetically in Mitsiefa. Similarly, when attempting to communicate with speakers of a language they do not know, coronals will be substituted freely for any consonants at all, often even with disregard to manner of articulation. A similar pattern is seen in syllable structure; dummy consonants will be inserted between vowel sequences and codas added to make speech seem more "foreign".

A bit of background is in order here. This grammar is being written ca. 30 years after first contact between Earth and whatever planet Mitsiefa is spoken on. As for when this actually occurs, where this planet is or how we got there, I'm being intentionally vague. Suffice to say humans are technologically more advanced than Mitsiefa people and fairly similar physiologically (except Mitsiefa people have sixteen heads with a hundred bulbous green eyes each, forty-foot-long poisoned tentacles and the ability to teleport through miles of solid rock and spark hydrogen fusion in their stomachs). Human/Mitsiefa interaction is partly trade-based, but various anthropologists and linguists and missionaries are just crazy about a whole new planet of people to study/convert.

Very recently however, Mitsiefa has become a language of high prestige, something of a lingua franca, thanks to the Mitsiefa's unique position as primary trading partners of the region with Earth, facilitated by their access to an area of suitable near-surface bedrock for construction of ports and an eagerness to obtain new technologies. The surrounding peoples, being forced to trade through the Mitsiefa for desirable Earth goods, have taken to learning Mitsiefa as a trade language, as have the most keen of human traders. It appears that there has not been sufficient time for the typologically marked features of Mitsiefa to resolve.