Korean Phonology Revisited

Sixteen letters are all you need

18 min readAug 27, 2022

Abstract

The traditional Seoul Korean analysis of nineteen consonants is reorganized into a simpler ten-consonant system. Additionally, it will demonstrate how the segmental system as a whole only needs sixteen segments, including two glides and four vowels, each with just four binary features.

1. Introduction

1.1. Consonants

Yu Cho (2016) lists the following consonants as its phonemic inventory: p, ph, p’, m, t, th, t’, s, s’, n, r, č, čh, č’, k, kh, k’, ŋ, and h. Yu Cho’s <c> is replaced with <č> to show that it is a postalveolar affricate, <l> with <r>, and the superscript h with a regular <h>. The original chart is available here.

1.1.1. The plain series

The plain stops, p, t, č, and k, are weakly aspirated voiceless stops wordinitially, and voiced stops between sonorants. They are unreleased voiceless stops at the coda position. They do not appear in other contexts.

Even between sonorants, the alveolar fricative s is frequently voiceless, though it appears that voicing is a free variation (T. Cho, et. al. (2002), p. 212). The glottal fricative h is regularly voiced to ɦ between sonorants.

1.1.2. The aspirated series

The aspirated stops, ph, th, čh, and kh, are strongly aspirated voiceless stops in all positions. They do not appear at the coda position.

1.1.3. The tense series

The tense stops, p’, t’, č’, and k’, are voiceless unaspirated stops in all positions. They do not appear at the coda position.

The tense alveolar fricative s’ is less aspirated than the plain counterpart (Park (1999)) and is voiceless even intervocalically.

1.1.4. The feature specifications

Kim (1965) describes the aspirated and tense series to be tense and the plain series to be lax. The aspirated series exhibits a higher VOT value than its tense counterpart, whereas the VOT values for the plain series are just allophonic. The two tense series require stronger muscle activity near the point of articulation. These facts allowed Kim to arrange these stops into a diagram of cross-cutting categories of tensity and voicing:

Kim’s analysis sufficiently captures the known phonemic contrasts and hence allows us to express them accordingly. Following the convention of Chomsky and Halle’s (1968) The Sound Pattern of English, we henceforce use the macron (¯) to represent tensity to replace p’, t’, č’, and k with b̄, d̄, ǰ̄, and ḡ, and ph, th, čh, kh, and s’ with p̄, t̄, č̄, k̄, and s̄.

The sonorants, m, n, r, and ŋ, and the glottal h are assumed to be lax or [–tense].

1.2. Vowels

Yu Cho’s vowel inventory admits ten vowels:

The use of the symbol æ in place of the low front vowel doesn’t imply it is as low as IPA [æ]. It simply indicates the vowel is lower than e by a sufficient distance to be contrastive. Also note that the contrast between e and æ is largely lost for younger speakers.

Kim’s (1968) analysis, however, reduces the vowel inventory into four members:

The unrounded vowels *ɨ, ə,* and a are treated as central vowels while they are treated as back vowels in Yu Cho’s. This is because Kim assumes three levels of backness while Yu Cho assumes two.

Under Kim’s analysis, front vowels are obtained by affixing the palatal glide y: /əy/ [e], /ay/ [æ], /oy/ [ö]. /ɨy/ is wordinitially [ɨy] and [i] elsewhere. Also, [u] and [yu] is obtained from /wɨ/ and /yɨ/, respectively.

Notice that, if we replace Kim’s ə with the unrounded low back vowel ʌ, the vowels will be classified into two cross-cutting binary categories. The conservative variant has ə: as a longer allophone of ʌ, so both seem to have right to claim the underlying standing.

We will later see why the latter is a preferrable vowel inventory.

1.3. References

2. Analysis

2.1. The phonetic inventory

The following inventory serves as good enough approximations of surface phonetic values for our analysis.

*ŕ, ń, š, and š̄* are interpreted as the palatalized forms of r, n, s, and s̄.

Vowels may be long or short wordinitially. The mid central vowel ə only appears as the long vowel ə: and is the longer counterpart of ʌ.

2.2. Phonological interpretations

2.2.1. Plosives

Consider the following:

All sonorants are voiced and are lax
All lax obstruents are voiceless or their voicing is not contrastive
All fricatives are voiceless or their voicing is not contrastive
Palatal is only phonemic place out of labial, dental, palatal, and velar that lacks a nasal
The postalveolar affricates are the only plosive series that lack their nasal counterparts

With these facts, consider the following interpretation of stops:

As the lax plosive series lacks a voicing contrast, they can be interpreted as voiceless in the phonemic forms. Given that, the only lax voiced stops left are the sonorant series.

The liquid r~ŕ is interpreted to be underlyingly palatal. Given the four places of articulation corresponds to the plosives, palatal is the only place that otherwise lacks a segment. Additionally, one may point out that as the liquid is the only oral sonorant consonant, the postalveolar (‘palatal’ in our phonemic framework) series are the only affricate plosives.

2.2.2. Continuant consonants

Consider the following:

Velar is the only phonemic place out of labial, dental, palatal, and velar that lacks a continuant consonant or glide

With that in mind, the continuants can be interpreted as the following:

labial: ẘ
dental: s, s̄
palatal: ẙ
velar: x

The velar continuant x is the underlying correlate of the glottal fricative h~ɦ as otherwise velar place is the only empty spot. Our analysis on consonat clusters provides further justifications of the debuccalization analysis.

All continuant consonants are deemed underlyingly voiceless. (The ring above the letters w and y indicates voicelessness.) This is a possible consequence of Kim’s (1968) analysis on glides: the labial glide [w] is closed to [p], not [m] nor [b], before a consonant, e.g., [küyʌwʌ]~[küyʌpd̄a] ‘is cute’. This is analogous to [s] being closed to [t] before a consonant as in [usʌ]~[utḡo] ‘to laugh’. These alternations are generalized that a voiceless segments are closed before a consonant. That is:

glides → [+voice] /_V.
[–voice] → [–continuant] /_C.

2.2.3. Canonical vowels

Our phonemic analysis admits four places of articulation: labial, dental, palatal, and velar. The feature specifications of these places are given with [±grave] and [±compact]:

The same features adequately capture our vowels: ɨ, o, a, and ʌ. Hence, if all vowels are underlyingly voiced, continuant consonants and vowels, both [+continuant], are told apart by [±voice]:

Notice that Kim’s (1968) original proposal is less economical due to the vertical arrangement of central vowels, ɨ, ə, and a.

2.2.4. ɨy and yɨy

Kim (1968) proposes /ɨy/ and /yɨy/ coexist in the vowel system, and while both realize as [i] phonetically, dental consonants are sensitive to their difference and palatalize only when followed by /yɨy/ and not by /ɨy/. This argument is highlighted in the following passage:

In Korean a palatalization of a dental C occurs before yV and certain cases of i, e.g. pinya [piɲə] ‘hair-rod’, hullyuŋ [huλuŋ] ‘excellence’, hɛtoti [hɛtoci] ‘sunrise’, kathi [kachi] ‘together’; but puti [puti] ‘by all means’ (not [puci]), thikkɨl [thikkɨl] ‘particle’ (not [chikkɨl]), etc. Unless we differentiate a morphonemically palatal /yi/ from morphonemically non-palatal /i/, even though the two are phonetically realized as the same [i], there is no way to state the palatalization rule with any generality.[] This justifies positing two underlying representations /ɨy/ and /yɨy/ for one phonetic [i]. (pp. 521–522.)

Here, he conflated two different kinds of palatalization. One is the automatic alternations between n and ń and between r and ŕ. The other is alleged alternations between two contrastive consonants, t and č.

Palatalizations of n and r to ń and ŕ found in pinya [sic.] ‘hair-rod’ and hullyuŋ ‘excellence’, in our phonetic notation [pińʌ] and [huŕŕuŋ], are purely allophonic realizations of the acute sonorants in question in the palatalizing environment. The nonpalatalized counterparts, n and r, never appear before the palatal glide or the vowel [i], e.g., *[ni], *[ri] (ok [ńi], ok [ŕi]). If one argues that the vowel /ɨy/ is nonpalatalizing, he then must establish a seperate rule that neutralizes /ɨy/ and /yɨy/ after acute sonorants to account for the absense of *[ni] and *[ri].

The alleged palatalization cases of t and t̄ into č and č̄ are of a different nature. Both of these series freely stand before [i], as Kim himself points out, as in puti [pudi] ‘by all means’ and kathi [kač̄i] ‘together’. This is because, unlike the previously mentioned sonorant series, the dental and palatal plosives are independent phonemic units. They can be stored in lexicon without a distinction between /ɨy/ and /yɨy/, in forms such as /putẙɨẙ/ [pudi] and /kačxẙɨẙ/ [kač̄i], and Kim provides no cases against such an assumption. As an additional case againt Kim’s palatalization account, the phonetic sequence [tyV] is attested in words such as [tyu:sɨ] ‘deuce’. Such forms would not be possible if the dental and palatal plosives were mere allophones.

If /ɨẙ/ is not necessary for [i], then, what is the phonetic realization of it? An obvious possibility, though I do not adopt it in this article, is complete neutralization in all noninitial positions. That is, /ɨẙ/ and /ẙɨẙ/ in noninitial positions are indiscriminately interpreted as [i] and accordingly palatalize dental sonorant and continuant consonants.

In the next section, I propose alternatively that /ɨẙ/ is [ay] before a vowel and [a] before a consonant.

2.2.5. Vowel harmony

Vowel harmony is observed in the ending ʌ~a when it attaches to the stem. If the last vowel of the stem is either [a] or [o], the ending realizes as [a], and [ʌ] otherwise.

It, however, breaks for the verb ha (or hay depending on the analysis) ‘to do’, which appears as [hayʌ] when the harmonizing ending. Since the last vowel of the form ha is [a], it seems that the ending should harmonize into [a] instead.

Note: In this article, the source of the vowel length is ignored for simplicity, and the vowels may be written as long or short in both underlying and surface notations. This is not meant to claim that vowels have underlying length. Similarly, some morphophonemic analyses are omitted for simplicity. For example, the underlying form of the verb ending ta is probably /ǰta/, but it is written as /ta/ in this article regardless. See Iida (2021) for an attempted analysis.

One may try to explain the form [hayʌ] by supposing that the underlying form of the stem is /xaẙ/ as the traditional generalization predicts that the ending becomes [ʌ] if the stem’s last vowel is a front vowel. Under our analysis, front vowels are /Vẙ/. The problem, however, is that it is not a front vowel. When the verb takes the ending ta, it realizes as [hada], not *[hæda]. Simply saying that the palatal glide is deleted when it is followed by a consonant does not solve the problem as many verbs have a front vowel at the last syllable of the stem when it takes the ending ta: [mæ:da] ‘to tie’, [sæ:da] ‘to leak’.

Now, recall that ɨ is the only vowel that is known to be not fronted when followed by the palatal glide wordinitially. Supposing that the verb ha has /ẙ/ after the vowel, the behavior of the vowel appears to be parallel to that of ɨ. This fact, coupled with the fact that the vowel [ɨy] appears only wordinitially, allows us to suppose the following rule.

ɨ → a /S_ẙ.

This gives us the underlying form of ha to be /xɨẙ/. The ending a~ʌ harmonizes into [ʌ] because the last vowel of the stem is ɨẙ. This correctly predicts the surface form to be [hayʌ].

Additionally, to capture the fact that it takes ta to form [hada], not *[hayda] nor *[hæda], after obtaining the low central vowel a, delete preconsonantal ẙ.

ẙ → 0 /SV_C.

With /ɨẙ/ being [a~ay], the vowel system is summarized as follows:

2.2.6. Tensity

Tense consonants are obtained from a consonant followed by the glottal fricative h, which is interpreted as /x/ in our analysis, e.g., [sæŋgak] ‘ thought’ + [hada] ‘ to do’ → [sæŋgak̄ada] ‘to think’. Thus, it is imperative to assume the following rules:

C → [+tense] /_x.
x → 0 /[+tense]_.

Now all tense consonants are removed from our underlying inventory, and all underlying segments are characterized by just four features: [±continuant], [±voice], [±grave], and [±compact].

2.2.7. Tones

The first syllable of a word receives a high tone if the syllable’s onset is a fricative or a tense consonant (T. Cho, et. al. (2002), p. 220).

Since tensity is given by postconsonantal x, we can simplify the previous statement as follows:

V → V́ /#(C)[+continuant, +obstruent](G)_. (The high tone is represented by the acute accent.)

That is, if the syllable includes x or s before the nucleus, it receives a high tone.

2.3. Consonant clusters

2.3.1. Postconsonantal tensing

All consonants are tense obstruent after an obstruent. Since in our inventory all voiceless consonats are obstruent, this restriction is adequately captured by a rule as simple as the following. (Recall that tense segments never appear before a consonant.)

C → [+tense] /[–voice]_.

This correctly explains the tensity found in the previously mentioned case of [küyʌpd̄a] ‘is cute’ in a way such as /kwɨẙẙʌẘ&ta/ → güyʌẘda → güyʌẘd̄a → [küyʌpd̄a]. The verbal ending /ta/ gains tensity because it is after a voiceless segment, in this case /ẘ/.

2.3.2. Preconsonantal sonorization

Now, consider the following facts:

Our rule ensures that all consonants before a consonant is closed. This means that there may be two sources of preconsonantal obstruent: One is voiceless continuant, ẘ, s, or x. The other is original voicelss stops, p, t, or k.
A consonant following a sonorant consonant can take all four forms: the sonorant, the lax obstruent, the tense voiced, the tense voiceless. (a) If the tensity is given by the voicelessness of the preceding consonant, we are forced to postulate a rule that turns a preconsonantal voiceless stop into a sonorant. (b) If the tensity is not given by the preceding consonant, we have to assume a sequence of three conescutive consonants for every post–sonorant consonant tensity.

Between the analyses (a) and (b), which is the better option? Recognizing that preconsonantal continuity would be redundant under the analysis (b), it seems that (a) is the truer analysis. That is:

[–continuant] → [–obstruent] /_C.

In fact, preconsonantal sonorization is observed in a few verbs such as the following:

/sẙɨẙt&ta/ → žitda → žitd̄a → [šind̄a] ‘to sit’.

2.3.3. Postconsonantal debuccalization

The glottal fricative h or ɦ is obtained from x followed by a vowel, but this process is not available when x is after a consonant since x in this position tensifies the preceding consonant and deletes itself.

Incidentally, as the velar nasal ŋ never appear after a consonant, we are led to assume postconsonantal ŋ debuccalizes into h or ɦ.

ŋ → ɦ /C_.

2.3.4. Summary of consonant clusters

Under our analysis, the phonotactics of Korean can be summarized as the following:

Notice that preconsonantal velar plosive k is only available without extra cost under the assumption that [h~ɦ] is underlyingly /x/.

2.4. References

3. Summary of the phonemic system

Rules are applied in the order they are given.

As a sample derivation, I show how the following sentence, taken from Shogakukan Korean–Japanese Dictionary, changes its form as it iterates rules.

어린아이들의 장난을 보고 자지러지게 웃었다.
/ʌǰẙɨẙdaẘẙɨẙtɨǰʌẙ | čagdadɨǰ | po&ko | čačẙɨẙǰʌčẙɨẙkʌy | ẘɨ:s&ʌs&ta/

3.1. Initialization

Although only four features, [±continuant], [±voice], [±grave], and [±comapct], are required to be specified in the lexicon, rules are sensitive to other features as we have seen in earlier sections. For this reason, all underlying segments are assumed to receive initial values of several other features at the beginning of the derivation.

Fill the feature values according to the following matrix.

어린아이들의 장난을 보고 자지러지게 웃었다.
ʌǰẙɨẙdaẘẙɨẙtɨǰʌẙ | čagdadɨǰ | po&ko | čačẙɨẙǰʌčẙɨẙkʌy | ẘɨ:s&ʌs&ta

‘C’, ‘V’, ‘G’, and ‘S’ are shorthands for [+consonantal], [–consonantal, +vocalic], [–consonantal, –vocalic], and a segment (as opposed to junctures), respectively.

3.2. Lone consonant lenition

We employed a boundary symbol <&> to indicate a morphophonemic boundary found in verbs such as /wɨs&ko/ [utḡo]. Before the boundary symbol is deleted, the consonant after the boundary needs to be lenited as can be seen in the process: /wɨs&ko/ → wɨs&go → usgo → [utḡo].

[C, +voice] → [–obstruent] /{&, #, V, G}_{&, #, V, G}.
[C, –voice, –continuant] → [+voice] /{&, #, V, G}_{V, G}.
& → 0.

As shown, these leniting rules affect all lone consonants.

ʌŕẙɨẙnaẘẙɨẙtɨŕʌẙ | čagdanɨŕ | po&ko | čačẙɨẙŕʌčẙɨẙkʌy | ẘɨ:s&ʌs&ta
ʌŕẙɨẙnaẘẙɨẙdɨŕʌẙ | ǰagdanɨŕ | bo&go | ǰaǰẙɨẙŕʌǰẙɨẙgʌy | ẘɨ:s&ʌs&da
ʌŕẙɨẙnaẘẙɨẙdɨŕʌẙ | ǰagdanɨŕ | bogo | ǰaǰẙɨẙŕʌǰẙɨẙgʌy | ẘɨ:sʌsda

3.3. Vowels and glides

We obtain the front and rounded vowels i, ü, e, ö, æ, and u, from our underlying vowels ɨ, o, a, and ʌ.

3.3.1. Long vowel adjustment

ʌ: → ə:.
ʌ → ə /_ẙ.

This may be generalized with respect to the weight or length of the vowel as the sequence əy would have two morae.

3.3.2. Fronting and lowering

[V, –diffuse] → [+acute] /_ẙ.
ɨ → i /ẙ_ẙ.
ɨ → a /S_ẙ.
ẙ → 0 /[V, +acute]_.

ʌŕẙɨẙnaẘẙɨẙdɨŕæẙ | ǰagdanɨŕ | bogo | ǰaǰẙɨẙŕʌǰẙɨẙgæy | ẘɨ:sʌsda
ʌŕẙiẙnaẘẙiẙdɨŕæẙ | ǰagdanɨŕ | bogo | ǰaǰẙiẙŕʌǰẙiẙgæy | ẘɨ:sʌsda
ʌŕẙiẙnaẘẙiẙdɨŕæẙ | ǰagdanɨŕ | bogo | ǰaǰẙiẙŕʌǰẙiẙgæy | ẘɨ:sʌsda
ʌŕẙinaẘyidɨŕæ | ǰagdanɨŕ | bogo | ǰaǰẙiŕʌǰẙigæ | ẘɨ:sʌsda

3.3.3. Rounding

Now that we obtained front vowels i, e, and æ, what is left is rounded vowels ü, ö, and u by prefixing ẘ.

V → [+flat] /ẘ_.

ʌřẙinaẘyidɨřæ | ǰagdanɨř | bogo | ǰaǰẙiřʌǰẙigæ | ẘʉ:sʌsda

3.3.4. Backness adjustment

The rounding rule generates ẘʉ from ẘɨ. We now need to adjust the backness to obtain u from ʉ. This backness adjustment is generalized with /ẙɨ/ [yu]. That is, a high central vowel becomes back and rounded when it follows a glide.

[V, +diffuse, –acute] → [+grave, +flat] /G_.

ʌŕẙinaẘyidɨŕæ | ǰagdanɨŕ | bogo | ǰaǰẙiŕʌǰẙigæ | ẘu:sʌsda

3.3.5. Onglide ẘ-deletion

After obtaining rounded vowels, the rounded glide ẘ is deleted before a rounded vowel.

ẘ → 0 /_[V, +flat].

This also ensures that w never appear before o.

ʌŕẙinaẘyidɨŕæ | ǰagdanɨŕ | bogo | ǰaǰẙiŕʌǰẙigæ | u:sʌsda

3.3.6. Offglide ẘ-deletion

The labial glide ẘ works as a syllable boundary before ẙ. Without the ẘ-shield, the grammar would not know whether the palatal glide ẙ is an onglide or an offglide (Kim 1968).

ẘ → 0 /_ẙ.

ʌŕẙinayidɨŕæ | ǰagdanɨŕ | bogo | ǰaǰẙiŕʌǰẙigæ | u:sʌsda

3.3.7. Onglide ẙ-deletion

Only after deleting the separator ẘ, delete ẙ before i.

ẙ → 0 /_i.

ʌřinaidɨŕæ | ǰagdanɨŕ | bogo | ǰaǰiŕʌǰigæ | u:sʌsda

3.4. Tones

If the first syllable of a word has a fricative before the nucleus, it receives a high tone.

V → V́ /#(C){x, s}(G)_.

This has to be applied before x is deleted.

ʌŕinaidɨŕæ | ǰagdanɨŕ | bogo | ǰaǰiŕʌǰigæ | u:sʌsda

3.5. Consonant clusters

3.5.1. Postconsonantal tensing

# → 0.
C → [+tense] /[C, –voice]_.

After applying this rule, tensity of sonorants must be neutralized.

ʌŕinaidɨŕæ | ǰagdanɨŕ | bogo | ǰaǰiŕʌǰigæ | u:sʌsd̄a

3.5.2. X-tensing

Consonants are tensed when followed by the velar fricative x, which is then deleted.

C → [+tense] /_x.
x → 0 /C_.

ʌŕinaidɨŕæ | ǰagdanɨŕ | bogo | ǰaǰiŕʌǰigæ | u:sʌsd̄a

3.5.3. Pre– and post–consonantal sonorization

[–continuant] → [–obstruent] /_C.
[C, –tense, +voice] → [–obstruent] /C_.

ʌŕinaidɨŕæ | ǰaŋdanɨŕ | bogo | ǰaǰiŕʌǰigæ | u:sʌsd̄a
ʌŕinaidɨŕæ | ǰaŋnanɨŕ | bogo | ǰaǰiŕʌǰigæ | u:sʌsd̄a

3.5.4. Sonorant nasalization

This step is necessary for palatalization of sonorants as before filling the nasality values, the liquid and the nasals are kept apart only by the place features.

[C, –obstruent, {+grave, –compact}] → [+nasal]

3.5.5. Preconsonantal closing

[–vocalic] → [–continuant] /_{C, &, #}.

ʌŕinaidɨŕæ | ǰaŋnanɨŕ | bogo | ǰaǰiŕʌǰigæ | u:sʌtd̄a

3.5.6. Velar debuccalization

The velar sonorant ŋ is debuccalized into the glottal fricative ɦ after a consonant. While ŋ is the only source of postconsonantal ɦ, this rule also ensures that the phonotactically prohibited postconsonantal ŋ never surfaces.

ŋ → ɦ /C_.
x → h /{#, +, V, G}_{V, G}.

ʌŕinaidɨŕæ | ǰaŋnanɨŕ | bogo | ǰaǰiŕʌǰigæ | u:sʌtd̄a

3.5.7. Prevocalic voicing

Lax segments including glides are voiced before a vowel or glide.

[–tense] → [+voice] /_{G, V}.

While this ensures that all glides and lax consonants are voiced before a vowel or glide, this also voices the alveolar fricative s, which tends to be voiceless on the surface. This receives treatment at some later stage of derivation.

ʌŕinaidɨŕæ | ǰaŋnanɨŕ | bogo | ǰaǰiŕʌǰigæ | u:zʌtd̄a

3.5.8. Devoicing

Wordinitial lax consonants and alveolar fricatives are devoiced.

[C, –tense] → [–voice] /#_.
z → s.

ʌŕinaidɨŕæ | čaŋnanɨŕ | pogo | čaǰiŕʌǰigæ | u:sʌtd̄a

3.6. Surface adjustment

3.6.1. Palatalization

Dental sonorants and fricatives are palatalized before y or i.

ŕ → r.
[C, –grave, {+sonorant, +continuant}] → [+compact] /_{y, i}.

ʌrinaidɨræ | čaŋnanɨr | pogo | čaǰirʌǰigæ | u:sʌtd̄a
ʌŕinaidɨræ | čaŋnanɨr | pogo | čaǰirʌǰigæ | u:sʌtd̄a

3.6.2. Front vowel hight adjustment

Front vowels receive [–compact] unless they are at the first syllable. This ensures that æ and e don’t contrast at a noninitial syllable.

[V, +acute] → –compact /$(C)(G)_. (‘$’ stands for a syllable.)

ʌŕinaidɨre | čaŋnanɨr | pogo | čaǰirʌǰige | u:sʌtd̄a

3.6.3. Diphthongization

Rounded low vowels are diphthongized.

0 → w /_[V, +compact, +flat].
[V, +compact] → –flat /w_.

Diphthongizations of mid and high vowels have some correlations with the preceding consonant. As Yu Cho (2016) admits, the observations do not allow a neat analysis.

3.7. References

4. Conclusions

4.1. Conclusions

It has been shown that the traditional nineteen-consonant analysis is reorganized into the present ten-consonant system.

Further, it has been shown that the entire segmental system requires only sixteen segments, including four vowels and two glides, characterized by just four binary features.

4.2. Supplementary remarks

The vowel ɨẙ might be further analyzed to be [ɨy~a~ay~ʌ~e] when the h–irregulars and ʌ–irregulars are taken into consideration.

The h–irregulars are those verbs that show the following alternations when taking the endings ta and a~ʌ:

[irʌt̄a] – [ire] (not *[irʌ]) ‘is like this’
[ḡamat̄a] – [ḡame] (not *[ḡama]) ‘is black’

Similarly, the ʌ–irregulars are those verbs that show the following alternations:

[kɨrʌda] – [kɨre] (not *[kɨrʌ]) ‘to do like that’
[ʌǰ̄ʌda] – [ʌǰ̄e] (not *[ʌǰ̄ʌ]) ‘to do how’

It seems that the second vowels of those verbs are in harmony with the first vowel; if the first vowel is either a or o, then the second vowel is a, and ʌ otherwise. Also, it seems that when the ending a~ʌ is attached, the otherwise latent y appears and the ending a~ʌ is deleted.

Hence, for a two-syllable stem that has /ɨẙ/ at the second syllable:

If the harmonizing ending a~ʌ, which is /ɨẙ/, is attached, delete it.
Lower the ɨ of the second syllable to a or ʌ to harmonize with the preceding vowel
If a consonant ending such as ta is attached, delete the last ẙ.

These rules seem to generalize the behaviors of said irregular verbs. As a bonus, by slightly modifying the rules so that they are applied to one–syllable stems, hæ: (or he if æ and e are marged), the modern or shortened form of hayʌ, may be explained.

However, it still fails to capture the following verbs:

[hayat̄a] – [haye] ‘is white’
[b̄uyʌt̄a] – [b̄uye] ‘to blur’

Assuming that the second vowels of these are ɨẙ fails to capture [ya] and [yʌ]. It becomes then tempting to think that [i] is just /ẙ/, but that would give us another problem: how does the grammar know if /CẙV/ is two syllables or one?

Further analyses may or may not solve these problems. (Edit: I did it.)

4.3. About this article

This article is a reorganization and modification of two of my ealier preprints though the morphophonemic analyses in the second one is largely omitted: