Extra-graphemic phonemes in Sumerian

Aleksi Sahala 2007-2010
University of Helsinki (updated 4.2.2011)


Sumerian phonology has puzzled the Assyriologists and Linguists since the language was rediscovered in the 1850s. Traditionally the Sumerian phoneme inventory has been considered to be almost identical to that of Akkadian, which is somewhat unsatisfactory as the languages are not genetically related. The similarities are most likely a result of the decipheriment process, which has relied strongly on Sumero-Akkadian lexical tablets written by Akkadian scribes and scholars.

The study of Sumerian phonology utilizes following methods:

1) Examining inconsistencies in writing, however excluding scribal errors, dialectal variation and diachronic change. For example, for some Akkadian scribes a foreign phoneme has sounded X and for some Y, the truth being somewhere between.

2) Examining graphemic concordances: When an identified morpheme such as /-a/ is systematically scribed in two different ways in a similar looking context, it is possible that both variants represent their own phonemes. E.g. if /-a/ is scribed both Ca1 and Ca2 after C, it may indicate that the suffix preceeding C may in fact represent two different consonant phonemes C1 and C2.

3) By examining sound changes in loan words between the Sumerian and Akkadian. E.g. Sumerian ~ Akk. ru ~ du may indicate that the Sumerian word had a consonant that sounded both, /r/ and /d/ to Akkadians.

Below, I will go through each suggested phoneme and their possible attestations. Colored entries are mentioned in the source books, uncolored are my own suggestions. Discussed phonemes are /dr/, /zd/, /gb/, /l2/, /r2/, /H/, /o/, /ĩ ã ẽ/, /ü ä ə/, /ph th kh/, nl w j/.

Statistical data has been calculated from PSD and ETCSL corpora.

1. Phoneme /dr/

The phoneme /dr/ is attested in syllable onset and coda, and can be speculated to exist in (1) words having r ~ d in free alternation; (2) Sumerian words which have underwent d ~ r change after being borrowed into Akkadian; (3) words where the next morpheme with initial /a/ or /e/ is scribed -rá and -re6 instead of -ra, -da, -dè, -re(2).

Exact quality of the phoneme is not know, but a single thrilled /r/ or dental fricative has been suggested by numerous scholars.

1.1. Onset

Word Reconstr. Source / Note
'to build' /*dru/ Sum. na4-dù-a 'erected stone' -> Akk. nâru 'stele'
/*dra/ Represents /a/ initial morphemes after /dr/.
re6 /*dre/ Represents /e/ initial morphemes after /dr/.
adab '~bird' /*adrab/ Syllabic a12-rá-bu, Akkadian arabû

1.2. Coda

Word Reconstr. Source / Note
bad, ba9 'to be remote' /*badr/ Morpheme distribution: -rá (22%) ~ -ra (3%)
dud, du6 'mound' /*dudr/ du6-du6-rá; Possibly dud ~ dul variation.
enkud 'tax collector' /*enkudr/ Following /a/-morphemes marked -rá
gud, gu4 'ox' /*gudr/ Following /a/-morphemes marked -rá
keš(e)d 'to bind' /*keš(e)dr/ Morpheme distribution: -rá (45%) ~ -ra (1%)
kud, ku5 'to cut' /*kudr/ Morpheme distribution: -rá (35%) ~ -ra (0%)
pad 'to break' /*padr/ Morpheme distribution: -rá (29%) ~ -ra (0%)
par, pa5 'canal' /*padr/ -> Akk. pattum 'canal'
sud 'to be distant' /*sudr/ Morpheme distribution: -rá (30%) ~ -ra (1%)
sukud '~headdress' /*sukudr/ In bird name: ka-sukud-da ~ ka-sukud-rá
kur9, ku4 'to enter' /*kudr/ Very unlikely. Rarely attested with -rá
šukur 'ration' /*šukudr/ Attested with -rá in ED III, later with -ra.

2. Phoneme /zd/

Suggested by Alster (1972 p. 352). Very rarely attested and thus may not necessarily represent a phoneme of its own.

In Emesal, main dialect /d/ is often replaced by /z/, e.g. dùg ~ zé-eb 'good', udu ~ e-zé 'sheep'. It is unclear if this change is related somehow to /zd/.

2.1. Attestations

Word Reconstr. Source / Note
uz 'goose' /*uzd/ Genitive uz-da, adessive uz-dè

3. Phoneme /gb/

Supported by numerous scholars and discussed in detail in M. Civil (1973 pp. 59-61). The exact quality of this phoneme is not known, but a labiovelar stop /gw/ (in ePSD /gb/) has been suggested. The phoneme can be recognized in words having /g/ ~ /b/ alternation.

It is also possible that /gb/ did not represent a distinct phoneme, but an allophone of /g/ occurring before round vowels. However, as it has also been (rarely) attested before /i/ and /a/, we should either consider it as a phoneme, or alternatively speculate that /a/ and /i/ could also represent round vowels, perhaps [ǎ] and [].

In Emesal, the main dialect's /g/ is regularly represented as /b/ (e.g. igi ~ i-bi 'eye'). According to Civil (1973) Emesal sound changes are not related with the /g/ ~ /b/ alternation in Emegir and thus all Emesal words have been excluded from my list.

All colored entries, unless marked with ePSD are from M. Civil.

3.1. Word initial

Word Reconstr. Source / Note
buru14 'harvest' /*gburu/ Alternation: buru14 ~ gur7 (ePSD)
buru7 'fruit; flower' /*gburun/ Alternation: buru7 ~ gurun
buru4 'crow' /*gburu/ Alternation: buru4 ~ gu-úru
buniĝ 'bowl' /*gbuniĝ/ -> Akk. kuninum 'a basin for liquids' (ePSD)
ugu-dù 'tuft of hair' /*ugbudru/ -> Akk. abbuttu '~hair style'
úbúr '~plant' /*gbur/ Alternation: úbúr ~ úkur
kuru7 'inspection' /*gburu(m)/ Alternation: kuru7 ~ bu-ru, gú-ru(-um)
bir6 'shred' /*(i)gbir(i)/ Alternation: bir7 ~ gi-ir, ki-ir
'to blow' /*gbun/ Alternation: bún ~ bu-ú, gu-u
gišbunin 'trough' /*gbunin/ Alternation: bu-ni-in ~ gu-ni-in

3.2. Intervocalic

Word Reconstr. Source / Note
abrig 'cultic functionary' /*agbrig/ Alternation: abrig ~ abrig
a-gar5 'lead' /*agbar/ Alternation: a-gar5 ~ a-bár
lu-lu-bu-na 'cloak' /*lulugbuna/ Alternation: lu-lu-bu-na ~ lu-lu-gu-na
uburrim '~priest' /*ugburrim/ Alternation: uburrim ~ ugurrim
zubud '~weapon' /*zugbud/ Alternation: zubud ~ zugud

3.3. Final

Word Reconstr. Source / Note
ur-zíb 'lion' /*urzigb/ Alternation: ur-zíb ~ ur-zig
sig-b '~' /*siggb/ Varying declination: sig-ba ~ sig-ga
šu-ru-ub 'to dry up' /*šurugb/ Alternation: šu-ru-ub-ba ~ šu-ru-ug-ga

4. Phoneme /l2/

Similarly to phoneme /dr/, /l2/ can be recognized by examining vowel initial morphemes after words ending in graphemic L. For example, /enlil-ak/ is scribed en.líl-lá but /til-ak/ til-la.

The phonetic quality of /l2 is unknown, but a dental lateral has been suggested.

4.1. Distribution of -lá and -la

Word Reconstr. Source / Note
líl 'spirit; air' /*liL/ -lá 100% (Edzard)
gal5 '~demon' /*gaL/ gal5-lá
asila 'joy' /*asiLa(la)/ asil(1,3)-lá
pel 'defile' /*peL/ -lá >99%
kisal 'courtyard' /*kisaL/ -lá ~66%, -la ~33%
gibil 'new' /*gibiL(a)/ -lá ~60%, -la ~40% (Edzard)
bil(1,2,3) 'burn' /*biL/ -lá ~33%, -la ~66%

It also seems that [l] is randomly dropped in auslaut. Th. Jacobsen, 1957 p. 92 and I.M. Diakonoff. 1976 p. 111. have suggested that the the retained and dropped [l] represented different phonemes. The apocope is most likely not related with the feature described in 4.1. as there is no coherence between the distribution of stable and unstable [l] and the distribution of -lá and -la.

I consider it possible that the apocope occurred in words which were originally monosyllabic. The few exceptions could be a result of analogy. However, this hypothesis cannot be validated satisfactorily unless the whole vocabulary and all attestations are studied carefully.

In the tables below, I have collected some examples of unstable and stable [l], excluding verbs which regularly drops the final vowel in marû, e.g. tìl ~ ti-ti 'to live'. Colored entries are listed in Thomsen 1984, p. 45.

4.2. Unstable lateral (~60% monosyllabic)

Variants Notes
lulu > lul > lu5 '(to be) false'
sùl > su6 '~fiber'
dul6 > du6 'hill?; steppe?; ~fish' Also variation: dul ~ dud (?)
lal > lá 'small'
lál > la(1,2,5?) 'to harness; strap'
bíl > bi5 'sour'
bila? > bil > bí 'to burn' -lá ~33%, -la ~66%
mala > mal > mà '?'

4.3. Stable lateral (~20% monosyllabic)

Variants Notes
asilala > asìlal > (a)sila > asil 'joy' Word seems to have both, stable and unstable [l]
bala > bal '~dig'
dala > dal '~insect'
dili > dil 'one; single'
dùl '?'
gala > gal 'big'
galla(x) > gal5 '~demon' gal5-lá /*gaLLa/
ĝala7 > ĝal 'to exist'
gibila > gibil 'new' also gibir
gulu > gul 'destroy; carve'
húla > húl 'to rejoice'
kisal 'courtyard' -lá ~66%, -la ~33%
líla > líl 'spirit; air' -lá 100%
melix > mèl 'neck'
mulu > mul 'to shine'
mùl '~insect' (poorly attested)
sala > sal ~ šal 'fine; thin'
sìla > sìl 'street; unit of measure; cut'
dúl, túl 'fountain'
zil, síl 'to boil'

5. Phoneme /r2/

In symmetry to liquids /l/ and /l2/, Sumerian may have also had two r-sounds /r/ and /r2/. The second tremulant has been suggested to exist in few words where Sumerian /r/ is represented as /h/ in Akkadian. Phoneme may have been uvular /R/ as in French.

5.1. Attestations

Word Reconstr. Source / Note
huš 'red' /*Ruš/ Akkadian ruššu, huššu.
šuhuš 'root' /*šuRuš/ Akkadian šuršu.

Similarly to /l/ and /l2/, also the other tremulant dropped in auslaut. Tremulants are mostly stable except in the following:

5.2. Unstable tremulant

Variants Notes
-ra > -r > -0 (dative suffix) Often omitted and placed in the prefix chain of the verb.
kurx > ku4 'to enter' Following suffix is rarely written -rá
zir > zi 'to break' Syllabic zi-ir ~ zi

There are also occasions where r alternates with l.

5.3. Alternation

Variants Notes
buluh ~ buru8 'vomit'
rib ~ líb 'surpassing' (Thomsen 1984, p. 45)
gibil > gibir 'firewood' (Thomsen 1984, p. 45)
bar ~ bala 'outside; back'

6. Phoneme /H/

Phoneme /H/ can be traced back in few Sumerian words, mainly those where 1) the Semitic correspondent has a /h/ where Sumerian has zero, or 2) where the Sumerian vowel initial morphemes having [e] do not assimilate with the preceeding vowel, e.g. /gala+ene/ > gala-e-ne, vs. /ama+ene/ > ama-ne, reconstructed words being thus /*galaH/ and /*ama/.

/H/ represented most likely a laryngeal fricative in contrast to pharyngeal /h/.

The colored entries in table are listed in Edzard 2003, p. 19-20.

6.1. Attestations

Word Reconstr. Source / Note
é 'house' /*He(y)/ Sum. é-gal ~ Ugaritic hkl 'palace; temple'
íd 'river' /*Hid/ Sum. I-di ~ Hebr. hideqqel 'Euphrates'
ú 'grass' /*Hu/ U2 represents /hu/ in Ugaritic
'to pay' /*laH/ la-e /la+ed/
gala 'cantor' /*galaH/ gala-e
u4-da 'modern' /*odaH/ u4-da-e-ne

Similarly to tremulants and laterals, one of the /h/ sounds drops in auslaut.

6.2. Unstable /h/

Attestation Note / Source
duh(u)? > duh > du8 'loosen' (Thomsen 1984, p. 45)
zúh/g > zú 'tooth; flint'
àh > a12 'dry'
buluh ~ buru8 'vomit'

7. Vowel /o/

Due to enormous amount of homonymy and overrepresentation of [u], it is almost certain that the Sumerian language had a phonemic vowel /o/. Unfortunately, as the language has been deciphered through Akkadian having only four phonemic vowels /a i u e/ the distribution of /o/ in Sumerian cannot be clarified.

I believe that some occurrences of /o/ could be also speculated by examining vowel changes in Akkadian words borrowed from the Sumerian, especially those where /a/ is borrowed as /u/ and vice versa.

7.1. Suggestions

Word Reconstr. Source / Note
dumu 'child; son' /*domu/ Unexpected syllabic spelling in Emesal: du5-mu.
u4 'day; time' /*od/ (S.J. Lieberman 1979)
u 'earth; gift; ten; totality' /*o/ (S.J. Lieberman 1979)
ù 'sleep' /*o/ (S.J. Lieberman 1979)
dara4 'goat' /*dorah/ Akkadian turāhu
guru7 'to heap up' /*goru/ Akkadian karû
buluh '~resin' /*boloh/ Akkadian balahhu

8. Nasal vowels /ĩ ã ẽ/

These nasal vowels are very controversial. It is unclear if they represented individual phonemes or not. The only context where they have been presumably attested is the neutral prefix ì written with cuneiform sign NI. In verbal prefix chain bilabial stops nasalize between the neutral prefix and a vowel: /ĩ-bi/ > im-mi-, /ĩ-ba/ > im-ma-, /ã-ba/ > àm-ma- etc. Nasalization does not occur elsewhere.

8.1. Suggestions

Word Reconstr. Source / Note
ì (neutral prefix) /*ĩ/ Nasalizes the following bilabial stop in prefix chain.
a (neutral prefix) /*ã/ Nasalizes the following bilabial stop in prefix chain.

9. Other vowels /ü ä ə/

By observing variant spellings and sound changes in Sumero-Akkadian loans (Sum. an ~ Bab. ēnuti, [*än?]) some scholars have suggested even a 8 vowel inventory (see. Hayes 1997, p. 11). However, most of these alternations are found from Old Babylonian period where the Sumerian language began to fade, thus possibly being misspellings by Akkadian scribes.

10. Aspirated stops /p' t' k'/

It has been frequently suggested that Sumerian stops did not have opposition voiced ~ unvoiced, but voiced ~ aspirated/post-glottalized voiced instead. On graphemic level BA, GA, DA represented most likely neutral stops and PA TA KA the colored ones. The hypothesis is based on the fact that the earliest Akkadian texts did not distinguish /p t k/ from /b d g/ even though the voice opposition existed in the Akkadian language. The reason for this may be that signs representing the Sumerian PA TA KA had stops alien to Akkadians, where the BA DA GA sounded like both, Akkadian voiced /p t k/ and unvoiced /b d g/.

11. Other phonemes /γ nl w j/

There are also numerous indications to even more phonemes in Sumerian. However the number of attestations is very minimal and thus drawing any conclusions is to be avoided.

11.1. Suggestions

Word Reconstr. Source / Note
/nl/ (prenasalized lateral)
nu (negative prefix) /*nlu/ /nu-ba/ > /la-ba/
lugal 'king' /*nlugal/ Ebla. nu-gal (perhaps an Emesal word)
/w/ (approximant)
arda, urdu, arad 'slave' /*war(a)da/ Akkadian wardum
uruk (City) /*wuruk/ Akkadian warka
/j/ (approximant)
ù-ru-ru '~lament' /*jurūru/ Akkadian jarūru
a-a 'father' /*aja/ Also written a-ia
/γ/ (velar fricative)
zalag(UD) > zala /*zalaγ/ UD is also read zalah (ePSD)
zug(KA) > /*zuγ/ KA is also read zuh (ePSD)
ha- (cohortative prefix) /*γa-/ varies with ga-, but only in first person.
ku6-a 'fish' /*γua?/ Also ha 'fish'.
hur-saĝ 'mountain' /*γursaĝ/ Also kur 'mountain'.

12. Sources

ePSD - Electronic Pennsylvania Sumerian Dictionary
Civil, Miguel 1973. From Enki's headaches to Phonology (JNES 32, No.1/2. pp. 57-61). The University of Chigaco Press.
Edzard, Diez Otto 2003. Sumerian grammar. Brill.
Liebermann, S.J. 1978. The phoneme /o/ in Sumerian (AOAT 203). Neukirchen-Vluyn, pp. 21-28. (Unfortunately I have not found this article anywhere)
Thomsen, Marie-Louise 1984. The Sumerian language. Akademisk forlag

Back to index