LINGUIST List 5.206

Tue 22 Feb 1994

Sum: Stem alternation

Editor for this issue: <>


  1. Jason Johnston, 5.156 Sum: Stem alternation (longish)

Message 1: 5.156 Sum: Stem alternation (longish)

Date: Sat, 19 Feb 1994 18:13:17 5.156 Sum: Stem alternation (longish)
From: Jason Johnston <>
Subject: 5.156 Sum: Stem alternation (longish)

Thank you to all those who replied to my query (5-156) about cases where
an inflectional category is realised by 'stem alternation alone', including
prosodic alternation. They are, in the order I received their (first) message:
 Tom Cravens, David Odden, Bob Beard, Larry Hutchinson, Ronald
Feldstein, Andrew Carstairs-McCarthy, J. A. Rea, Richard Ogden, Bruno
Tersago, Tapani Salminen, Stavros Macrakis, Norbert Strade, John E.
Koontz, Marcia Haag, Alec Marantz, Erika Mitchell, Marit Julien.
 Thank you particularly to Tapani Salminen, Larry Hutchinson,
Marcia Haag, Marit Julien and Alec Marantz for offering to send me further

These are the instances suggested by correspondents:

Choctaw (Muskogean, North America) -- formation of the 'intensive aspect'
by stress shift to initial syllable, accompanied by gemination of first
internal consonant and second vowel. For instance, from _falAma_ 'to return'
youget _fAllaama_. (Interestingly iconic for an intensive aspect.) "If a word
has only 2 syllables, /yy/ may be inserted to make the requisite number of
syllables, or alternatively, the medial vowel may be bent [sic] to provide
enough length". Apparently this is the regular formation for all verbs. Also
in Choctaw, "the second person singular imperative is formed by a final
accent on the stem vowel: _binili_ 'sit' becomes _biniLI_ 'sit down!'"
 Information from Marcia Haag.

Lakhota and Omaha-Ponca (Siouan, North America) -- in Lakhota,
inflection of "y-stem verbs as follows:
 ex. yuha' 'to hold, have'

 Agent 1, Patient 3 b-luha'
 Agent 2, Patient 3 luha'
 Agent 3, Patient 3 yuha'
Third person is zero marked. [...] in modern Omaha-Ponca, edh-stems
 ex. dhathe' 'to eat'

 Agent 1, Patient 3 b-dha'the
 Agent 2, Patient 3 na'the
 Agent 3, Patient 3 dhatha'
Third person zero marked. The final vowel shift in the last form is not
person marking per se. It is the mark of the plural or the proximate third
person singular..." Further details "in Boas & Deloria's grammar of Dakota
or in Boas & Swanton's Academy of Sciences sketch of Siouan (Dakota
with Omaha and Winnebago) in the old Handbook of North American
Indian Languages." A diachronic "solution" was suggested by Dorsey c.
 Information from John Koontz.

Maori -- formation of "plurals such as _waahine, maatue, tiipuna_ for
_wahine, matua, tipuna_ 'woman, parent, ancestor'. This is restricted to a
few nouns denoting persons, while most nouns don't distinguish number at
all (so there's no 'rival' number affix); but it is inflectional inasmuch as
number is also expressed obligatorily on determiners and possessives".
 Information from Andrew Carstairs-McCarthy.

(Standard) Italian -- alternations in stress and/or length/gemination in the
verbal system. Examples _pArlo_ 'I speak', _parlO_ 'he/she/it spoke';
_parlerE:mo_ 'we shall speak' _parlerEm:o_ 'we would speak'; _krE:de_
'h/s/i believes', _kredE_ 'h/s/i believed'; _pArti_ 'you ( leave',
_partI:_ 'I left', _partI_ 'h/s/i left'; _kA:de_ 'h/s/i falls', _kAd:e_ 'h/s/i
"All are productive except the ['kad:e] type. Vowel length is predictable (i.e.
long in stressed open syllable); consonant length isn't." I'm not sure what
to make of the form _partI:_ vs. _partI_.
 Information from Tom Cravens.

Italian (dialects) -- in "some north Italian dialects, say Bergamask",
formation of the plural of nouns ending in coronals by "palatalizing" them.
So the plurals of _det, an, al_ 'tooth, year, valley' are _dec, an~, aj_ with
<c>=English <ch>, <n~> as in Spanish, <j>=English <y>. [?Possible not
very abstract source in _det+j, an+j, al+j_]
 Also, "some south Italian dialects that form plurals by vowel change
of the stem, as in Abruzzese /ap/ 'bee' (that's a schwa for a second
vowel) plural /ip/; /dulor/ 'pain' plural /dulur/, and so on." More info
in Gerhard Rohlfs, _Grammatica Storica della Lingua Italiana_, Vol.2
'Morfologia', especially sections 375-6 (= _Historische Grammatik der
Italienischen Sprache_).
 Information from J. A. Rea.

Nilotic languages -- "things that look like broken plurals. Especially in
Dinka, the plural (a nice coincidence) is formed by various random stem
changes (including total suppletion). The kinds of things that happen
include vowel lengthening, shortening, diphthongization, tone change,
voice-quality change, change in the final consonant, monophthongization,
and vowel-quality changes (raising and lowering) [...] Similar things exist
throughout Nilotic."
 Information from David Odden.

Temne (Sierra Leone) -- plural formation by mutation of initial consonant,
e.g. _kuma, tuma_ 'box, boxes' and _ngola, yola_ 'cola tree, cola trees'.
LH believes that an analysis which treats the initial consonants as prefixes
doesn't work. Whereas other cases of consonant mutation may be the result
of a formerly allophonic variation which has lost its conditioning
environment, entailing "close phonetic similarity of the mutating
consonants", "Temne mutation arose by a quite different process, which
accounts for the phonetic dissimilarity of the mutations".
 Information from Larry Hutchinson.

Nenets -- [TS] "the formation of accusative plural (which also functions as a
stem for further morphological processes) is formed by affixation for
consonant stems, but by vowel modification for vowel stems. A basic
substem for verbs known as general finite stem is formed by affixation for
consonant stems and certain vowel stems, but by vowel modification for the
remaining vowel stems, both types of vowel stems being very common."
Example of noun inflection: _wano_ 'root' (with other case/number forms
_wanom, wanoh, wanoq_ etc.) has accusative plural _wanu_ and genitive
plural _wanuq_. TS is aware of a possible morphophonemic analysis but
rejects it: "we can (in principle) derive wanu from *wanou by deleting o or
from *wanoy by changing o to u before y and then deleting y etc. Actually,
both of these approaches create more problems than they solve, and
therefore (and for other more theory-internal reasons) I prefer to claim that
wanu comes into existence via vowel modification of o to u."
 Information from Tapani Salminen.

Dutch -- "stem alteration in the preterit and past participle of the so-called
'strong' verbs", as in German.
 Information from Bruno Tersago.

Russian -- stress alternations such as _vOlos_ 'hair,', _volOs_
'hair,'. "In this case, both instances are usually interpreted as
having a zero ending. This is not the usual, productive gen. plur. ending,
and is considered an isolated relic."
 Information from Ronald Feldstein.

Finno-Ugrian (Finnish, Estonian, Sa(a)mi) -- most mentioned was the
Estonian genitive singular, where the realization may be [ACMcC] "by
truncation of the final consonant (_kuningas/kuninga_ 'king') or
substitution of the strong 'grade' of the stem for the weak (_tarve/tarbe_
'need', _mure/murde_ 'break')."
 RO further mentions that "[m]any (but not all) cases can be
generated by sticking the suffix on to the genitive singular form" and gives
the following reference for a thorough description: Valter Taulli, _Standard
Estonian Grammar_, Part I: Phonology, morphology, word formation. Acta
Universitatis Upsaliensis, Studia Uralica et Altaica 8, Uppsala 1973.
 Something similar goes on in Northern Saami, where [MJ] "the
accusative/genitive of nouns is formed by stem consonant gradation" , and
there are three consonant grades: weak, strong and extra-strong or 1, 2 and
3. In even-syllabled nouns, the acc/gen is in a grade one level down (i.e. 2
if nom is 3, 1 if nom is 2). Examples:
 Nom Sg Acc/Gen Sg
 johka /johka/ joga /joga/ 'river'
 vahkku /vahhku/ vahku /vahku/ 'week'
 guossi /guosssi/ guossi /guossi/ 'guest'
 luossa /luossa/ luosa /luosa/ 'salmon'
[Evidently the orthography doesn't always distinguish strong from extra-
strong grade.] In odd-syllabled nouns, the alternation is weak in nom,
strong in acc/gen, whereas in "contracted nouns" (where a syllable has been
lost, historically) the alternation is weak in nom, extra-strong (if possible)
 In Finnish, where consonant gradation usually [EM] "appears as a
weakening of syllable (p,t,k) onsets when a syllable becomes closed by the
addition of a consonant through affixation or whatever. E.g. luke 'read' +
-n (1SG) -> (lu.ken) -> lu.en", there are instances where consonant
gradation occurs without an apparent triggering affix. For instance,
"Imperative = (Consonant Gradation): luke -> ( -> lu.e". EM supports
an analysis which posits a "phantom consonant" to trigger the alternation,
citing as evidence "the fact that these phantom consonants can actually be
heard (at least in Finnish); they take on the place and manner of following
initial consonants, or else are realized as glottal stops: lue[s] se! 'Read
lue[?] artikkeli! 'read the article'." It is not clear if a similar "phantom-
consonant" analysis would be feasible for Estonian.
 Speaking of these Finno-Ugrian gradations, which are historically
phonological in motivation but arguably not so synchronically, NS observes
that "[t]he formerly agglutinating languages are being transformed into
inflecting ones."
 Information from Andrew Carstairs-McCarthy, Richard Ogden,
Marit Julien, Erika Mitchell and Norbert Strade.


Several correspondents asked the nature of my interest in this question. It is
simply to gather data which may be crucial in assessing a currently
proposed - and disputed - claim.

 A strong version of this claim is that inflectional categories always
have a 'principal exponent' (other apparent exponents being conditioned
side effects) and furthermore that this principal exponent is always an affix
(or lexical item or Vocabulary Item or call it what you will - but in any case
a *thing*). This claim is made by proponents of Distributed Morphology,
including Morris Halle, Alec Marantz and Ralph Noyer. By contrast,
proponents of A-Morphous Morphology, notably S.R. Anderson, deny that
there is any fundamental difference between affixation and other forms of
'realization', which are all considered to be the side effect of the operation
of rules.

 A slightly weaker version of the claim would be that, even if both
affixation and various forms of mutation or stem alternation can expound
inflectional categories, nevertheless they behave differently with respect to
various principles of the grammar. For instance, Andrew Carstairs-
McCarthy has argued that only rival affixes, not rival stem alternants or
prosodies, count for the purposes of his Paradigm Economy Principle.

 Two factors complicate the evaluation of evidence for these claims.
The lesser, I think, is that even the most processual looking alternations
can, with varying degrees of ingenuity, be reformulated as 'things', e.g.
CV skeletons, morphemic tiers, floating autosegments, what have you,
even if some 'overwriting' of existing structure is called for. For instance,
one of my correspondents mentioned the analysis of the Arabic broken
plural (my own example) as containing a 'discontinuous infix'. My own
feeling with these was that, as the precise choice of vocalism is a lexically
idiosyncratic property, i.e. it has to be listed anyway, they're reasonable
candidates for 'stem alternation', even if some internal structure can be
discerned and described.

 A more important problem is the possibility of an analysis in terms
of zero morph(eme)s. Clearly if we are free to posit zero morph(eme)s very
freely, the claim that inflectional categories are always realized by affixation
is unfalsifiable, since we can always say a zero affix is doing the realization
and merely triggering the stem change. Proponents of Distributed
Morphology do not, to their credit, throw zero morph(eme)s around with
complete abandon, but they do allow them under certain theory-defined
circumstances. Others, like my correspondent Tapani Salminen, are quite
suspicious of their use altogether.

 In any case, that was the point of my request for "examples where,
for one reason or another, there is little or no evidence to posit zero
morphemes." The most convincing examples would be those where even
the linguist who is fairly free (though obviously not totally unconstrained)
in positing zero morphemes would be disinclined to do so: cases, for
instance, where there is a lack of any parallel forms with overt affixation.
The most interesting example from that point of view is the Choctaw
'intensive aspect' described to me by Marcia Haag. But in this case,
curiously, the very simplicity and regularity of the prosodic/templatic
change involved makes analysis as some sort of 'affix' relatively appealing.

 Finally, I could have been clearer about what I meant by 'stem
alternation'. I leave the last word to Alec Marantz:

>What I find interesting about the claim that stem modifications can
>realize inflectional features is that, for the most part, what are
>discussed are root modifications. The distinction with affixation is
>that the inflectional affixes stack up and you'll find a certain number
>of affixes in a row. So an outer affix clearly is attaching to a
>combination of the root and the inner affixes. Stem modifications don't
>apply to combinations of root plus inflectional affixes, as far as I can
>tell. I think that this observation alone is sufficient to sink an
>Anderson-style "affixation is just one of several possible modifications
>to a stem performed in the presence of inflectional features" analysis
>of inflectional morphology.

Please contact me if you want bibliographic references to the writers I've
mentioned. I've omitted them only to save some space in an already longish

Jason Johnston
Dept of Linguistics, F12
University of Sydney, 2006
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue