Andrew Carstairs-McCarthy: on Morris Halle on Latin verbs

Fri Nov 19 15:49:16 UTC 1999

I would like to compare the interesting suggestions by Morris Halle and
David Embick on Latin verb inflection with some ideas of mine, published
over the years, on the same topic.  What do DM-listers think of the
differences between two approaches?

1.  The 3rd, mixed and 4th conjugations

First I will comment on Halle & Embick's point (3), about the difference
between 3rd-conjugation and 'mixed' verbs.  H&E suggest:

"The /i/ that appears in third conjugation verbs such as

	ducere -- ducimus

is underlyingly different from the /i/ that appears in

	capere -- capimus.

The latter is [-back], but the former is [+back]."

H&E's suggestion belongs to a tradition of distinguishing between the 3rd,
mixed and 4th conjugations in terms of the underlying phonology of the
theme vowel (TV) or theme glide (TG).  An earlier suggestion of this kind
is that of Rochelle Lieber (1981).  I will set it alongside H&E's
(substituting /k/ for Latin orthographic _c_):

             1SgPres  Infinitive  Lieber   Halle & Embick
3rd conj    du:ko:   du:kere     TG y     TV i[+back]
mixed conj  kapio:   kapere      TV i     TV i[-back]
4th conj    awdio:   awdi:re     TV i:    TV i:[-back]

Probably both of these analyses can be made to work, provided one posits
appropriate deletion rules.  For H&E, the rule must apparently delete a
[+back] vowel if followed by a vowel.  (I think that that must be what
Halle means, though he what he says is 'a rule that deletes a vowel if
followed by a [+back] vowel'.)  A question then arises about whether or not
such a rule can be related to some phonological universal (nowadays
perhaps, one of the universal inventory of violable constraints), or is
just a Latin idiosyncrasy.  But there are further questions, and to my mind
more fundamental ones:
(a) Does this analysis really accord with all the relevant facts of Latin?
(b) How does this analysis fit in with what is independently known about
how inflection class systems are organized?

As to question (a), the relevant facts include the fact that membership of
the mixed and 4th conjugations is much more fluid than that of the 1st, 2nd
and 3rd.    Firstly, there is a tendency in early Latin texts for 'mixed'
verbs to be assigned to the 4th conjugation, and the fact that this merger
resurfaces in modern Romance languages, along with other features of early
Latin, suggests that it persisted as a colloquial feature throughout the
Classical period.  For example, Italian _fuggire_ and French _fuir_ point
towards a 4th-conj proto-Romance ancestor /fugi:re/, in contrast to the
mixed-conj Classical Latin /fugere/ 'to flee'.  Secondly, the mixed and 4th
conjugations are largely distinct as regards the phonological shapes of
roots.  We have:

monomoraic roots ending in obstruents    mixed conj
polymoraic roots                         4th conj
monomoraic roots ending in sonorants     some mixed, some 4th

(For the facts, with references to philological authorities, see
Carstairs-McCarthy 1994:753.)  It looks very much as if different groups of
Latin speakers groped for ways of imposing order on an inflectional pattern
which was really for them as confusing as its *surface* phonology suggests.
But why should this be if, as H&E claim, inflection class membership can be
handled entirely in terms of differences in *underlying* phonology that are
just as clearcut for the mixed and 4th conjugations as for the 1st and 2nd?

As to question (b), H&E ignore evidence that inflectional affixes are
distributed in such a way that children learning them can exploit the
Principle of Contrast proposed by Clark (1993) to account for how
vocabulary is acquired -- provided that *unambiguous* identification of
inflection class can count as part of the information content that an affix
encodes (Carstairs-McCarthy 1994).  (It is easy to imagine inflection class
systems which would not satisfy that proviso, but I am aware of no
clearcut, uncontroversial example of such a system.)  This approach
accounts for the 'paradigm economy' (PE) phenomenon (Carstairs 1987) in a
fashion that makes sense in developmental psycholinguistic terms.  But it
presupposes that inflection classes are real, i.e part of I-language, not
just mere pedagogical conveniences.  So if, like H&E, one proposes an
analysis of Latin verbal inflection class membership solely in terms of TV
differences and their phonological effects, one saddles oneself with the
task of providing an alternative explanation for PE effects.

Noyer (1994) attempts such an explanation, relegating PE effects to a
theory of lexical learnability situated outside UG.  But such an approach
would not discriminate, so far as I can see, between actual Latin, where it
is the mixed and 4th conjugations that are fluid, and a hypothetical Latin
in which it is the 1st and 2nd conjugations that are fluid.  If the
/i[+back]/ theme vowel that H&E posit for /du:ko:/ was genuinely part of
the I-language of some speakers of Latin, why should it not have sufficed
to keep the 3rd, mixed and 4th conjugations securely apart?  The trouble
with Noyer's type of explanation is that it will work in this instance only
if one rejects precisely the kind of abstract phonology that H&E posit!

2.  Realization of Agr

Let me turn now to H&E's point (1):

"The basic morphological structure of the word is

	Main Verb -- ASP -- TENSE -- MOOD -- Agr -- VOICE,

with the special provision that the nodes in caps are filled by unary
features, which have the property of collapsing the node when not
present.  Thus, we have all but the VOICE node realized phonetically
in the Pluperfect Subjunctive: lauda: - ui - s - se: - mus, but none
of them appears in the Present Indicative: lauda:-mus."

As H&E imply, in finite verbforms, it is only the Agr node that is always
'present', i.e. only Person-Number that is always overtly realized.  What
interests me especially about Person-Number is the variety of ways in which
it gets realized; and I would be interested if any DMer has any thoughts
about the distribution of these various realizations.

Following Aronson (1982), let us use 'screeve' to mean 'a set of
Person-Number forms that are identical in all other respects, i.e. Tense,
Aspect, etc.'  So one can talk of e.g. the Perfective Past Subjunctive
Active screeve or the Imperfective Present Indicative Passive screeve.  Now
let us use 'screeveshape' to mean 'a set of affixes used in one or more
screeves'.  Thus one can say, for example, that the Imperfective Past
Subjunctive Active and the Imperfective Future Indicative Active
screeveshapes are the same in the 3rd conjugation but different in the 1st
conjugation.  If one abstracts theme vowels or anything that looks like a
theme vowel, one finds that Latin has just three screeveshapes for Active
non-Imperative verbforms:

o-screeveshape:  o:, s, t, mus, tis, nt
m-screeveshape:  m, s, t, mus, tis, nt
i-screeveshape:  i, isti:, t, mus, istis, e:re

(Notice the 'economy' here: for four of the Person-Number cells there is
more than one affixal option, and there are in fact 3x2x1x1x2x2 (or 24)
ways in which these affixes could be combined into distinct screeveshapes;
yet only three screeveshapes exist.)

Now, where do these three screeveshapes appear?  One's inclination is to
expect that they should divide the labor, so to speak, on straightforward
morphosyntactic lines: one screeveshape should be used for Perfective and
another for Imperfective, or one for Indicative and another for
Subjunctive, or something like that.  But in fact, what we find is the
following (Carstairs-McCarthy 1998):

i-screeveshape:  Perfective Present Indicative
m-screeveshape:  Subjunctive AND/OR Past
o-screeveshape:  elsewhere

What's intriguing is the distribution of the m-screeveshape, which is
'L-shaped' or 'T-shaped': it occupies both the Past 'row' and the
Subjunctive 'column', so to speak.  This could be just an accidental quirk
of Latin.  But it's remarkable that something very like this distribution
crops up also in Turkish.  Turkish also has three screeveshapes (ignoring
vowel harmony), even though, as in Latin, the mathematical upper limit is
much higher:

Shape 1:  im, sin, (dir), iz, siniz, (dir)ler
Shape 2:  m, n, =D8, k, niz, ler
Shape 3:  eyim, esin, e, elim, esiniz, eler

Their distribution is as follows:

Shape 3:  Subjunctive 'Simple'
Shape 2:  Conditional AND/OR Past
Shape 1:  elsewhere

Morphological systems thus seem to favor, or at any rate not to discourage,
disjunctive or 'AND/OR' meanings for screeveshape.  I have some ideas about
why such meanings should not be discouraged (Carstairs-McCarthy 1998), but
none about why they should be (or might be) favored!  Can DM shed any light
on this?

Of course, talking of meanings for *screeveshapes* as opposed to individual
affixes presupposes that screeveshapes are recognized as part of
I-language, just like inflection classes.  But, again, I do not see such
recognition as fundamentally incompatible with DM.

References
Aronson, H. 1982. Georgian; a reading grammar. Columbus: Slavica.
Carstairs, A. 1987. Allomorphy in inflexion. London: Croom Helm.
Carstairs, A. 1988. Paradigm economy: a reply to Nyman. Journal of
Linguistics 24: 489-99.
Carstairs-McCarthy, A. 1994. Inflection classes, gender and the Principle
of Contrast. Language 70: 737-88.
Carstairs-McCarthy, A. 1998. How lexical semantics constrains inflectional
allomorphy. Yearbook of Morphology 1997, ed. by G. Booij & J. van Marle,
1-24.
Clark, E. 1993. The lexicon in acquisition. Cambridge: CUP.
Lieber, R. 1981. On the organization of the lexicon.  Bloomington: IULC.
Noyer, R. 1994. Paradigm structure and lexical generative capacity. NELS
24: 427-41.
___

Having criticized Halle and Embick, it is only fair that I should offer
something of my own for criticism.  I have a couple of drafts on morphology
nearing completion, one (joint with Thea Cameron-Faulkner) called 'Stem
alternants as morphological signata: evidence from blur avoidance in Polish
nouns', and one called 'How an affix can acquire a stem alternant as its
'meaning': evidence from umlaut in German noun inflection'.  I am too
web-incompetent, unfortunately, to have these in a form downloadable from
the web; but I can send them as attachments to anyone who is interested, or
else supply them for archiving on the DM webpage.

Andrew

Andrew Carstairs-McCarthy
Associate Professor
Department of Linguistics, University of Canterbury, Private Bag 4800,
Christchurch, New Zealand
phone (work) +64-3-364 2211; (home) +64-3-355 5108
fax +64-3-364 2969
e-mail a.c-mcc at ling.canterbury.ac.nz
http://www.ling.canterbury.ac.nz/adc-m.html