[tied] Morphology (10/20)

From: Miguel Carrasquer
Message: 14727
Date: 2002-08-29

10. The middle

Besides the active and the stative, PIE also had a category called the middle.
In Ancient Greek, the middle is primarily used to express an action which
affects the subject in some way (as opposed to e.g. the active transitive, where
the action primarily affects the object). Although middle forms can be
transitive or intransitive, the nature of the middle makes a diachronic
transition towards an intransitive or passive meaning understandable, and it is
indeed in that direction that the middle (> the passive) in e.g. Latin has
progressed. But even in Latin, the older situation is preserved in the
so-called deponent verbs, which couple active meaning with passive/middle
morphology. Only Balto-Slavic has completely lost the middle voice.

The endings of the middle are closely related to those of the stative.
Historically, they must at one time have been the same, i.e.:

*-h2- *-mw-
*-th2- *-dhw-
*-0- *-0-

In order to explain the enormous variation that we see in the middle endings of
the various Indo-European languages, it is necessary to go back to the origin of
the middle itself. The curious endings in -r and -m, which are typical of the
middle, provide us with an important clue. The r-forms, as found in Hittite,
Tocharian, Celtic and Italic, have a tendency not to occur in the 2nd person
(Tocharian is the exception, Hittite forms in -r(i) are rare in the 2nd persons,
and the exceptions of Armenian -ar/-aruk` (with -r- from *-d(h)- or *-z-) and
Latin -re/-ris (with -r- from *-z-) are only apparent). The forms ending in -m
occur only in the 2nd and 3rd persons, but never in the first person (the only
exception being Greek *-ma:m > -me:n). Now if the middle were in origin a
conjugation with incorporated indirect object (dative), and if we assume that
*-m is, as elsewhere, a 1st person ending, and that *-r (< **-n ?) is a second
person marker, the distribution of these forms would make perfect sense, since
in conjugational grids involving an indirect object, the first person dative
markers generally do not occur together with a first person subject, nor do the
second person dative markers occur with second person subjects. So, having
identified *-m(i) and *-r(i) as 1st and 2nd person dative markers, the rest of
the grid can be filled in as follows:

Middle grid:

1sg/pl 2sg 3sg 1du 2du 3du 2pl
1 -- -h2ar(i) -h2a(i) -- -h2ah2r(i) -h2ah2(i) -h2adh(i)
2 -th2am -- -th2a(i) -th2ah2m -- -th2ah2(i) --
3 -(t)om -(t)or(i) -(t)o(i) -(t)ah2m -(t)ah2r(i) -(t)ah2(i) -(t)odh(i)

1 -- -h2mWor(i) -h2mWo(i) -- -h2mWah2r(i) -h2mWah2(i) -h2mWodh(i)
2 -h2tWom -- -h2tWo(i) -h2tWah2m -- -h2tWah2(i) --
3 -h2tom -h2tor(i) -h2to(i) -h2tah2m -h2tah2r(i) -h2tah2(i) -h2todh(i)

1 -- -mWor(i) -mWo(i) -- -mWah2r(i) -mWah2(i) -mWodh(i)
2 -dhWom -- -dhWo(i) -dhWah2m -- -dhWah2(i) --
3 -ntom -ntor(i) -nto(i) -ntah2m -ntah2r(i) -ntah2(i) -ntodh(i)
-rom -ror(i) -ro(i) -rah2m -rah2r(i) -rah2(i) -rodh(i)

The structure is: subject-object-indirect object, where the subject markers are:
*-h2, *-th2, *-(t); *-h2mw, *-h2t(W)/*-h2dh(W), *-h2(t); *-mW, *-dhW, *nt/*-r.

The object marker in the above grid is the thematic vowel (*o/*e), ie. a 3rd
person singular object, but surely there were also forms with 3pl. object
agreement, such as the one underlying the attested 1pl. form *-mesdh(W)o(i)
(Hitt. -wasta, Grk. -mestha, OIr. -mis).

The dative person markers are:
*-m, *-n (> *-r), *-0; *-h2m, *-h2n (> *-h2r), *-h2; *-m(W), *-dh(W), *-0?,
optionally followed by the dative/locative suffix *-i. In some languages this
was later associated with the present tense marker *-i (perhaps in origin also a
locative marker), and an opposition was secondarily established between middle
presents in *-(o)i and middle preterites ending in *-o.

When the original dative meaning of the middle was lost, probably before the
break-up of PIE, the individual languages selected different forms from the old
grid to create their new middle paradigms (including new distinctions between
middle past and present), resulting in the apparent chaos of the attested PIE
middle endings:


1. -ha, -hari, -hahari < *-h2a, *-h2ar(i), *-h2ah2r(i)
2. -ta, -tati, -tari < *-th2a
3hi -a, -ari < *-o, *-ori
3mi -ta, -tari < *-to, *-tori
1pl -wasta(ti) < *-mWesdh(o) [3pl. object: *-mW-es-dh(W) + -o]
2pl -duma < *-dhWo
3pl -anta(ri) < *-nto, *-ntori

1. -ha(ha)t(i) < *-h2adh(i)
2. -ta(ti) < *-th2a (+ -dh(i))
3. -(t)a(ti) < *-(t)odh(i)
1. -wastat(i) < *-mwesdho + -dh(i)
2. -dumat(i) < *-dhwo + -dh(i)
3. -antat(i) < *-ntodh(i)

In the 3sg., the forms with -t- are of the mi-conjugation (active), the forms
without -t- belong to the hi-conjugation (stative). The opposition between
present and past tense was created by generalizing *-ri (2sg. dat.) in the
present forms, and *-dhi (2pl. dat.) in the past, leading to such unetymological
forms like 2sg. -tari or 2pl. -dumat. In the 1pl., the form with 3pl. object
and 2pl. dative agreement *-mWesdhi was reshaped to *-mWesdho to better agree
with the other forms, and could subsequently be extended with -ti or -ri
(-wastari, -wastati) to mark the emerging middle tenses.

Tocharian A/B:

-ma:r/-mar < *-m-h2ar
-ta:r/-tar < *-th2ar
-tär < *-tor
-mt(t)är < *-medh(W)or
[-cär]/-tär < *-dhWor
-ntär < *-ntor

-(w)e/-mai < *-(mW-)h2ai
-te/-tai < *-th2ai
-t/-te < *-to
-mät/-mt(t)e < *-medh(W)o
[-c]/-t < *-dhu < *-dhWo
-nt/-nte < *-nto

In Tocharian, as in Hittite, the forms in -r were generalized to build a middle
present, while the forms with 3rd person dative agreement (with -i in the 1/2sg,
without it elsewhere, cf. the Latin perfect) were put to use to mark the middle



1. -da ?
2. -za < *-so
3. -da < *-to
P. -nda < *-medho/*-nto

The 2pl. *-so (*-sa), as in Latin, Greek and Sanskrit, is a normalization of the
ending for original *-th2a. The 1sg. appears to have the 3sg. form, perhaps
following the model of the preterites (strong and weak), where the 1 and 3sg.
forms were also identical. In the plural, the 1st and 3rd persons *-m(e)dho and
*-nto merged phonologically, and expected 2pl. *-dhwo (> *-da) was carried
along, giving -nda as the unitary plural ending of the Gothic middle.


aorist middle:

-ay < *-a:-ai < *-h2ai
-ar < *-a:-dha < *-th2a
-aw < *-a:-to < *-to
-ak` < *-a:-sW < *-mW(e)dho (?)
-ayk`/ -aruk` < *-a:-tesW
< *-a:-dhWesW < *-dhwo
-an < *-a:-nto < *-nto

The middle forms in Armenian are characterized by a suffix a (< *a: < *eh2)
between the stem and the middle endings. This may have originated in the 1sg.
(thematic vowel *-e- + *-h2ai > *-a:i), or it may be the a:-preterite extended
with middle endings.
The 2sg. ending -r may derive from *-th2a(i) if *th2 was weakened to *dh (>
*-r). The 2pl. -a-ruk` (besides analogical -ayk`) clearly points to middle
*-dhW- extended with the plural marker *-sW (> -k`) [-a:dhwesW > -aruk`, cf.
drak^wesW > artasuk` "tears"]. The 1pl. ending -ak` (which was extended to all
preterites in Armenian) is phonologically difficult to explain. One would
expect *-awk` < *-a:-mWosW (like the thematic ending of the present conjunctive,
*-o-mWosW > *-owok` > -uk`, or the noun awr "day" < *awur < *a:m(W)o:r).


(deponent/passive conjunct)
1. -ur, -or < *-o:r
2. -ther < *-th2ar-es (?)
3. -thar < *-t-ro
3P. -(th)ar < *-(t)or
1. -mar < *-mor
2. -ith < *-te[s]
3. -tar < *-nt-ro
3P. -tar < *-ntor

1. -(i)n(n) < *-oih1-m-h2ai
2. -(i)tha < *-oih1-th2a:-s
3. -(i)th < *-oih1-to
1. -(i)mis < *-oih1-mosdho(i)
2. -(i)the < *-oih1-teso(i) (for *-dhwesoi)
3. -(i)tis < *-oih1-nto, analog. -(i)tis after 1pl. -(i)mis

In Old Irish, as in Hittite and Tocharian, the forms with 2sg. dative agreement
in -r were used to make the present tense of the mediopassive. As in Latin,
1sg. -o:r is a regularization of original *-h2ar. The 2sg. -ther is
problematical. We would expect *-th2a(r) > -thar in the conjunct, *-th2ares >
-ther in the absolute. Perhaps the absolute form was generalized to avoid
confusion with the 3rd. sg. dep./pass. -thar < *-tor. The difference between
the Old Irish 3rd. person deponent and passive, besides the retention of stative
*-or > -ar in a number of passive forms, is that the vowel before the deponent
ending -thar is rarely elided, whereas in the passive ending -thar, the vowel is
regularly elided if it used to be the second vowel in the word. This indicates
that in the deponent ending there was originally no vowel between *t and *r,
while such a vowel *was* present in passive -thar < *-tor(i). The origin of
3sg. *-tro must lie in the 3pl. form *-nt-ro (> -tar), a portmanteau formation
based on the two PIE endings *-nto and *-ro (cf. Sanskrit -ran < *-ro-nt). The
3pl. passive regularly reflects PIE *-ntor.

The middle endings as reflected in the Old Irish imperfect lack the -r, and are
very similar to the Sanskrit middle preterit endings. The 2sg. form *-th2a:s,
both in Celtic and Sanskrit, may reflect a form with 3du. dative agreement
(*-th2ah2), secondarily extended with 2sg. active -s. If Old Irish 1pl. -mis
reflects something like *-mesdh(o)i, it is the third attested IE middle form
derived from this prototype, together with Hittite -wasta and Greek -mestha
(var. of -metha).


-or < *-o:r -mur < *-mor
-ris, -re < *-so -mini: < *-dhwomoi ?
-tur < *-tor -ntur < *-ntor

Latin has regularized the 1/2sg. forms (*-o:-r for *-h2ar, *-so for *-th2a).
The 2pl. -mini: is problematical. We know that *d(h)w gave *b in Latin, and
nasal assimilation may explain *-bini: > -mini:. But that only gets us to
something like *-dhwenoi, which is not a recognizable middle form. Perhaps we
have to posit dissimilation as well as assimilation, whereby we can start from a
form *-dhwom (as in Sankrit), extended with presentic *-oi, and then *-dhwomoi >
*-b&moi, with dissimilation *-bini:, with assimilation -mini:.


1. -mai < *-m-h2ai
2. -sai for *-th2ai
3. -toi, -tai < *-toi
1. -me(s)tha < *-mWe(s)dh + a
2. -sthe < *-(s-)dhwe
3. -ntai, -ntoi < *-ntoi
D. -sthon < *-(s-)h2th2am, *-(s-)h2tom

1. -me:n < *-m-h2ah2-m (?)
2. -so for *-th2a
3. -to < *-to
1. -me(s)tha < *-mwe(s)dh + a
2. -sthe < *-(s-)dhwe
3. -nto < *-nto
2. -sthon < *-(s-)h2th2am
3. -sthe:n < *-(s-)h2tah2m

The -s- before endings in a dental probably derives from roots ending in another
dental, where *-TT- > *-TsT- > -sT-. The a-vocalism of the 1/2 sg. has been
transferred to some of the other forms (3sg./pl. -tai, -ntai, 1pl. -metha). The
past tense 1sg. ending -me:n is perhaps best compared to the 2sg. past ending of
Sanskrit (and the Old Irish imperfect) -tha:s, from *-th2ah2 + -s, so similarly
1sg. *-h2ah2 + -m > m- + -a:m > -ma:m.


1. -e < *-h2ai
2. -se for *-th2ai
3. -te, -e < *-toi, *-oi
1. -mahe < *-mwodh + -oi
2. -dhve < *-dhwoi
3. -nte, -re, -rate < *-ntoi, *-roi, *-r-ntoi
1. -vahe < *-h2mwodh + oi
2. -ithe/-a:the < *-h2th2ai
3. -ite/-a:te < *-h2toi

1. -i (analogical: -mahe ~ -mahi --> -e ~ -i)
2. -tha:s < *-th2ah2 + -s
3. -ta, -a < *-to, *-o
1. -mahi < *-mWodhi
2. -dhvam < *-dhWom
3. -nta, -ran < *-nto, *-ro+nt
1. -vahi < *-h2mWodhi
2. -itha:m, -a:tha:m < *-h2th2ah2m
3. -ita:m, -a:ta:m < *-h2tah2m

Besides the indicative, we also have middle optatives and conjunctives. The
middle conjunctive/thematic is in some languages (Gothic, Tocharian)
characterized by o-grade of the thematic vowel in all persons. Since the
quality of the thematic vowel depends on the voicedness of the following
segment, this cannot be original. The o-grade is original to the whole plural
(*-o-mw-, *-o-dhw-, *-o-nt-/*-o-r-), but the singular originally had *e
throughout (*-e-h2-, *-e-th2-, *-e(-t-)), with *e > *a in the 1sg, of course.

As mentioned previously, Greek and Sanskrit have what were probably middle dual
forms in the active dual (Skt. 2du./3du. preterite -tam, -ta:m, Grk. 2du./3du.
preterite -ton, -te:n, Greek also 2/3 du. present -ton), from *-h2tWom, *-h2tom
(2/3du. x 1sg./pl.) and *-h2tah2m (3du x 1du), respectively, in the grid above.

Besides middle forms, Sanskrit also has a 3rd sg. aorist passive, with strong
root in o-grade, and ending -i (< *-H?). It's unclear what the origin of this
peculiar form is. Formally, it looks like a "verbal collective", whatever that
may mean.

Miguel Carrasquer Vidal