Agreement Suppression Effects and Unification via Agree

        In this paper I argue that despite being an apparently disparate range of empirical facts
and having received deferent analytical treatments, Subject-verb agreement (1), Clitic-doubling
(2), and Negative Concord (3) are all cases of agreement. I show that just as Subject extraction
alters agreement and yields Anti-Agreement Effects (AAE) (1) vs. (4) (Ouhalla 1993, 2005b,
Richards 2001), Object extraction bars the occurrence of clitics i.e. bars clitic-doubling (2) vs.
(5), and Negative Concord Item (NCI) extraction affects Negative Concord (NC) (3) vs. (6).
These effects are just as expected if Clitic Doubling and NC are themselves forms of agreement
obtained via the same mechanism namely: Agree.

        Subject-Verb agreement is obtained when T "probes" a DP subject. As a result of the
Agree operation, the unvalued and un-interpretable ? -features on T and the unvalued un-
interpretable Case feature on the DP are valued and spelled out before reaching the PF and LF
interfaces. Extending the same analysis to Clitic-Doubling I argue that the clitic head enters in an
Agreement relation with its DP double. Clitic-Doubling, as is cross-linguistically attested,
requires feature agreement/matching between the clitic and the doubled DP (7) vs. (8). In fact a
few researchers have analyzed clitics as heads of Agreement projections or akin to agreement
heads heading clitic projections (Sportiche 1992, 1998). The position the clitic is merged into c-
commands the clitic double and appealing to probe-goal Agree we can account for the agreement
between the clitic and its double without appealing to relations like Spec-Head.

        Negative Concord as a type of agreement is not in fact a novel idea. The idea that NCI's
and NPI's are licensed by a negation element goes back to Klima (1964). Also, the hypothesis
that the structural relation under which Negation licenses NCI's and NPI's is C-Command goes
back at least to Laka (1990, 1993). Adopting the same basic idea, I argue that the licensing
relation between the negation and the NCI's is a matching relation obtained by the application of
the Agree operation as defined in Chomsky (2001a, 2004) (see also Zeijlstra 2004, Watanabe
2002, 2004).

        The probe in this case is a Neg head and the goal is an NCI. Following Watanabe (2004),
NCI's are cross-linguistically specified for an un-interpretable focus feature which renders them
active and visible to the probe. In Berber for example, Neg head (Neg1) is lexically specified for
Neg feature, carries Neg force and can not be dropped, whereas the negative concord sha (Neg2)
is optional. I hypothesize that the NCI sha is specified for an unvalued un-interpretable Neg
feature which gets valued and deleted as a result of the Agree operation between the Neg head
and the NCI.

        The negation head also licenses NPI's like agidge `no one'; the subject agidge in (11)
enters into multiple Agreement relations as represented in (12), first with the f -complete T
which values and deletes its case feature. However the subject's Neg feature does not get valued
by T hence the subject remains active and enters into a second Agreement relation, with Neg1,
which values its Neg feature. Extraction of the NPI subject not only inhibits subject-verb
agreement and yields AAE but it also affects NC (13) and (14). What is interesting is that NC is
not only prohibited when an NPI subject is extracted but also when any type of subject is
extracted (Wh-Subject (15), Subject-relative (16), and Clefted-Subject (17)). As mentioned
before Clitic doubling is prohibited when the object DP is extracted (5), and as predicted NC is
also inhibited in these contexts (18) vs. (19).

        These effects are not coincidental and provide, under proper analysis, further empirical
evidence of unification under Agree (and elimination of construction specific accounts).
Following Chomsky (2005), the "third factor" in "language design" is a set of general principles
of efficient computation. Agree, to the extent that it is a computationally efficient operation,
holds for any form of agreement. When this operation is inhibited, the suppression effects are
uniformly displayed across all the seemingly disparate domains within which it applies.

(1)    th.qra         thamttut       thabrat                  (data is from Tamazight Berber)
       3sf.read.PERF woman           letter
       `The woman read the letter'
(2)    th.wshith-as           lkthaab i-Merriam
       give.PERF.2s-her       book to- Merriam
       `You gave the book to Merriam'
(3)    ur      iddi                  sha     ag-idge
       Neg1 leave.PERF.neg.3sm Neg2 not-even-one
       `No one read left'
(4)    ma-ag          qra.n          ___              thabrat ?       Anti-Agreement Effect
       who-that       read.Neu       ___              letter
       `Who read the letter?'
(5)    i-mi th.wshith-(*as)          lkthaab          _____ ?         "Anti-Clitic Doubling Effect"
       to-who give.PERF.2s-her book                   _____
       `To whom did you give the book?'
(6)    ag-idge        ur      iddin         (*sha)            ____           "Anti-NC Effect"
       not-even-one Neg1 leave.PERF (*Neg2)                   ____
       `No one left'
(7)    ughen -as              tumubil        i-yelli-sn
       buy.PERF.3p-her         car           to-daughter-their
       `They bought a car to their daughter'
(8)    *ughen-ak              tumubil        i-yelli-sn
       buy.PERF.3p-you        car            to-daughter-their
(9)    ur      thugh          sha    Meriam           lekthaab
       Neg1 3f.give.PERF Neg2 Meriam                  lekthaab
       `Meriam didn't buy the book'
(10)   Sha-ur thugh           ____ Meriam             lekthaab
       Neg2-Neg1              ____ Meriam             lekthaab
       `Meriam didn't buy the book'
(11)   ur      iddi                  agidge           gher lhefla
       Neg1 go.PERF.neg.3s           no one           to       party
               `no one went to the party '
(12)   [NegP ur [TP T... iddi [vP agidge iddi gher lhefla ] ] ] ]
(13)   agidge         ur      iddi.n         ____ gher lhefla
         no one       Neg1 go.PERF.Neu ____ to-party
       `no one went to the party'
(14)   *agidge ur iddi               sha     ____ gher lhefla
         no one Neg1 go.PERF.3s Neg2 ____ to-party
       `no one went to the party'
(15)   ma      ur     iddi.n         (*sha)           gher-lhefla
       who Neg1 went-Neu             (Neg2)           to-party
       `who didn't go to the party?'
(16)   argaza ay-     ur      iddi.n (*sha)           gher-lhefla
       man-this that- Neg1 went-Neu (*Neg2) to-party
       `it was this man who didn't go to the party'
(17)   argaz ay-      ur      iddi.n         (*sha)           gher-lhefla...
       man that- Neg1 went.Neu               (*Neg2)          to-party...
       `The man who didn't go to the party...'
(18)   ur-     as     th.wshith-     sha     lkthaab i-Merriam
       Neg1- her      give.PERF.2s Neg2 book to- Merriam
       `You didn't give the book to Merriam'
(19)   i-mi ur-       (*as) th.wshith        (*sha) lkthaab ____
       to-who Neg- (*her) 2s.gave            (*Neg2) book ____
       `Who didn't you give the book to?'