Verb placement and agreement
Olaf Koeneman
(Radboud University Nijmegen)
Overview
This course discusses two case studies related to the interaction between verb placement and agreement:
- The correlation between V to I movement and rich agreement (the so-called ‘Rich Agreement Hypothesis’). This is collaborative work with Hedde Zeijlstra.
- This is a case of inter-language variation: a language either moves or does not move the verb to I, depending on its subject-agreement properties.
- It will be argued that the correlation is real.
- It will be argued that this correlation holds in its strongest (bi-directional) formulation. If and only if a language has rich agreement does it move the verb to I.
- It is proposed that the correlation has the status of a potential language universal.
- The correlation between pre- or post-subject verb placement and having ‘inversion morphology’. This is collaborative work with Jan Don & Paula Fenger.
- This is a case of intra-language variation: the verb carries different morphology depending on its place with respect to the subject.
- It will be argued that to understand this phenomenon better one has to take into account inter-language (or better: inter-dialectal) variation as well.
- It will be argued that there are clear restrictions on inversion morphology.
- It will be argued that these restriction form an argument against one prominent account of inversion morphology, namely the ‘double paradigm theory’.
- It will be argued that post-syntactic operations (most notably ‘impoverishment’) help tremendously in accounting for the (restrictions on) the variation.
Case study 1
The Rich Agreement Hypothesis Rehabilitated
- Introduction
Language universals under fire: Evans & Levinson (2009):
“This target article summarizes decades of cross-linguistic work by typologists and descriptive linguists, showing just how few and unprofound the universal characteristics of language are, once we honestly confront the diversity offered to us by the world’s 6,000 to 8,000 languages.”(Evans & Levinson, 2009, p. 429)
See Lingua (2010, Volume 120 No 120) for extensive replies, criticizing various aspects of their analysis. One common thread: not the proposed language universals but perhaps the criticism on them is ‘unprofound’.
Structure of this first case study:
- Propose the generalization in a way that is explicit enough to be testable.
- Identify the counter-examples.
- Contra Evans & Levinson, do not stop here but study the counter-examples in a profound way.
- Only then conclude whether the counter-examples are real counter-examples.
An example isGreenberg’s universal 42 (1963: 96):
All languages have pronominal categories involving at least three persons and two numbers.
(1)Kuman (cf. Cysouw 2003)
Person / Singular / Plural1 / na / no
2 / ene
3 / ye
Counter-examples exist (Tvica 2014):
(2)Pronouns in Japanese(Martin 2004: 144,adapted)
Person / Singular / Plural1 / wata[ku]si
2 / anata
3 / ano-hitoM_bookmark70,an´o-koF
(3)Pronouns in Wāmbule (Opgenort 2002:197, adapted)
Person / Singular / Plural1 / uṅgu~uṅ
2 / unu∼un
3 / ɑ̄ṅgu~ɑ̄ṅ
But as Tvica notes:
- Whenever a language lacks forms for the plural, there are forms that are restricted to occurring in singular contexts only (Classical Chinese).
- Whenever a language lacks pronouns for a particular person (i) either a null form appears (so that a contrast is still observable) or (ii) the missing features are compensated for by morphology on the verb.
(4)Wāmbule
- uṅgu hepIbi-ljɑ̄ː-ø-me
1cooked.grainyoursoc-loceat-1.sg -res
‘I eat rice at your place.’
- unimbī-lcamdopɑ̄-sīcɑ̄b-du-m
2thatsoc-locgamedo-infcan-2.sg-res
‘You.sg can play with that [boy].’
(Opgenort 2002: 169)
So these counter-examples all disappear after a more profound look.
The Rich Agreement Hypothesis states that there is a correlation between V to I movement and richness of agreement.
Two formulations of the Rich Agreement Hypothesis (henceforth: RAH):
- Strong (bi-directional): Rich inflection V-to-I movement
- Weak (one-directional):Rich inflection V-to-I movement
Counterexamples to the strong definition (language with poor agreement but with V to I movement) would allow one to maintain the weak version. A counterexample to the weak version kills the whole hypothesis.
Structure of this case study:
(i)What is V to I movement? section 1.1
(ii)What is rich? section 1.2
(iii)What is the empirical motivation? section 2
(iv)What are the counterexamples? section 3
(v)Why are they not really counterexamples? section 3
(vi)Why would this correlation exist in the first place? section 4
1.1What is V to I movement
Some languages move the finite verb to I, others do not (Emonds, 1976).
(5)a.John oftenkisses MaryEnglish
b.Jean embrasse souvent Marie
Jean kisses often MarieFrench
- Jean asouvent embrassé Marie
Jean has often kissed Marie
(6)IP
SUI’
INegP/VP
NegVP
tSUV’
VOB
diagnostic
Note that this distinction holds:
-irrespective of the size of the inflectional domain;
-irrespective of the size of the verb phrase (distinction between vP and VP);
-irrespective of the precise analysis of negation (adverb or functional head);
-irrespective of whether the subject originates within the VP.
So why not assume that the verb is fixed and the position of the adverb/negation is flexible?
-Note that the generalization is not about verbs but about finite verbs ((5)b versus (5)c).
-As we will argue, there is a correlation with rich agreement.
If you assume that the position of the verb is fixed and the order of adverbs/negation is flexible, then adverb/negation placement depends on the finiteness and richness of another category (the verb) that it has to be ordered with. This requires a funky theory of adverbs/negation. If the position of the diagnostics is fixed and the verb can move, then movement is contingent on finiteness and richness of the verb itself, which makes more intuitive sense. Alterntively, you could assume that each language just picks a particular word order, but that’s the disinterested answer, as far as I am concerned.
1.2What is rich?
Ideally, the definition of what counts as rich is:
(i)empirically accurate. It should correctly predict when a language has V to I movement.
(ii)conceptually sound. It either allows us to relate richness in V to I movement to something else that requires a similar definition and/or it helps us in trying to understand why richness would be related to V to I movement at all.
(7)Hypothetical definition:
V to I movement takes place if in a language the affix for the 1st person plural is distinct from the affix for the 3rd person singular.
Even if (7) turns out to be empirically correct (and actually, it may very well be!), it would be a mystery why it should hold. There are in fact quite a few empirically correct definitions, so the conceptual requirement is in fact quite decisive.
Our proposal? Back to Greenberg and Kuman!
(8)Greenberg’s universal 42 (1963: 96)
All languages have pronominal categories involving at least three persons and two numbers.
(9)
Person / Singular / Plural1 / na / no
2 / ene
3 / ye
Na [+speaker][-plural]
Ene [-speaker][+participant]
Ye [-speaker][-participant]
No [+speaker][+plural]
(10)Richness (our definition):
A language exhibits rich subject agreement iff agreement involves at least the same featural distinctions as those manifested in the smallest (subject) pronoun inventories universally possible.
Given (8), every language requires at least three pronominal features to constitute subject pronouns. If languages exhibit agreement with at least these three features, agreement in such languages counts as rich and V to I movement is expected.
(11)a.sgplb.sgplc.sgpl
1stA1st1st
A
2nd2nd2ndBA
B
3rd3rdB3rd
A = [+speaker]A = [+participant] A = [+plural]
B = [-speaker] B = [-participant] B = [-plural]
(i)Emprically accurate? We will show.
(ii)Conceptually sound? Well, it uses a definition we independently need. And it suggests that movement of the verb is somehow related to the fact that rich agreement on the verb shares properties with pronominal subjects. So it moves us in a very specific direction (although the correlation is not instantaneously explained).
2.Empirical motivation
Below, we will show (i) that richness in some intuitive sense makes the right cut (Kosmeijer 1986, Pollock 1989, Platzack & Holmberg 1989, Holmberg & Platzack 1991, 1995, Roberts 1993, Rohrbacher 1994, Vikner 1995, 1997, Bobaljik 1995, Bobaljik & Thráinsson 1998) and (ii) that richness in our precise sense also makes the right cut. And it will make the right cut in the realms of synchronic macro-variation, synchronic micro-variation and diachronic variation respectively.
2.1 Synchronic macro-variation
- Standard French: rich V-movement across ‘often/neg’
Standard English: poor No V-movement across ‘often/neg’
(12)a. Englishb. Standard French
inf. walk-øinf. Parl-er
SG PL SG PL
1stwalk-øwalk-ø1stparl-eparl-ons
2ndwalk-øwalk-ø2ndparl-esparl-ez
3rd walk-swalk-ø3rdparl-eparl-ent
(13)a.John oftenKISSES MaryEnglish
b.Jean EMBRASSEe souvent Marie
Jean kisses often MarieFrench
- Icelandic and Yiddish: rich V-movement across ‘often/neg’
Norwegian and Danish: poor No V-movement across ‘often/neg’
(14)a. Icelandicb. Yiddish
inf. seg-ja inf. loyf-n
SG PL SG PL
1stseg-i seg-jum1stloyf-ø loyf-n
2ndseg-ir seg-ið2ndloyf-st loyf-t
3rd seg-ir seg-ja3rdloyf-t loyf-n
These languages show V to I movement in non-V2 contexts:
(15)a. Ég veit ekki af hverju kýrin HEFURoft staðið í herberginuIcelandic
I know not why the.cow has often stood in the room
b. Ikh veys nit ven di ku IZ oyfn geshtanen in tsimerYiddish
I know not when the cow is often stood in the room
(16)a. Danish b. Norwegian
inf. kast-e inf. elsk-a
SG PL SG PL
1st kast-er kast-er 1stelsk-er elsk-er
2ndkast-er kast-er 2ndelsk-er elsk-er
3rd kast-er kast-er 3rd elsk-er elsk-er
(17)a. at Peter ofte HAVDElæst den Danish
that Peter often had read it
b. Vi tenkte ikkeat han aldri VILLE ha penger Norwegian
we thought not that he never would have money
2.2 Synchronic microvariation
- Standard Swedish: poor No V-movement across ‘often/neg’
Älvdalen Swedish: rich V-movement across ‘often/neg’
(18)a.Standard Swedish b.Älvdalen Swedish
inf. bit-a inf. kast-a
SG PL SG PL
1st bit-er bit-er 1stkast-ar kast-um
2nd bit-er bit-er 2ndkast-ar kast-ir
3rdbit-er bit-er 3rdkast-ar kast-a
(19)a. att Johan inte köpteboken Standard Swedish
that Johan not bought book-the
b. ...ba fo dye at uir uildumint fy om Älvdalen Swedish
just because that we would not follow him
Hallingdalen Norwegian: poor No V-movement across ‘often/neg’
(20)a. Älvdalen Swedishb. Hallingdalen Norwegian
inf. kast-a inf. kast-æ
SG PL SG PL
1stkast-ar kast-um 1stkast-a kast-æ
2ndkast-ar kast-er 2ndkast-a kast-æ
3rdkast-ar kast-a 3rdkast-a kast-æ
(21)Noko gamlæ mænna som ikki haddævore mæ ve kyrkjaHallingdalen
some old men who not had been along at church
2.3 Diachronic microvariation
Old Swedish, Middle English: rich V-movement across ‘often/neg’.
Early Mod. Swedish, Early Mod. English: poor No V-movement across ‘often/neg’
(22)a. Old Swedish b. Middle English
inf. älsk-a inf. sing-en
SG PL SG PL
1stälsk-ar älsk-um 1stsing-e sing-en
2ndälsk-ar älsk-in 2ndsing-est sing-en
3rdälsk-ar älsk-a 3rdsing-eð sing-en
(23)æn han sivngærægh thigianda messu Old Swedish, 1290
if he sings not silent mass
(24)By thy thanks I SETnot a strawMiddle English
The same point can be made for older stages of Norwegian and Danish. Germanic VO languages used to be rich at some point but a subset of them lost rich agreement as well as V to I movement along the way.
3.Trouble in paradise?
Over the years, a substantial number of problems have been conjured up for the RAH.
Empirical:
-The existence of languages that seem to exhibit V-to-I movement in absence of rich agreement. These examples then seem to falsify the strong version of the RAH;
-The existence of languages that seem to lack V-to-I movement in spite of rich agreement. These then seem to falsify the weak version of the RAH;
-The fact that loss of V-to-I movement does not directly follow verbal deflection. These diachronic gaps would show that there is no contentful correlation between richness and V to I movement.
We will argue that all these data have been misanalyzed. Result: the RAH should be maintained in its strong form.
Theoretical:
-The idea that in syntactic derivations RAH requires access to paradigmatic knowledge;
-The idea that late morphological spell-out blocks a close connection between morphology and syntactic operations (morphology does not drive syntax);
We will argue that these arguments are basically correct but do nothing to refute the RAH. It only shows that the link between morphology and syntactic movement is indirect. The missing piece of the puzzle is language acquisition.
3.1 Evidence against the weak RAH
Note that if this evidence is convincing we are left with no RAH: If the weak RAH is falsified, then there is no RAH.
Icelandic (Bentzen et al 2007) and Älvdalen Swedish (Garbacz 2010) Rich agreement but “V3” orders allowed
(25)Mér fannst skrýtið þegar hann oftlékhróknumIcelandic
I found strange when he often moved rook.the
‘I thought it was strange when he often moved the rook’
(26)a.Eð ir biln so an intWILL åvå.Älvdalen Swedish
it is car.def. that he not wants-to have
b.Eð ir biln so an WILLint åvå.
it is car.def. that he wants-to not have
In Russian the finite verb (both in the perfective and the imperfective) follows manner or frequency adverbs:
(27)a.My vnimatel'noPROČITALIpravilaRussian
we carefully PERF.read rules
‘We have carefully read the rules'
b.My častoČITALIpravila
We often IMP.read rules
'We read the rules often
Let’s start with Icelandic and Älvdalen Swedish.
- Note that these examples do not show that the verb remains in situ. They only show that the verb is preceded by an adverb/negation.
- Another logical possibility is that the effect of movement is neutralized by either subsequent movement or higher base-generation of the verb movement diagnostic (See for Icelandic Angantýsson (2007) and Thráinsson (2009)).
(28)a.[CP [FP SU ADVjVi[vPtjti…]]]
b.[CP [FP SU (NEG)Vi(NEG) [vP …ti …]]]
Why is the lack of V to I in these constructions an implausible analysis for Icelandic?
(i)Motivation for the “V3” word order:
- It is ‘severely restricted and heavily marked’.
- The adverb must be stressed.
- The subject must be an unstressed pronoun.
- Optional movement generally yields information-structural effects, and stress effects on the adverb are thus expected. The fact that the subject must be unstressed and not introduce a new referent, then, follows from the adverb already being in focus.
- Optional movement standardly leads to (information-structural) marked sentences (fronting takes place to mark a constituent for focus- or topichood). If Bentzen et al. were right, the reverse would apply: not moving gives you the marked option, which is typologically extremely odd.
(ii) “V3” orders can co-occur with object shift. Example (23) (from Angantýsson, p.c.) shows that verbs that follow adverbs can still precede shifted objects. Under the standard assumption that the object shifts out of vP (in the example indicated by the fact that it precedes the negation), the latter must have subsequently moved across the object, countering Bentzen et al’s account. (cf. Thráinsson (2009) for further argumentation along these lines.).
(29)Mér fannst skrýtið þegar hann oft lék hróknum ekkií tímahraki
I found strange when he often moved rook.the not in lack.of.time
‘I thought it was strange when he often didn’t move the rook through lack of time’
Why is the lack of V to I in these constructions an implausible analysis for Älvdalen Swedish? Answer: Because there is independent evidence showing that negation can appear even higher.
(30)Eð ir biln so int an will åvå.
it is car.def. that not he wants-to have
The fact that the finite verb in Icelandic and Älvdalen Swedish is able to appear to the left of the marker of sentential negation at all (in contrast to English and standard versions of Mainland Scandinavian) is what provides the evidence for V-to-I movement.
Conclusion:
Icelandic and Älvdalen Swedish do provide evidence for V-to-I movement. Hence, at least the weak RAH is maintained.
Let’s turn to Russian. Again, there are two possible analyses. Either the adverbs are good diagnostics and Russian lacks V to I movement, contra RAH (as in (31)a) or adverbs are generated higher as a rule (as in (31)b) and Russian neither confirms nor falsifies the RAH.
(31)a.[IP My [vP vnimatel'no pročitali pravila ]]
b.[FP My [IP vnimatel'no [IP pročitali [vP tVfinite pravila ]]]]
Evidence for high placement of adverbs in Russian, as in (31)b, comes from negative sentences. Negation in Russian is a head and moves along to C with the verb in negative imperatives and negative questions (cf. Han 2001 and references therein):
(32)a.Ne pey vodku často!
Neg drink vodaka often
‘Do not often drink wodka’
b.Ne p’eš li ty vodku často?
Neg drink Q you wodka often?
‘Don’t you often drink wodka’
Now, suppose that adverbs could adjoin to vP, thus functioning as a diagnostic for V-to-I movement. Then the predicted order of a negative clause containing an adverb ‘often’ should be as in (33). But these are bad. Such sentences, however, are seriously degraded if not just ungrammatical, as shown both for perfective and imperfective verbs.
(33)a.??/*Ty ne pročitalavnimatel'no pravila
You NEG PERF.read carefully rules
'You haven't read the rules carefully'
b.??/*Ty ne čitalačasto pravila
You NEG read often rules
'You haven't read the rules carefully'
The only possible way to include an adverb in a sentence where the negative marker left-attaches to the finite verb is to let it precede the negated verb or to put it clause-finally.
(34)a. Vnimatel'noty ne pročitala pravila
Carefullyyou NEG PERF.read rules
'You haven't carefully read the rules'
b.Často ty ne čitalapravila
Often you NEG IMP.read rules
'You haven't often read the rules'
This shows that vP is not a proper left-adjunction site for these adverbs.
Conclusion:
Russian neither confirms nor falsifies the RAH.
3.2 Evidence against the strong RAH
Spoken French, Faroese and Regional Northern Norwegian (ReNN) dialects Poor, agreement but Vfinite – NEG orders allowed.
(35)a. Spoken French (phonological)b. Faroese I & II
inf. parl-[e]inf. kast-a
SG PL SG PL
1stparl-əparl-ə * 1stkast-ikast-a
2nd parl-ə parl-[e] 2ndkast-ir kast-a
3rd parl-əparl-ə3rd kast-irkast-a
*1st PL in spoken French is not (nous) parl-[õ] but (on) parl-ə, making French observationally as poor as Modern English.
(36)Jean MANGE souvent des pommes.French
Jean eats often apples
Jonas (1995): There is a dialect split within Faroese. Faroese I and II have the same paradigm but a different verbal syntax. Faroese I has optional verb movement (cf. 27a), which Faroese II lacks (cf. 27b):
(37)a.Taþ var ovæntaþ at dreingirnir <VORU> als ikki <VORU> ósamdir
it was unexpected that boys-the were at-all not disagreed
b. Taþ var ovæntaþ at dreingirnir <*VORU> als ikki <VORU> ósamdir
it was unexpected that boys-the at-all not were disagreed
Regional Northern Norwegian dialects have the same paradigm as Standard Norwegian (i.e. no person/number distinctions whatsoever. Yet, Vfinite can precede sentential adverbs.
(38)…ettersom nån studenta {sannsynligvis} leverte {sannsynligvis} oppgaven
… as some students probably handed.in probably assignment.the
‘… as some students probably handed in the assignment’
Solution for French:
Its pervasive subject doubling property indicates that subject clitics function as agreement markers, rendering the language’s inventory of the formal pronominal features rich (see Auger 1994, Legendre et al 2004 and others).