Maximally Distributing Morphology

advertisement
J. Emonds, May 2009, Topic 3
Structure of the Syntax-Lexicon Interface
INFLECTIONAL MERGER = PF MERGE OF BOUND MORPHEMES
University of Newcastle, Structure of the Syntax-Lexicon Interface
I. DM Merger: is it Move or Merge or neither?
These lectures have been focusing on syntax within “words,” defined simply as maximal X0
domains (Embick and Noyer 2001: 574). Words encompass compounds and bound morphology.
The only universal difference between words and phrases seems to be (1), an idea proposed by
several linguists, given as (19) in Topic 1. It holds of both compounds and bound morphology.
(1)
Nonmaximality. A productive X0 domain cannot contain categories characteristic of
maximal projections, such as D, I, Degree Words, Intensifiers, Focus Particles, etc.
We have been considering whether some separate component of “Morphological Structure” (MS)
contains not only (1) but also other special principles that apply only within words, i.e. X0.
One such principle is proposed is an early paper in Distributed Morphology. Halle and Marantz
(1993: 116) derive Chomsky’s (1957) classic English “affix movement” by an MS operation
“Merger,” which “generally joins a head with the head of its complement XP.”
H and M’s Merger accounts for how words sometimes spell out syntactic characteristics of
constituents structurally “nearby,” what traditional grammar calls inflection, in ways that seem
unlike transformational relations. English tense/number suffixes are one clear example.
As welcome consequences, this step eliminates (i) an instance of transformational “lowering” (H
and M 1993: 134) as well as (ii) the rule’s ad hoc restriction to –MODAL values of English I.
Just below, we examine affix movement in detail, agreeing that Merger is the right approach.
But if Merger is not itself an ad hoc way to eliminate lowering, there must be other instances of
it, and in addition, some general characterization of all possible Mergers, stated here as (2).
(2)
Merger/ Alternative Realization. A syntactic feature F canonically interpretable on β
can be realized in a closed class item under γ0, provided projections of β and γ are sisters.
Since Emonds (1985), I have used the term Alternative Realization (“AR”) for such operations,
where Marantz has used Merger. A difference between the two conceptions will emerge later.
II. English finite suffixes: Merger as PF Merge of “Dissociation”
Chomsky (1957) provides a convincing generative treatment of English finite affixes. Let ±φ
represent features of person/ number agreement with a subject. His proposal in today’s terms:
(3)
Affix Movement. The features of I, [-MODAL, +PAST, ±φ], are moved to, and spelled
out under, a c-commanded head V if and only if these I and VP are adjacent sisters.
But in retrospect, there is a problem: almost all convincing transformational analyses (i) have
involved raising of constituents αj and (ii) affect all instances of such αj.
1
J. Emonds, May 2009, Topic 3
Structure of the Syntax-Lexicon Interface
Since Affix Movement (3) contradicts both these general properties, it doesn’t seem to be a
transformation. But (3) is a special case of Merger/ AR (2), where F is ±PAST, β is I, and γ is V.
As a result of Merger/ AR, a single feature, e.g., PAST, splits as it were into two realizations,
one in I where it can be universally interpreted in LF (its “canonical position”), and another on
V where it is “spelled out.” Plausibly, only canonical positions contribute to interpretation.
The physically observed Merged/ AR position is then only a PF phenomenon. In other words,
Merger/ AR results from vocabulary items, in particular many affixes, entering trees only in PF.
This derives Chomsky’s stipulation that Affix Movement must “apply late” in a derivation.
From this perspective, PF Insertion of α is a subtype of Merge α. Such “Late Merge” occurs
(only) when α’s position does not contribute to LF and has certain other characteristic properties.
What I call Alternative Realization is exactly what Embick and Noyer (2001) call “Dissociation”:
These morphemes [Case suffixes on N & A, JE] must be added postsyntactically…such
inserted morphemes are called dissociated, since the information their signalization
conveys is partly separated from the original locus of that information in the phrase
marker [my “canonical positions,” JE]….Dissociated morphemes are not interpreted in
LF, since they are inserted only at Spell Out. (Embick and Noyer 558)
The limitations imposed by Merger/ AR on where non-interpretable features occur imply that
Affix Movement (3) has no status as a movement. Rather, English lexical entries specify only
output values of Merger/ AR, which in this case are two (or three) suffixes on a lower head V:
(4)
English lexical representations:
a. -(e)d, <V___>, +PAST
b. -(e)s, <V___>, -PAST, -PLUR, +3rd
c. Ø, <V___>, -PAST, { +PLUR/ -3rd }, or possibly:
Ø, <V___>, “elsewhere”
The affix movement statements (4) are thus not transformational. They are rather contextual
specifications inside lexical items. The Merger/AR Principle (2) defines a space of possible
lexical entries, limiting configurations that can appear them.
No analysis of English finite forms, no matter what terms it uses (feature checking, etc.) can
avoid the minimal descriptive statements of (4). Without them, an analysis is just an abstraction.
III. Extended Sisterhood
The present tense inflection (4b) requires a refinement in AR. A suffix –(e)s on V spells out
features of a subject D, yet no projections of V and D appear to be sisters—at least I0 intervenes.
In fact, AR correctly predicts that any overt I, such as a modal, blocks Affix Movement.
To capture these facts, the phrase “are sisters” in Merger/ AR (2) must be interpreted exactly as
in Chomsky’s early technical use of “is an α.” That is:
(5)
“Extended” Sisterhood. γ “is a” sister of β if all lexical material dominated by an actual
sister δ of β is in fact also dominated by γ.
2
J. Emonds, May 2009, Topic 3
Structure of the Syntax-Lexicon Interface
α
(6)
β
δ
ε
Ø
ζ
η
γ
Ø
lexical material
This interpretation of “γ is a sister of β” permits empty heads between β and γ in Merger/ AR
configurations. The bound morpheme –(e)s in (4b) alternatively realizes the features [-PLUR,
+3rd ] of a subject D, because VP “is a” sister of DP whenever an intervening I is empty.
In (4), these PF inserted finite suffixes are (a) bound morphemes, and (b) enter a derivation via
Merger/ AR. The main question now is, how widely does Merger/ AR (2) have effects?
The answer here is that Merger/ AR covers all of what traditional grammar calls inflection.
IV. Applicative morphology as Merger/ Alternative Realization
In “applicative” constructions from various language families, some DPs that typically appear in
PP complements, such as a benefactive, instrumental or “inner locatives,” can alternatively be
“promoted” to direct object status. Baker (1988: Ch. 3) provides several detailed analyses.
An essential property of applicatives is that language-particular prepositions [P,+Fi] (introducing
the DP complements that can “promote”) are replaced by suffixes on a selecting higher head V.
(7)
a.
b.
(8)
a.
b.
Ali [V mem-beli ] telefisi [P untuk ] ibu-nja. (Indonesian; Chung 1976)
Ali TRN-buy television for mother-his
‘Ali bought a television for his mother.’
Ali [V mem-beli-[σ kan ] ] ibu-nja telefisi.
Ali TRN-buy-APPL mother-his television
‘Ali bought his mother a television’
Mbidzi [V zi-na-perek-a ] msampha [P kwa ] nkhandwe. (Chichewa; Baker 1988)
Zebras SP-PAST-hand-ASP trap
to
fox
‘The zebras handed the trap to the fox’
Mbidzi [V zi-na-perek-[σ er ]-a ]
nkhandwe msampha.
zebras SP-PAST-hand-APPL-ASP
fox
trap
‘The zebras handed the fox the trap’
Clearly, the PPs projected by these P are sisters of V0. AR accurately describes this
configuration, where in (2) β = P and γ = V. Merger/ AR thus sometimes realizes a syntactic
feature F lower than in its canonical position (Affix Movement) and sometimes higher
(Applicatives). 1 Exercise. Can English double object constructions be analyzed this way? How?
1
An exact analysis of applicative constructions is not at issue here. My view is that an empty P is a
necessary and sufficient condition for promoting (moving) the prepositional object DP to direct object. It
is then this promotion, not the location of the empty P, which causes the essentially universal reordering of applied and deep direct objects (Emonds and Ostler 2006).
3
J. Emonds, May 2009, Topic 3
Structure of the Syntax-Lexicon Interface
This type of AR, the insertion of applicative suffixes, then permits and in fact requires the Ps in
(7b)-(8b) to be zero. Similarly, affix movement requires English I to be zero.
What causes the canonical (interpreted) position of a feature F to be always or sometimes empty
when F is alternatively realized? The motivation for AR can be related to Economy, i.e., AR
permits derivations with fewer morphemes—or in phonological terms, fewer potential stresses.
Consequently, canonical positions are either (a) always or (b) whenever possible empty in the
presence of AR. Emonds (2000: section 4.4) discusses these two options with several examples.
V. A disagreement about English adjectival comparison
Since AR is a condition on the possible feature make up of vocabulary items in the lexicon, we
expect that AR features can sometimes depend on phonological contexts. This appears to be so.
As is well known, the English degree words more/ most can merge (= be alternatively realized)
and surface as the bound suffixes –er/ -est on the head of an AP.
Now the insertion contexts A___ for the entries for -er and -est also satisfy a phonological
condition: The category A here can consist at most of a single trochaic foot: sillier, earlier,
stupider, *contenter, *intensest, *wonderfuler, *explicitest.
The condition is necessary, not sufficient: *deafer, *tautest, *morbider, *aridest.
The syntactic aspects of this “AR of Comparison” closely parallel Affix Movement (3). In both
cases, the least marked members of c-commanding functional categories (DEG and I) surface
as bound suffixes on the lexical heads A and I of their respective sister phrases.
But Embick and Noyer (2001: 564)) deny full syntactic parallelism between Affix Movement
and the AR of Comparison. They attribute Comparison to a special Morphological Structure
operation, “Local Dislocation,” which is not an instance of Merger/ AR ( = their “Lowering”).
Their argument rests on one example: *Mary is the amazingly smartest person. They conclude
that placement of –en/ –est does not involve a structural A head. But the syntactic input (and
semantics) of their example is actually (9), which explains why –est cannot lower to smart.
(9)
Mary is the [ [most amazingly] smart ] person.
In other inputs (10a) whose degree words clearly modify and are higher than the head A,
lowering is possible (10b).
(10)
a. The [ most [ intellectually free ] ] couple on the planet just walked in.
Of the two, Mary has a [ more [ truly wide ] ] vision of our future.
b. The [ Ø [ intellectually freest ] ] couple on the planet just walked in.
Of the two, Mary has a [ Ø [ truly wider ] ] vision of our future.
Consequently, there is here no need for E and N’s special operation of Local Dislocation.
4
J. Emonds, May 2009, Topic 3
Structure of the Syntax-Lexicon Interface
The structural parallelism between adjectival comparison and verbal tense affixes is thus entirely
captured by AR. This further argues against affix movement as a transformation, since a parallel
transformation in comparative APs would then be subject to a phonological condition. 2
Merger/ AR (2) thus subsumes the morphological operations termed Merger by Halle and
Marantz (1993: section 4), and several more as well. Clearly, Merger/ AR is a central tool for
characterizing bound inflectional morphemes, that is, “morphology,” as suggested in Section I.
VI. Examples of Merger/ AR inside nominal projections
Embick and Noyer (2001: 561) divide the “Morphological Merger” of Marantz (1984: 261),
into two types: Local Dislocation and Lowering, where only Lowering is a subcase of AR (2).
(11)
Lowering of X0 to Y0. [XP X0 …[YP...Y0…]]  [XP…[YP…[Y0 Y0 + X0 ]...]]
They argue in detail that English Affix Movement exemplifies Lowering, and I concur.
Their section 5 goes on to demonstrate in detail that the Bulgarian nominal suffix –ta/ -ə ‘DEF’
is another example of Lowering, i.e., AR. That is, a feature of D is realized on/ merged with N.
In the same section, Embick and Noyer further argue that affixation and cliticization are at most
slightly different variants of Lowering (AR); again I agree.
Their section 7.1 discusses a Scandinavian definiteness suffix on nouns –en.
 In Danish, the suffix is in complementary distribution with definite determiners D
(Embick and Noyer, 584).
 In Swedish, -en doubles all definite determiners, zeroing only definite articles (580-581).
Both case exemplify the Lowering subcase of Merger/ AR (2), with X0 = D and Y0 = N.
Again, these inflections exemplify AR of a D feature on N. The Danish use of AR here is what I
now define as unmarked, and the Swedish use is marked.
(12)
Unmarked AR. Unmarked alternatively realized features are phonetically realized only
when their canonical source is null in PF. (English ±PAST marking on V).
(13)
Marked AR. Marked alternatively realized feature are phonetically realized whenever
they can be, doubling their canonical source. (English ±PLUR, ±3rd marking on V).
In other work, I underlining for marked AR features in lexical entries such as (4b). Upon
recognizing non-canonical spell outs of features, language learners assume (12). If a
construction then exhibits any doubling, they re-analyze unmarked AR as marked AR (13).
Embick and Noyer (section 6.2.1) treat Latin –que ‘and’ on second heads of conjoined
NPs as “Local Disloaction,” i.e. different than Lowering of Bulgarian definiteness.
E and N admit that they cannot limit phonology-based insertion conditions to their “Local
Dislocations,” since they later observe (573): “…elements whose distribution or provenance must be
essentially syntactic [i.e. due to their Lowering, their term for Merger, JE] can occasionally interact with
“lexical” phonological processes. … Additional cases of this type are presented in Hayes 1990.”
2
5
J. Emonds, May 2009, Topic 3
Structure of the Syntax-Lexicon Interface
Yet both their “Local Dislocation” of Latin –que and their “Lowering” of Bulgarian –ta
into NPs alternatively realize higher heads on noun-modifying adjectives in the same
way. So I see no reason to introduce a Local Dislocation that differs from Merger/ AR.
Besides spelling out definiteness, N inflections can also spell out the features of case and
number from higher heads within a DP, as many Germanic and Slavic inflectional patterns show.
At the same time lexical gender, convincingly argued in Ritter (1993) to be a syntactic feature
of N, can also appear on D and sometimes even on low numerals (Veselovská 2001).
Additionally, all these features can in varying ways appear on attributive adjectives within the
extended DP projection. All these standard cases of traditional inflection fall under AR.
“Free Riders” in the AR framework. The complex Germanic and Slavic systems of agreement
within DPs suggest that AR (2) not only can license canonical features of β on heads of β’s
sisters, but also may further "spread" features already alternatively realized.
Thus in the Czech (14), the underlined case features on D and N and PLUR feature on N are
already alternative rather than canonical realizations. These features are additionally overtly
marked on intervening adjectives, under the structural condition (extended sisterhood) set by AR.
(14)
...s(e) [D,PL,INST všemi ] [ D,PL,INST těmi ] [A,PL,INST krásnými ] [N, PL,INST děvčaty]
with
all
those
beautiful
girls
As formulated, AR (2) only displaces features in their canonical positions to the next lexical
head, upward or downward in a tree. And overall, we do not want to weaken AR by allowing it
to "apply to its own output" in order to derive paradigms typified by (14).
Rather, certain alternatively realized features which co-occur with canonical features should
"piggy-back" on the latter's alternative realizations, similar to the way in which formal features,
called free riders, move with interpreted ones in Chomsky (1995: 270).
(15)
Free Riders. If a morpheme μ alternatively realizes (under X0) a canonical feature F of a
category β, and some F' is also spelled out under β, then μ can also spell out F' on X0.
Thus (14), an attributive adjective inflection μ alternatively realizes the lexical gender feature of
an N, and CASE and NUMBER are also spelled out under N (as alternative realizations).
Consequently the adjectival inflection μ can also spell out CASE and NUMBER.
VII. Are Fusion, Fission and Local Dislocation special operations in MS?
Distributed Morphology generally agues against those specifically morphological principles
inherited from lexicalist approaches to word-internal structure. But as is somehow endemic in
studies of morphology, DM has undertaken a quest for other such principles of its own making.
Three of these presented as specific to MS are Fusion, Fission and Local Dislocation. Fusion is
a technical term for familiar porte-manteaux morphemes of structuralism.
6
J. Emonds, May 2009, Topic 3
Structure of the Syntax-Lexicon Interface
(i) Fusion. In discussing their “Morphological Structure,” Halle and Marantz (1993: 116)
distinguish between operations of Merger and Fusion. Slightly reordering their phrasing:
“Merger…maintains two independent terminal nodes under…a zero-category
node…On the other hand, fusion takes two terminal nodes…and fuses them into a
single terminal node….A simple example of fusion is the single affix signaling number
and case encountered in many Indo-European languages….”
In addition to Indo-European case morphemes, some other examples are (a) French fusing of
two unmarked Ps with definite articles (à ‘to’ + les ‘the’  aux), and (b) English finite copulas:
were = [I, -MODAL, +PAST, +PLUR], where +PLUR is an AR φ-feature of a subject DP.
Exercise: Write some lexical entries for some morphemes exhibiting Fusion.
Fusion is simply a special case of AR (2). It need not be counterposed to it, or to H and M’s
Merger. Under AR, fusions are just lexical items that are specified for multiple syntactic features
whose canonical positions happen to be on adjacent heads. English –(e)s (4b) is an example.
This view simplifies H and M’s conception. While in both views a “fused” lexical item is
specified to spell out a certain set of combined features, H and M additionally postulate Fusion
as a separate preliminary operation in Morphological Structure.
But a system incorporating AR as defined in (2) renders a separate MS operation of Fusion
redundant. AR subsumes and accurately limits the variety of syntactic configurations that
individual morphemes can express, using word-internal context features such as <Y___>.
(ii) Fission. Halle and Marantz (1993: 116-120) coin this term for structures where features
from a single category are realized in different morphemes under a single X0. As an example,
they show cases where first person is realized as both a prefix and a suffix on a Georgian verb.
This term calls to mind traditional circumfixes, such as the discontinuous German past/ passive
participle ge-...-t/ ge-...-en. The suffix [A –t/-en ] is indeed independent of the prefix, since it can
also occur alone. It can be analyzed like its English counterpart (Emonds 2000: Ch. 5).
The prefix ge- is then a separate vocabulary item that alternatively realizes [ A –t/–en ] (provided
other prefixes are absent). The locality restriction of sisterhood built into AR (2) automatically
permits “fission” of a bound morpheme’s features, here an A0, into two spell outs under one X0.
(iii) Local Dislocation. Embick and Noyer (2001) claim that relations targeting “subwords,”
that is, non-maximal X0, require a kind of movement that is specific to morphology. Section V
has shown that their brief foray into English adjectival comparison (9) is incorrect.
Less superficially, they use Local Dislocation to analyze the Lithuanian reflexive –si. This
puzzling morpheme is always bound on its left either by an unprefixed V or by the leftmost
prefix within V.
I don’t know enough about Lithuanian to be more than suggestive, but in fact their analysis is
also quite sketchy, relying e.g. on a novel device (and formal symbol) for “subword adjunction.”
7
J. Emonds, May 2009, Topic 3
Structure of the Syntax-Lexicon Interface
A more profitable approach to problematic constructions is to hypothesize lexical specifications
that combine well-tested familiar devices in unfamiliar ways. My proposed lexical entry:
(16)
Lithuanian reflexive. –si, REFL, <X0___> and <V>
The first context feature specifies that -si always has a single word-internal X0 sister on its left,
transparently true. It appears this X0 must not be interpreted as a sequence. Perhaps X0 = [-N].
The second context feature can be plausibly interpreted as follows: -si must be a sister of some
V-projection that heads a full VP. If this projection were some exterior phrase V1, this would
violate <X0___> by (1). So –si must be both a sister of V0 and under some Y0, as appears correct.
Exercise: What if anything in (9) of Topic 2 needs to be modified for this?
A position inside Y0 is not canonical for a pronominal feature like REFL, so (16) must be AR.
In light of this type of approach to Lithuanian -si, Local Dislocation as a device specific to only
Morphological Structure seems unnecessary. Yet –si is E and N’s only real argument for LD.
This leaves only their Lowering, which is as observed earlier a special case of Merger/ AR.
References for Topics 3
Baker, Mark. 1988. Incorporation: A Theory of Grammatical Function Changing, University of
Chicago Press, Chicago.
Chomsky, Noam. 1957. Syntactic Structures, Mouton, The Hague.
Chomsky, Noam. 1995. The Minimalist Program. Cambridge: MIT Press.
Chung, Sandra. 1976 An object-creating rule in Bahasa Indonesa. Linguistic Inquiry 7, 41–87.
Embick, David and Ralph Noyer. 2001. “Movement Operations after Syntax,” Linguistic Inquiry
32.
Emonds, Joseph. 2000. Lexicon and Grammar: the English Syntacticon. Mouton de Gruyter,
Berlin.
Emonds, Joseph. 2002. “Formatting Lexical Entries: Interface Optionality and Zero,”
Theoretical and Applied Linguistics at Kobe Shoin 5: 1–22.
Emonds, Joseph. 1985. A Unified Theory of Syntactic Categories. Foris, Dordrecht.
Emonds, Joseph and Rosemarie Ostler. 2006. “Thirty Years of Double Object Debates,” in M.
Everaert and H. van Riemsdijk, eds., The Blackwell Companion to Syntax. Blackwell
Publishers, Oxford.
Halle, Morris and Alec Marantz. 1993. “Distributed Morphology and the Pieces of Inflection,”
in K. Hale and S. J. Keyser, eds., The View from Building 20: Essays in Linguistics in
Honor of Sylvain Bromberger, MIT Press, Cambridge, 111-176.
Hayes, Bruce. 1990. “Precompiled Phrasal Phonology,” in S. Inkelas and D. Zec, eds., The
Phonology-Syntax Connection. University of Chicago Press, Chicago.
Hoeksema, Jack. 1988. “Head-types in Morpho-syntax,” Yearbook of Morphology, G. Booij and
J. van Marle, eds. Foris Publications, Dordrecht, 123-137.
Marantz, Alec. 1984. On the Nature of Grammatical Relations, MIT Press, Cambridge.
Ritter, Nancy. 1993. “Where’s Gender?” Linguistic Inquiry 24, 795-803.
Selkirk, Elisabeth. 1982. The Syntax of Words, MIT Press, Cambridge.
Veselovská, Ludmila. 2001. “Agreement Patterns of Czech Group Nouns and Quantifiers” in N.
Corver and H. van Riemsdijk, eds., Semi-Lexical Categories. Mouton de Gruyter, Berlin.
8
Download