35.1699, Review: Icelandic Nominalizations and Allosemy: Wood (2023)

[The LINGUIST List]

LINGUIST List: Vol-35-1699. Fri Jun 07 2024. ISSN: 1069 - 4875.

Subject: 35.1699, Review: Icelandic Nominalizations and Allosemy: Wood (2023)

Moderator: Francis Tyers (linguist at linguistlist.org)
Managing Editor: Justin Fuller
Team: Helen Aristar-Dry, Steven Franks, Daniel Swanson, Erin Steitz
Jobs: jobs at linguistlist.org | Conferences: callconf at linguistlist.org | Pubs: pubs at linguistlist.org

Homepage: http://linguistlist.org

Please support the LL editors and operation with a donation at:
           https://funddrive.linguistlist.org/donate/

Editor for this issue: Justin Fuller <justin at linguistlist.org>

LINGUIST List is hosted by Indiana University College of Arts and Sciences.
================================================================


Date: 07-Jun-2024
From: Malhaar Shah [mpshah at umd.edu]
Subject: Linguistic Theories, Morphology, Semantics, Syntax: Wood (2023)


Book announced at https://linguistlist.org/issues/35.236

AUTHOR: Jim Wood
TITLE: Icelandic Nominalizations and Allosemy
SERIES TITLE: Oxford Studies in Theoretical Linguistics
PUBLISHER: Oxford University Press
YEAR: 2023

REVIEWER: Malhaar Shah

In ‘Icelandic Nominalizations and Allosemy’ (INA), Jim Wood
investigates the interpretation of Icelandic derived nominals (DNs)
in-depth. The book is full of well-presented, carefully collected
data, and has rich theoretical discussion within the framework of
Distributed Morphology (DM). This review contains a brief summary of
the book and a critical evaluation of some of the main theoretical
claims.

INA relies heavily on a distinction between two kinds of derived
nominals: derived Complex Event DNs (CENs) and derived Referential DNs
(RNs) (Grimshaw 1990). Only CENs have obligatory arguments, a
necessary event-reading, allow agent-oriented modifiers (e.g.
‘intentional’), have argumental ‘by’-phrases, allow implicit argument
control, allow aspectual modifiers, and allow modifiers like
‘constant’/’frequent’ without pluralization. While the diagnostics are
not hard and fast, the distinction seems real (examples from Alexiadou
& Borer 2020).

Complex Event (Derived) Nominals (CENs)
(1) a. the instructor's (intentional) examination of the student
   b. the frequent collection of mushrooms (by students)
   c. the monitoring of wild flowers to document their disappearance
   d. the destruction of Rome in a day

Referential (Derived) Nominals (RNs)
(2) a. the instructor's examination/exam
    b. John's collections
    c. the frequent destruction*(s) took their/*its toll
    d. that exam of Mary's
    e. this kind of destruction (?? of cities)


SUMMARY

The puzzle at the heart of INA is, when CEN and RNs are so
semantically distinct, why are they systematically homonymous? Wood’s
answer is that they share the same syntactic structure. He challenges
‘phrasal layering’ accounts, which embed the layers of VoiceP, AspP,
and VP all under NP in CENs (3a), but not in RNs (3b).
Phrasal-layering theories then identify the CEN/RN distinction as
homophony between two different structures. The intuition behind both
Wood’s account and Phrasal Layering is that addition of “of the
concerto” in (3b) is bad for the same reason that ‘John recited the
concerto’ is bad.

(3) a. [NP [VoiceP [AspP [V recite]]]-al]
(=ACT OF RECITING; CEN)
John’s (intentional) recital of three poems (in ten minutes) caused
controversy.

b. [N [V recite]-al]
(=SOLO PERFORMANCE; RN)
John’s recital (*of the concerto) caused controversy.

Phrasal-layering theorists like Borer (2013) sometimes stipulate that
heads like Voice and Asp are always null and need to be licensed by an
overt free morpheme or a stem adjoining to them. Rather than
proliferate null heads, Wood’s proposal is that Icelandic derived
nominals always have the same syntactic representation. This makes
their homonymy unsurprising.

(4) [n [v [√recite]]-al] (ambiguous between CEN and RN)

Wood proposes that the differences between CENs and RNs emerge from
the nature of post-syntactic interpretive rules applying at Logical
Form (LF), which assign differing interpretations to v and n. These
rules are blind to the allomorphs for v and n selected at Phonetic
Form (PF), which explains why the semantic difference is not
correlated with a morphophonological difference. Wood contends that
(3a)-type structures are attested in the world’s languages — just not
Icelandic. Universal Grammar (UG) must then provide a
language-specific choice between (3a) and (4) as the structure
underlying a CEN.

The implementation of these post-syntactic rules must be carefully
constrained, in order to capture two important generalizations about
derived nominals. Wood dubs them Borer’s Generalization (after work in
Borer 2003; 2013) and Lieber’s Generalization (after work in Lieber
2017). Here they are as stated by Wood (p. 29):

(5) Borer’s Generalization: Complex Event Nominals are always built
off of an existing verb with the same meaning.
(6) Lieber’s Generalization: Every nominalizing affix that has an
eventive meaning also has one or more referential meanings.

(5) is the point that a CEN like ‘recital’ entails the existence of a
morphophonologically-related verb like ‘recite’ with the same meaning
and argument structure. (6) is the point that none of the nominalizing
affixes ‘ation’, ‘ment’, ‘al’, etc. disambiguate between a CEN or RN
meaning.

The main argument against phrasal layering is that, in Icelandic,
V-assigned lexical case is present on the arguments in clauses but not
on arguments of nominalizations. The reasoning is, roughly, as
follows:

(6)
1. Being in [Comp/Spec, XP] is necessary and sufficient for the
appearance of dative case on a theme when the verb is X.
2. Dative case does not appear in a nominalization whose verbal source
is X.
3. Therefore, the [Comp/Spec, XP] position is unavailable in the
nominalization of X.

Wood proposes that the arguments are base-generated in the nominal
extended projection (as [Spec, PossessorP] or [Spec, nP]), so that the
embedded verb cannot assign lexical case. Wood contrasts Icelandic to
Lithuanian, where lexical case is inherited (Šereikaitė 2020).
Lithuanian has, it is argued, two genitives (default and lexical).
Nominalizations of verbs which assign dative or lexical genitive to
their arguments retain those cases on the arguments. Wood argues that
Lithuanian has (3a)-type CENs and Icelandic has (4).

In Chapter 2, Wood motivates a number of empirical generalizations,
which I will list without illustrating.

1. The choice of nominalizing affix is determined by the root when v
is silent in [[[√ ] v] n], and by v when v is overt.
2. The gender of the noun is determined by the choice of n.
3. Verbs which select a preposition/particle (‘drive on’=ACCIDENT in
Icelandic) prefix P in the nominalization (like ‘on-driving’), even
when ‘he on-drove’ is ungrammatical. (Chapter 4 has a rich discussion
of this.)
4. The same telicity-diagnosing PPs (‘in/for an hour’) are available
with nominalizations as with the verbs they contain. Aspectual
modifiers and implicit argument control is possible, just like in
English.
5. Agents can be possessors. But ‘by’-phrases cannot be used with
Icelandic nominalizations. A special PP can be used to introduce
agents, but not the same PP as appears in passive.
6. Icelandic has no productive affix like English -ing (i.e., there is
no productive verb-to-noun affixation comparable to English gerundive
-ing).

There is also rich discussion of ECM and Raising in English
nominalizations.

The last two chapters develop Wood’s theory of allosemy in more
detail. A system is described in which the heads in (4) can each
receive different interpretations. The aim is to derive Borer’s
Generalization. Like Borer (2013), Wood rules out root suppletion in
order to ensure that verbs and their nominalizations are
morphophonologically related (i.e., to block the derivation of
‘metamorphosis’ from ‘change’, etc.).

To fully capture Borer’s Generalization, allosemy must be constrained.
Wood stipulates that roots are assigned meanings only after the
functional heads are interpreted. If a functional head is interpreted
as null (i.e., as the identity function), then it is deleted (‘pruned’
from the tree, in Ross’s phrase). This changes the locality relations.
So, in (4), if v is interpreted as null, then it deletes, making n and
√recite local. This means, if v is interpreted as null, n can
condition the choice of meaning for √recite.

On the other hand, if v receives a non-null interpretation, n cannot
trigger an idiosyncratic meaning of the root. So, suppose v is
interpreted as it would be in the clause — i.e., as a predicate of
events — then, the embedding n cannot affect the meaning of the root
past a meaningful v. Therefore, whenever the embedded root/v is
interpreted as a predicate of events (as in a CEN), then the
nominalization is compositional with respect to the embedded verb.
When v is interpreted as the identity function, then n can change the
meaning of the root, leading to RNs with unpredictable meanings with
respect to the meaning of the source verb, if that verb had appeared
in a clause. Wood ends up proposing six allosemes for n (null, simple
event, simple state, simple entity, result, and location), and at
least two allosemes for v (event, null).


EVALUATION

With respect to the main puzzle — the systematic ambiguity of CENs and
RNs — INA certainly provides an answer. Nominalizations are
homophonous between two readings because the negotiation of allomorphs
and the negotiation of ambiguity take place independently. The data
collection is also thorough, with many well-categorized examples. INA
really shines here. The theory is mostly laid out very clearly, with
many worked through examples of derivations for particular phenomena.

To move on to the bigger picture, INA is couched within the framework
of Distributed Morphology (DM). However, Wood’s proposals shift the
theory away from some influential claims in DM. A phase-theoretic
approach to categorizing heads (Marantz 2001; Embick 2010) is weakened
by allowing pruning at both PF and LF. Consider three categorizing
heads, n, v, and a, in the configuration [a [v [n [√]]]]. If n gets
pruned at PF, then v and √ are local at PF, but not at LF. And if v
gets pruned at LF, then n and a are local at LF, but not at PF. This
is unlike stricter phase-theoretic implementations as presented in,
e.g., Embick (2010, p. 48), which bans n from interacting with the
root in [n [v [√ ]]] even when the intervening v is null. Wood’s
implementation of DM then brings it closer to ‘non-simultaneous’
phase-based approaches (Marušič 2009; Borer 2013; and relatedly,
Preminger 2022), which allow for PF and LF locality to mismatch.
Perhaps this highlights one way in which consensus could be emerging,
from frameworks sometimes presented as very different.

Allosemy itself is a major part of the book, and the mechanism
receives explicit enough treatment to evaluate in depth. It is
contextual interpretation — item-specific rules of interpretation that
refer to its syntactic environment. Unlike most treatments of
allophony/allomorphy, allosemy does not typically determine a unique
output, in Wood’s implementation. I will outline what I think are some
core issues with it, before also mentioning some problems with complex
head formation.

I think Wood’s statement of Lieber’s Generalization is problematic.
Lieber’s observation that the same affixes are found in both CENs and
DNs is a little different from saying that an ‘affix that has an
eventive meaning also has one or more referential meanings’. It is odd
to say affixes like ‘ation’, ‘al’, etc. have meanings at all. They
change syntactic category from verb to noun, but why say that they
mean ‘refers to entity’? It is worth emphasizing this point to bring
out an asymmetry between allomorphy and allosemy. DM’s assignment of
(possibly zero) allomorphs to every head is intended to derive
blocking generalizations: why is √sing-PAST not ‘sing-ed’ or
‘sang-ed’? DM models this allomorphy as competition for realization of
a terminal node, PAST. In such cases, the ∅-allomorph wins over -ed,
because ∅ is specified to occur with √sing, and specific beats
general. But allosemy is not like this. Idiomatic readings do not
always block compositional readings (though there are such cases,
e.g., the meaning of ‘stand’ is absent in ‘understand’). At least for
the case of deverbal nominalization, the unpredictable meaning
coexists with the predictable one. This is more like ‘dived’/’dove’ in
allomorphy, but is also the usual case in deverbal nominalization.
Trying to extend the head-by-head interpretation to the
entity-referring/null/event-referring readings of little n/v/a
obscures how allomorphy and allosemy differ, and so, obscures the
motivation for DM’s handling of allomorphy.

There are some additional problems with LF-pruning. As mentioned,
LF-pruning is required to make n and √ LF-local in the RN reading for
[n [v [√]]]. But this risks having an unfortunate consequence. There
is no architectural constraint on non-compositionality. To see the
argument, consider the following Lithuanian diminutives (Veronika
Gvozdovaitė, p.c.):

(7a) ‘varna’ [n [√varn]-a] = √varn-F = CROW
(7b) ‘varniukas’ [n [n [√varn]-iuk]-as] = √varn-DIM1-M = SMALL
CROW/CROW CHICK
(7c) ‘varnėnas’ [n [n [√varn]-ėn]-as] = √varn-DIM2-M = STARLING
(7d) ‘varnėniukas’ [n [n [n [√varn]-ėn]-iuk]-as] = √varn-DIM2-DIM1-M
= SMALL STARLING/STARLING CHICK

Here, it seems that diminutives like -ėn can trigger special meanings
of the root. Lithuanian here creates no special problem for an
allosemy system. But it is desirable to rule out, say, (7d) possibly
meaning SMALL CROW. This is a part of Marantz’s (2013) aim, in ruling
out ‘semantic flip-flopping’. Introducing LF-pruning no longer
excludes that architecturally. To see this, we need to see that the
system would be fine with a hypothetical Lithuanian*, where (7d) meant
(7b). This would require only pruning of a semantically null -ėn in
(7d), making (7d) and (7b) identical at LF. Assuming such cases are
unattested, there is a real sense in which that is a coincidence for
the system. Wood himself is sensitive to this issue (INA, p. 297):
“this proposal does seem to resemble the one in Borer (2013, 2014) or
in many lexicalist accounts where an arbitrarily large structure can
get a special, idiosyncratic “encyclopedia” entry. Borer and others
have proposed that the Voice head, or more descriptively the external
argument layer, is where idiosyncratic meaning is ‘fixed’.”

It is worth noting that Borer’s theory is actually somewhat different
on this point, in that it does have a way to force compositionality.
Borer’s system only allows non-compositionality with what she calls
‘C-functors’ — roughly, category-changing morphology, which is
semantically erratic but morpho-phonologically more regular. This
allows for non-compositional readings of, e.g.,
‘natur-a-liz-ation’=BECOME CITIZEN, and ‘edit-or-ial-ize’=GIVE
OPINION. She calls heads like Tense, Asp, Num, etc., ‘S-functors’,
which are irregular in their realization, but consistent semantically.
For her, merging any S-functor forces compositionality (so, the
idiomaticity-domain is actually smaller than Voice for her). In her
system, S-functors are not subject to an allosemy-like mechanism,
placing sharp constraints on non-compositionality. This is quite
different from Wood’s system, which allows heads like Voice to be
subject to allosemy.

Bringing out this difference is important, because in Wood’s system
Voice could, in principle, have a null, prunable alloseme. Why, then,
does Tense never condition an idiosyncratic interpretation of V? Or,
why do we not see C-conditioned allosemy when Tense is semantically
null in infinitives? One reason to favor Borer’s Phrasal Layering
system over INA’s, then, is that it has a mechanism for delivering
compositionality as the rule, and for seeing idiomaticity as the
exception.

On a narrow-syntactic note, given the centrality of the
phrase-structural assumptions of the proposal, it is lamentable that
these are not clearly laid out. It is utterly crucial to Wood’s
proposal that complex heads can be built without movement. The analogy
is to adjunction — adjunction does not change the bar level of its
target. Why stipulate adjunction to not operate on heads? Why
stipulate it to only happen as a species of movement? Yet, even if we
grant that argument and relax those conditions on adjunction, I think
this does not get us the required results. Wood wants the nominalizing
head to be constrained in Icelandic to merge/adjoin to v-heads but not
to vPs. This is richer than not affecting the bar-level of the target;
n must also choose the bar-level of the target.

I am not convinced that this is stateable in a Minimalist grammatical
architecture. Let me lay out the argument more fully. √fjöl means
‘increase’ (INA, p. 114). In its verbal instantiation (8a), it takes a
dative theme, but not when nominalized (8b).

(8a)
Addamshjónin fjöl-g-uðu börnunum
the.addamses.NOM increase-VBLZ-PST children.the.DAT
‘The Addamses had more children.’

(8b)
Fjöl-g-un barnanna vakti athygli
increase-VBLZ-NMLZ children.the.GEN drew attention
‘The increasing of the children drew attention.’

(8c) *Fjöl-g-un börnunum vakti athygli
increase-VBLZ-NMLZ children.the.DAT drew attention

Wood’s argument from these examples is that the nominalization must be
deverbal — it contains an overt verbalizer -g, yet we see that the
theme loses dative (from the ungrammaticality of (7c)). If we grant
the assumption that the -g in (8a) and (8b) is the same, at some point
in the derivation, we have (9a) built up.

(9a) [v [√fjöl]-g]

Let us grant that this is a head. We need the nominalizer to be
restricted in its distribution, such that it can merge with (9a), but
cannot merge with (9b).

(9b) [vP [√fjöl]-g [DP börnunum]]

(9c) [n [v [√fjöl]-g]-un]
(9d) *[n [vP [√fjöl]-g]-un]

To the extent that Bare Phrase Structure (Chomsky 1995) is adopted,
‘phrase’ and ‘head’ are relational notions. A head is something drawn
from the lexicon, and a phrase is a maximal instantiation of a head.
If BPS is adopted, then (9c) definitionally contains a vP. So, Wood
needs a richer system than BPS to distinguish (9c) and (d). But
enriching the grammar with bar-levels is not enough. The nominalizer
must also be sensitive to the bar level of the item it externally
merges with. That is, the nominalizer must know whether the v it
merges with has projected to a phrase (by taking an argument) or
remained a head. I am not sure how to do this without introducing some
sort of selection of particular bar levels like X⁰s.

As mentioned, this is the core theoretical proposal in INA. If it, in
fact, requires enriching the grammar with not only bar-levels, but
also the ability of heads to select for heads (as distinct from the
phrases their complements project), then more must be said about the
phrase-structural system. I do not want to suggest that this is
unworkable, but would have been nice to see the issue addressed.

I do not think these are minor concerns, but they do not detract from
the overall high quality of the book. INA is an excellent in-depth
study of Icelandic nominalizations. A wealth of puzzles are described
in detail. Readers interested in the state of the art in Distributed
Morphology will surely learn much from the book. The empirical
generalizations which INA seeks to capture are clearly described and
motivated. However, the theoretical proposals are often in conflict
with some other attractive, established theories. The major avenues
for research in this direction are to either further develop the
theory of phrase structure and allosemy in INA such that its problems
can be resolved, or to reanalyze the Icelandic data without appeal to
such mechanisms. It is a testament to its empirical thoroughness and
theoretical imagination that, whichever direction the reader finds
more promising, INA is a first-class resource.


REFERENCES

Alexiadou, A., & Borer, H. (Eds.). (2020). Nominalization: 50 years on
from Chomsky’s remarks. Oxford University Press.
Borer, H. (2013). Structuring Sense Volume III: Taking form. Oxford
University Press.
Chomsky, N. (1995). The minimalist program. The MIT Press.
Grimshaw, J. B. (1990). Argument structure (LI Monographs 18). MIT
Press.
Embick, D. (2010). Localism versus globalism in morphology and
phonology. The MIT Press.
Halle, M., & Marantz, A. (1993). Distributed Morphology and the pieces
of Inflection. In S. J. Keyser & Hale, Ken (Eds.), The View from
Building 20 (pp. 111–176).
Lieber, R. (2017). English nouns: The ecology of nominalization.
Marantz, A. (2001). Words. Ms. MIT.
Marantz, A. (2013). Locality Domains for Contextual Allomorphy across
the Interfaces. In O. Matushansky & A. Marantz (Eds.), Distributed
Morphology Today (pp. 95-115).
Marušič, F. L. (2009). Non-Simultaneous Spell-Out in the Clausal and
Nominal Domain. In K. Grohmann (Eds.), InterPhases (pp. 151-181).
Preminger, O. (2022). Natural language without semiosis. Slides.
Šereikaitė, Milena. 2020. Voice and Case Phenomena in Lithuanian
Morphosyntax. University of Pennsylvania Doctoral Dissertation.

ABOUT THE REVIEWER

Malhaar Shah is a PhD student in Linguistics at the University of
Maryland. He holds a BA from Oxford University, in Philosophy &
Linguistics. He is interested primarily in syntax and how it interacts
with semantic composition and morphophonology, as well as experimental
semantics.



------------------------------------------------------------------------------

Please consider donating to the Linguist List https://give.myiu.org/iu-bloomington/I320011968.html


LINGUIST List is supported by the following publishers:

Cambridge University Press http://www.cambridge.org/linguistics

De Gruyter Mouton https://cloud.newsletter.degruyter.com/mouton

Equinox Publishing Ltd http://www.equinoxpub.com/

John Benjamins http://www.benjamins.com/

Lincom GmbH https://lincom-shop.eu/

Multilingual Matters http://www.multilingual-matters.com/

Narr Francke Attempto Verlag GmbH + Co. KG http://www.narr.de/

Wiley http://www.wiley.com


----------------------------------------------------------
LINGUIST List: Vol-35-1699
----------------------------------------------------------