ZigaForm version 4.6.9

Cooperation and coercion

Cooperation and coercion
Jenny Audring
University of Amsterdam
j.audring@uva.nl
Geert Booij
Leiden University
G.E.Booij@hum.leidenuniv.nl

Abstract: Coercion is a much-discussed topic in the linguistic literature. This article expands
the usual range of cases at the most subtle and the most extreme end: it demonstrates how
coercion extends into semantic flexibility on the one hand and into idiomaticity on the other.
After discussing a broad variety of coercion cases in syntax and morphology and briefly
reviewing the equally diverse literature, we identify three mechanisms – selection, enrichment
and override – that have alternatively been proposed to account for coercion effects. We then
present an approach that combines all three mechanisms, arguing that they can be unified
along a single axis: the degree of top-down influence of complex structures on lexical
semantics.
Keywords: coercion, constructions, semantics, polysemy, idioms
Acknowledgements: We thank Ray Jackendoff and two anonymous reviewers for comments
and advice. Jenny Audring is grateful to the Netherlands Organisation for Scientific Research
(NWO) for a Veni grant, #275-70-036.
1 Introduction
Sometimes, words are used in surprising ways. Consider example (1) from Dutch:
(1) a. Een joggende boerka is nog geen terrorist
‘A jogging burka is not yet a terrorist’
b. Rara, een joggende djellaba
‘Guess what, a jogging djellaba’
(NRC newspaper, 21 June 2014)
The process that allows for (1a) and b) to be felicitous sentences of Dutch is known as
coercion. In coercion, the utterance context favors or enforces a particular reading of a word,
here, a person reading for the two nouns denoting items of clothing. Since this reading lies
outside the usual semantic range of boerka and djellaba, the coercion effect is clearly
noticeable and can be exploited stylistically.
The person reading in (1) is the consequence of a semantic clash: joggende requires a human
agent, a condition not satisfied by the inanimate subject nouns. This confirms the well-known

2

observation that coercion is a repair strategy or “compromise construal” (De Swart 1998;
Michaelis 2004) – a bridge between a phrase and an incompatible word appearing in it.
While in (1) the head of the NP is coerced into a person reading to accommodate the
participle, in other cases it is the verb that undergoes coercion. In a classic example
(discussed, among others, in Verkuyl 1972; Talmy 1978; and Jackendoff 1991), a temporal
adverbial changes the aspect of a verb from point-event to state by introducing iteration (2a).
A similar effect occurs with all night in (2b) and with the result adverbial in (2c).
(2) a. The light flashed until dawn.
b. We played poker and she won all night.
c. He knocked on the door until his knuckles ached.
Again, we see a semantic conflict – here, between point-event and time adverbial – and a
form of resolution. Resolution can follow a wide variety of strategies. In many cases it
involves established classes, such as the types of verbal aspect as in (2), or semantic/
pragmatic strategies such as metonymy, as in (1). The observation of typical resolution
pathways has led researchers to assume that coercion is effected by conventionalized
operators or constructions (Jackendoff 1991, 1997, 2013; Fillmore and Kay 1993). One such
operator with the meaning ‘activity commonly associated with N’ might be proposed for (3):
(3) I’m done with…

  • the windows
  • the letter
  • the salt
  • the taxes
  • the sunflowers
  • the route
  • the wardrobe

The semantic conflict in (3) arises from combining the expression done with X, which selects
for an activity, with a noun denoting an entity. Repair is realized by inserting an implicit
predicate, which by default is chosen to denote the activity commonly associated with the
entity in question, e.g. cleaning the windows, reading or writing a letter, using the salt, paying
taxes, drawing or painting sunflowers, planning a route, and tidying a wardrobe.

1 While the
spirit of the solution is the same in all instances, the range of the implicit predicates is
virtually infinite: any noun could in principle trigger a new semantic type.

1 Complement coercions as in (3) have sparked a substantial literature, theoretical as well as
psycholinguistic (e.g. Briscoe, Copestake and Boguraev 1990; Pustejovsky 1995; Jackendoff
1997; Lapata and Lascarides 2003; Lapata, Keller and Scheepers 2003; Pustejovsky and Jezek
2008; Kuperberg et al. 2010).

3

Another operator, which might be characterized as ‘model/ replica of N’ can be seen in
example (4) from Jackendoff (1997: 65).
(4) a wooden turtle
In (4), the adjective wooden conflicts with the noun, which denotes an animate being.
Resolution goes by way of coercing the meaning of the noun into ‘replica turtle’ (here
probably construed as ‘toy turtle’). The example nicely illustrates that coercion runs on
incompatibility; as Jackendoff points out, the ‘replica’ reading does not arise with objects for
which being wooden is unremarkable, as in a wooden spoon.
In yet other cases, resolution is more specific and less easy to capture by a standard operator.
Consider example (5), from our own observations.
(5) Zo lukt Kerst!
‘This way, Christmas will be a success!’
In this Dutch advertisement, the verb lukken ‘to work out right, to be a success’, which
usually selects a challenging activity, is combined with a holiday event, subtly capitalizing on
the fact that the management of such an event may indeed be felt to be a challenge.
In yet other instances, resolution in coercions is entirely fixed and does not vary with the
semantics of the coerced element. This can be witnessed in constructional coercions such as
the Dutch example in (6) below (also discussed in Cappelle 2014). The constructional idiom
REFL DET X schrikken is used to express degree of alarm when a person is startled, similar to
English ‘scared to death’. The quasi-resultative NP element in the construction can be
instantiated by a (limited) number of article-noun combinations, a selection of which is given
in (6). The effect is identical in all cases: the NP is coerced into an intensifier reading, such
that all variants mean ‘I’m startled a lot’.
(6) Ik schrik me… ‘I’m scared to … (lit. ‘I frighten myself …’)’

  • een hoedje ‘a hat (DIM)’
  • een ongeluk ‘a mishap’
  • de tering ‘the consumption/ tuberculosis’
  • een aap ‘a monkey’
  • de tandjes ‘the teeth (DIM)’
    What is most interesting is the diversity of fillers for the intensifier slot. While nouns denoting
    sickness or misfortune are not unexpected in such contexts, hats, monkeys and teeth are
    definitely unusual. We will return to such cases in the discussion of radical coercion below.
    The examples in this section show that coercion is a diverse and pervasive phenomenon. The
    aim of this article is to situate it in a larger context. We argue that, on the one end, coercions

4

constitute the tip of a metaphorical iceberg, whose body consists of contextual adjustments of
word meaning. On the other end, we claim, coercion runs into idiomaticity, an extreme case
of modulating lexical meaning in composition.
2 ‘Soft coercion’ – semantic flexibility
While examples such as (1) catch the eye of readers and linguists alike, the overwhelming
majority of coercions is so subtle that they quite go unnoticed. Indeed, the phenomenon is so
ubiquitous that the term might be reserved for the more drastic cases. However, it is worth
considering how smoothly the classical instances of coercion run into the negotiations that are
part and parcel of the construal of a word’s meaning in context. Consider, for instance, the
words lemon, egg and knocked in the following sentences:
(7) a. I’d like some water with a bit of lemon.
b. You’ve got egg on your shirt.
c. He knocked on the door.
The three words, strictly speaking, have alternative readings: the lemon in (7a) can be a slice
of lemon or a quantity of lemon juice, the egg in (7b) can be raw or cooked and knock in (7c)
can denote a one-time or a repeated activity. In the utterances (8) to (10) below, by contrast,
the ambiguities are resolved.
(8) a. Cut the lemon in half.
b. Drizzle the lemon over the cake.
(9) a. Put some egg into the potato salad.
b. Massage some egg into your hair and rinse.
(10) a. He knocked on the door and waited.
b. He knocked on the door until his knuckles ached.
Lemon clearly refers to a piece of fruit in (8a), but to juice in (8b); the egg in (9a) is cooked,
but raw in (9b), and the event in (10) switches from one-off (10a) to iterative (10b). In the
linguistic literature, (10b) is considered a case of coercion, while (8a) to (10a) are not.
However, the six examples have something in common: the context of the utterance triggers a
particular reading of a word. Hence, the more canonical cases of coercion can be seen as an
extension of an everyday phenomenon. The difference lies in the amount of force exerted by
the context. Alternations such as (8) – (10) are subtle, while the better-known cases of
coercion involve more drastic changes. An example are type coercions, which will be
discussed next.
3 Type coercion

5

Semantic coercion may involve type coercion, in the sense that a word is moved into another
word class or subclass. Type coercions can be found with all the major word classes.
For verbs, aspectual coercions are common, as shown in (2) above. Similar examples are the
sentences in (11), where for months introduces repetition or habituality to the event, while the
adverb suddenly gives a state verb know an inchoative reading (De Swart 1998: 359).
(11) a. For months, the train arrived late.
b. Suddenly, I knew the answer.
A different case of verbal coercion can be found in the use of the English prefix un-, which
can change the semantic class of the base verbs. As Bauer et al. put it (2013: 374): “un- can
take a stative, activity or other kind of verb and force it into a causative/inchoative verb that
implies a reversible result.” Examples are the verbs un-inhabit, un-grow, un-see, un-have and
un-hit. Social media provides further examples, from unfollow (on Twitter) to unfriend
somebody or unlike (i.e. remove the “like” on) a post (on Facebook). Another telling example
from our own observations is an instrument maker discussing the precision required in
manufacturing a flute and remarking that “you cannot undrill a hole”.
For nouns, typical kinds of type coercion are those between count and mass noun and between
proper name and common noun (see (12a) and (b), courtesy of Michaelis 2003), though many
other variants are possible (see Pustejovsky 1991: 432 for a list of common types).
2

(12) a. some pillow (count > mass)
b. a pudding (mass > count)
c. We’ve got three Pauls in the family. (proper name > common noun)
d. I’m practicing Brahms. (proper name > common noun)3
For adjectives, coercion can result in shifts from a relational to a qualifying adjective (13).
(13) a. This building looks very American.
b. You’re renting your home? That’s really un-Dutch.
Type coercion can also involve conversion, i.e. a change of word class. An interesting
example comes from Dutch (Booij and Audring forthcoming). Dutch has V > N conversion,

but its productivity is marginal. However, in the context of certain constructions, coercion-
driven conversions are surprisingly productive. This is the case in the aan de [V]N gaan-
construction (literally meaning ‘go at the …’). Internet search provides an impressive number

of instances; a small selection is given in (14):

2 We will give a different, more nuanced analysis of these examples in Section 6.3 below.
3 Examples such as (12d) are often discussed under the name of reference transfer (Nunberg
1979; Fauconnier 1985).

6

(14) a. Het gebeurt vaker dat ze ineens aan de vreet gaan
‘It happens more often that they start eating all of a sudden’
b. Zelfs een dynamo kan aan de ratel gaan
‘Even a dynamo can start rattling’
c. Gewoon buiten aan de ren gaan.
‘Just get running outside’
d. Als het warm is wil hij af en toe een stukje aan de zwem gaan.
‘When it is warm, he wants to go swimming a bit from time to time.’
In the above examples, the verb stem is used in a noun slot and thereby coerced into a class it
does not usually belong in. The semantic effect is inchoative: the construction focuses on the
start of the activity. None of the nominalizations in (14) exist outside the construction, at least
not with the relevant meaning.
Concluding the rough overview of the formal consequences of coercion, (15) provides an
example of incidental type coercion involving a change in word class from pronoun to noun.
(15) Wie dient, denkt niet alleen in ‘ík’.
‘Who serves, does not only think in ‘I’.’
Wie dient, denkt niet alleen in ‘zij’.
‘Who serves, does not only think in ‘they’.’
Wie dient, denkt ook in ‘wij’.
‘Who serves, also thinks in ‘we’.’
(Trouw newspaper, 14 January 2014, general army commander Peter Uhm)
In the first line of (15), the interpretation of ik is not that of a 1st person singular pronoun;
instead it denotes ‘self-interest’. Similarly, zij ‘they’ in the second line is interpreted as
‘others’, and wij ‘we’ in the third as ‘common interest’. Hence, these words function as nouns
semantically.

4 In this case, we know for sure that this correlates with a change of word class:
if the words ik, zij, wij still had the status of pronouns, they would have to appear in oblique
form after a preposition, which would be mij, hen and ons, respectively.
4 Coercion in constructions
A number of the cases explored above involve constructions, i.e. conventionalized pairings of
(complex) form and meaning. Coercion effects are of particular interest to Construction
Grammar, in a similar fashion as they were interesting to formal semantics in earlier decades:
they are prime examples of holistic meaning. In coercions, the utterance meaning is more than
(or different from) the sum of the lexical meanings of the individual words. From a
constructionist perspective, this is expected behavior for larger unitary structures: “coercion
constitutes a major argument in favor of the existence of constructions as independent
4 Note that (15) does not constitute a case of use-mention: there is no reference to the (use of
a) word; instead, the pronouns are given a new meaning.

7

form/meaning pairings, since it can be used as a heuristic means to discover the independent
constructional semantics” (Lauwers and Willems 2011: 1220).

5 At the same time,
constructions are a particularly fruitful environment to look for coercion data, as “[a]ny
construction that selects for a specific lexical class or phrasal daughter is a potential coercion
trigger” (Michaelis 2011: 1383-1384).
An illustrative example of (type) coercion in constructions is the (all) X-ed out construction
described by Jackendoff (2013 and elsewhere):
(16) a. If you’re not all festivaled out this summer head for The Moors Festival
b Just in case you’re not all Biebered out already, here’s the full studio version of
“Mistletoe” (referring to a Justin Bieber recording)
c. By midnight:30 I was all Amsterdammed out.
(examples from www.google.com, searched 19 November 2014)
This conventional pattern, which could be schematized as (all) [[X]-ed]V out refers to a state
induced by an activity, ‘exhausted from X-ing to excess’. When X is not a verb but a noun
and thus lacking the activity reading, the effect is coercive: ‘exhausted from experiencing X
to excess’ (Jackendoff 2013: 89, though without reference to coercion). The effect is as
predictive as the construction is productive: almost any noun can be coerced into a tiring
activity in this manner.
A very similar construction can be found in Dutch (example 17). Here, the noun kleuter
‘toddler’ is used as the root of a separable complex verb, resulting in the participle
uitgekleuterd ‘toddlered out, done with raising toddlers’. This coinage is particularly
interesting, as Dutch does not have the particle verb uitkleuteren (nor, for that matter, the base
verb kleuteren, at least not with the relevant meaning), so the noun has been inserted directly
into the participle schema.
(17) Mijn dochter is nu uitgekleuterd
‘My daughter is done with raising toddlers now’
The semantic effect of the coercion is in tune with other uit-X participles, which mean ‘done
with X’. (For details and a more extended discussion of construction-dependent morphology,
see Booij and Audring 2007).
Another example is the use of nouns in adjective slots in French (Lauwers 2014: 206):
(18) des costumes très ‘théâtre’
‘very theatre-like costumes’
5 Holistic meaning is often referred to as “non-compositional”. However, this term does not
seem appropriate if coercions are taken to involve a combination of lexical and constructional
meaning, in which case the result would be compositional after all (see e.g. Michaelis 2003).

8

The construction is similar to its English equivalent, which enjoys a special productivity with
years (19):
(19) This is so 2013!
In both English and French, the noun or number is forced into an adjectival reading, in
particular that of a qualifying adjective meaning ‘typical of X’.
Interestingly, constructional coercion, including type coercion, is not limited to syntax. Booij
and Audring (forthcoming) discuss coercion from the point of view of Construction
Morphology, demonstrating how inflection and derivation can produce type changes. An
example (from Grandi et al. 2011) is given in (20). In this Italian newspaper headline, the
noun bomba ‘bomb’ is used with a superlative suffix usually reserved for adjectives. The
semantic effect is that of intensification: ‘news like a bomb, breaking news’.
(20) Notizia bombissima! Priest Holmes si ritira?
‘Breaking news! Is Priest Holmes withdrawing?’
Leaving type coercions, but staying with constructional coercions, we find idiomatic
expressions such as the Dutch example in (6) above, which constitute an extreme case of
construction-determined meaning. Another remarkable case is the German idiom jemandem
auf DET X gehen ‘to get on somebody’s nerves’. Next to die Nerven ‘the nerves’, a wide
variety of items can be found in the DET X position; no overall semantic or formal pattern is
discernible (though some words appear to be euphemisms for certain body parts, and neuter
nouns are rare). A selection is given in (21):
(21) Er geht mir auf… ‘He’s getting on my… (lit. ‘He goes me on the…’)’

  • die Nerven ‘the nerves’
  • den Sack ‘the sack’
  • den Keks ‘the cookie’
  • den Wecker ‘the alarm clock’
  • die Eier ‘the eggs’
  • die Ketten ‘the chains’
  • den Docht ‘the wick’
  • den Trichter ‘the cone’
  • den Zeiger ‘the pointer/ the clock hand’
  • den Senkel ‘the shoelace’
  • das Schwein ‘the pig’
    The essential observation is that the lexical semantics of the nouns does not contribute to the
    utterance meaning. Instead, as in (6) above, the expression means exactly the same, whatever
    item is inserted. While the NP slot in the construction is partially open – new variants cannot

9

be added completely at random, but the broad range of filler items suggests a light degree of
productivity – the expression is fully and non-compositionally idiomatic and any NP is
coerced into the appropriate function.
5 Theoretical accounts
Considering the wide range of coercive modifications from contextual adjustment to
embedding into idioms, it is not surprising that theoretical accounts are diverse and
heterogeneous. The diversity is apparent in the metaphors used to describe the observed
effects. Across the literature, three views are prevalent.
The first view discusses coercion as a cooperation between lexical and phrasal meaning,
exploiting alternative readings of a lexical item.

6 Such cooperation is sometimes called CO-
COMPOSITION (Pustejovsky 1991, 1995). An illustrative example is the double functionality of

words such as bake or lunch in (22). As argued in Pustejovsky (1995: 47), bake can either
encode a change of state (22a) or a means of creation (22b). Lunch, in turn, can be an event
(22c) or a kind of food (22d) (Pustejovsky 2011: 1403).
(22) a. John baked the potato
b. John baked the cake.
c. Mary left school after lunch. (event)
d. Mary brought lunch to school. (food)
Such cases appear to involve SELECTION (Pustejovsky and Jezek 2008; Pustejovsky 2011): the
context selects the appropriate meaning from within the semantic range of the word. While
selection is often excluded from coercion proper, it is frequently discussed in the context of
coercion and therefore clearly perceived as a closely related phenomenon. In fact, examples
such as the author will discuss her book, which involves the selection of aspects of lexical
semantics, i.e. the informational content associated with book, are considered coercions in
Pustejovsky and Jezek (2008: 188).
A second view represents coercion as an operation involving the addition of unexpressed
semantics to the utterance. This view is endorsed by Jackendoff (1991, 1997, 2013 and
elsewhere), and phrased most helpfully in Culicover and Jackendoff (2005: 228). Coercion is
said to mediate between a function and an incompatible argument the way an adapter
mediates between an incompatible socket and plug. For example, the implicit predicate in
(23), informally represented as […], connects the verb began, which requires an action
argument, with the object argument the book.
(23) Mary began […] the book.
6 A discussion of theoretical semantic issues, such as the difference between polysemy,
ambiguity, and vagueness, is beyond the scope of this paper (but see e.g. Willems 2013 for a
recent account in relation to coercion).

10

The relevant operation is taken to be ENRICHMENT: the compositional utterance meaning is
enriched with the help of a semantic operator lacking overt form. The argument in favor of
this approach is that the book in (23) is still an object in the final interpretation of the sentence
– the reading event merely augments the semantic composition. Note how this analysis is
confirmed by pronominalization: the anaphoric pronoun it in (24) refers to the book as an
object.7
(24) Mary began the book after it had been sitting on her shelf for years.
The third view, which is the most widespread and found from the earliest to the most recent
accounts (e.g. Partee and Roth 1985; De Swart 1998; prominently Pustejovsky 1991, 1995,
2011; Pustejovsky and Jezek 2008; Michaelis 2003, 2004 and 2011), characterizes coercion as
an actual change of properties. In this literature, coercion is discussed in terms of a “type
shift”, “transition” , “modulation”, “conversion” or “meaning shift”. Here, the relevant
mechanism is captured in the term OVERRIDE. The contrast between enrichment and override
can be illustrated with (23) above, for which Pustejovsky (2011: 1406) gives the traditional
account as in (25):
(25) Mary began [the book]event.
Rather than inserting a mediator between the verb and the misfitting argument, the argument
is taken to be changed into a matching type. While this effect can be attributed to an operator
as in Jackendoff’s account, the difference between the two approaches lies in the effect the
operator is taken to have: adding to the lexical meaning vs. modifying it.
The latter kind of approach has enjoyed recent popularity in constructionist accounts of
coercion, in particular the work by Michaelis (2003, 2004, 2011), who formalizes it as the
Override Principle (Michaelis 2003: 9).
Override principle: If lexical and structural meanings conflict, the semantic
specifications of the lexical element conform to those of the grammatical structure with
which that lexical item is combined.
Informally put, the argument is that larger linguistic structures are a stronger force than the
individual word and are therefore able to modify properties of the coerced item.
By contrasting the three types of account in this brief (and admittedly superficial) way, we do
not intend to compare or argue in favor of one or the other approach. Instead, we suggest that
together they form a useful typology for coercion phenomena, each accounting for different
cases. The ‘spine’ of the typology is formed by the degree to which the original meaning of a
coerced item is preserved in the final interpretation.
7 We thank an anonymous reviewer for reminding us of this fact.

11

In coercion by selection, the resulting meaning is a part of the semantic repertoire of the
coerced word to begin with. From this perspective, coercion works largely ‘bottom-up’, with
only a light role for the context selecting one interpretation from a range of alternative
readings.
In coercion by enrichment, lexical semantics is preserved, but augmented in context. The
“adapter plug” represents a stronger ‘top-down’ influence, adding meaning to the utterance.
In coercion by override, in turn, contextual ‘top-down’ force is strongest; it modifies, replaces
or removes properties of the coerced item.
In the remainder of this article, we will revisit a number of the examples presented above in
the light of this three-way typology.
6 Reconciling accounts
6.1 Coercion by selection
Selection is the lightest form of coercion; in fact, many researchers consider it to be outside
the range of the term. However, we argue that contextual adjustment involving selection
constitutes the invisible mass of the iceberg, the tip of which are the more noticeable cases of
coercion. Consider again examples (8) and (9), repeated here as (26) and (27).
(26) a. Cut the lemon in half.
b. Drizzle the lemon over the cake.
(27) a. Put some egg into the potato salad.
b. Massage some egg into your hair and rinse.
While lemon has alternative readings as an item or a liquid and egg allows interpretation as a
raw substance or prepared food, some of the possible readings are incompatible in (26) and
(27). Thus, the liquid reading of lemon fails to match with cut in (26a), just as the item
reading does not fit the semantic frame of drizzle in (26b). In (27), the effect is more subtle
and depends on world knowledge; yet, a raw egg reading in (27a) and a boiled egg reading in
(27b) would produce odd semantic effects. The context thus coerces the lexical meaning into
appropriateness just as in the more familiar cases, only in a less noticeable manner.
Another argument for coercion-by-selection can be illustrated with the help of the contrast
between (28a) and b) (inspired by Pustejovsky 1995: 47).
(28) a. drop/ discuss the book
b. want/ finish the book

12

While (28b) is a considered a classical example of coercion and (28a) is not, the effect is
virtually the same: in both examples, the verb selects among the various semantic properties
(“qualia”) of book. Drop and want select the ‘physical object’ quale, while discuss and finish
refer to the ‘informational content’ quale. Treating the two cases as fundamentally different
would be missing an interesting point.
The view that coercion is actually a widespread, common and unremarkable process, with
only the borderline cases attracting attention (of speakers and linguistic researchers), is hinted
at in several places in the literature (e.g. Pustejovsky 1995: 109; Jackendoff 1997: 62;
Michaelis 2003; Harder 2010: 247; Suttle and Goldberg 2011: 1238). We support this view by
including contextual adjustment in our account.
6.2 Coercion by enrichment
The mechanism of enrichment can account for a large variety of coercions. A classic case are
aspectual coercions, discussed in many sources (e.g. Talmy 1978; Pustejovsky 1991;
Jackendoff 1991, 1997 and 2005). We gave three examples in (2) above, repeated here as
(29).
(29) a. The light flashed until dawn.
b. We played poker and she won all night.
c. He knocked on the door until his knuckles ached.
While such cases are occasionally taken to be transitions from one aspectual class, say
punctual, to another, say iterative, the two classes are actually not in opposition. In fact, for
the coercion to work, the short duration of the event is necessary to trigger the iterative
reading. The reason is, as Jackendoff (1997: 51) argues, that until or all night refer to an
ongoing event and that flashing, winning and knocking have an inherent ending and hence
cannot be construed as ongoing. The semantics of the utterance needs to be enriched by an
operator, a construction or a function denoting iteration in order to create an activity that can
be bounded by until or all night. Thus, the utterance ends up having a piece of semantics that
has no phonetic form but restores its semantic felicity.
The same effect can be seen in (30a), where arrive requires stretching-by-repetition in order
to be boundable by for months, while in (30b), suddenly homes in on the only point event in
the state denoted by know: the beginning.
(30) a. For months, the train arrived late.
b. Suddenly, I knew the answer.
In all instances, the coerced meaning is negotiated cooperatively, as a combination of
expressed and unexpressed meaning, the core claim of the enrichment approach.

13

Note again that – as argued for (28) above – there is no sharp break between aspectual
coercions and ambiguity resolution. Compare the sentences in (31).
(31) a. Chico drank the wine.
b. Chico drank the wine in an hour.
c. Chico drank the wine all night.
(31a) is ambiguous as to whether Chico finished the wine or not (though there might be a bias
towards completion). (31b) and c) are disambiguated by the adverbials: b) refers to a
completed action, c) does not. Here, the elements that resolve this ambiguity are the same
elements that create aspectual coercion as in (29) above.
Another example illustrating enrichment is (32), repeated from (3) above.
(32) I’m done with…

  • the windows
  • the letter
  • the salt
  • the taxes
  • the sunflowers
  • the route
  • the wardrobe
    This kind of coercion – complement coercion – is another classic in the literature, usually
    discussed with the verbs begin or finish, as in begin the book or finish the beer. In all
    instances, the verb or construction requires an activity predicate as complement. If instead of
    a predicate we find a nominal referent, the coercion enriches the utterance by an implicit
    predicate to fit the semantics of the noun.
    Finally, the creative word choice in (1)/(33) can be explained with recourse to enrichment.
    (33) a. Een joggende boerka is nog geen terrorist
    ‘A jogging burka is not yet a terrorist’
    b. Rara, een joggende djellaba
    ‘Guess what, a jogging djellaba’
    (NRC newspaper, 21 June 2014)
    Since the semantics of the participle requires an animate agent, an ‘adapter’ is needed to ‘plug
    in’ an inanimate noun. Here, the inserted function is ‘person wearing X’, a subtype of the
    function ‘person contextually associated with X’, proposed by Jackendoff (1991, 1997, 2013)
    in order to account for reference transfers. Again, enrichment appears to be a better account
    than, say, override, because the semantics of boerka and djellaba are not lost in the coercion,

14

they are merely augmented. This is the core argument in favor of enrichment: the original
semantics of the coerced lexical item is still present in the final interpretation.
6.3 Coercion by override
Depending on the exact understanding of the process, enrichment and override have a lot in
common. In both cases, a construction or an operator is assumed to add a semantic or formal
property to the utterance. The difference lies in the effect this has on the original lexical
properties. In enrichment, they remain intact and are merely augmented or ‘wrapped’ with the
new specification. In override, the original properties are replaced or removed. Still, there are
cases where either analysis seems possible. Consider (34), repeated from (12) above.
(34) We’ve got three Pauls in the family. (proper name > common noun)
In (34), Paul is coerced from a proper name to a common noun, which licenses the use of the
numeral and the plural ending. In this view, the example is a case of override: the category
[proper name] is replaced by [common noun], a classical type coercion. However, we could
also posit an implicit nominal referent meaning ‘persons named’, of which Paul is the
argument. This referent would sit between three and Pauls and enrich the utterance in a way
discussed in Section 6.2. Under this analysis, Paul is still a proper name and the countability
is a property of the implicit referent.
Analyses in terms of enrichment are harder to maintain when the new property is
incompatible with the old. (35) is a case in point.
(35) some pillow (count > mass)
The properties count and mass are mutually exclusive; hence, a change from count to mass
suggests an override. However, as defended in Jackendoff 1991, it is possible to assume an

operator or rule that changes the perspective on the item rather than its properties: the non-
countability of pillow in (35) comes from ‘zooming in’ on the material until the boundaries of

the item are “outside the current field of view” (Jackendoff 1991: 19). He stresses that “[t]his
does not entail that the entity is absolutely unbounded in space or time; it is just that we can’t
see the boundaries from the present vantage point”. Thus, the boundedness of the pillow in
(35) is not overridden, but merely ignored for the purpose of the utterance.
However, arguments for enrichment are harder to make when the coercion involves a formal
category change, as the conversion from noun to adjective in (36), repeated from (20). The
formal proof of the conversion (and from a production perspective, its trigger) is the
superlative suffix, which requires an adjectival base.
(36) Notizia bombissima! Priest Holmes si ritira?
‘Breaking news! Is Priest Holmes withdrawing?’

15

For type coercions involving category changes and concomitant morphological behavior, an
analysis in terms of override seems the better choice.
However, there can also be semantic reasons that speak for override. Consider again (19),
repeated as (37).
(37) This is so 2013.
Both on the form side and the semantic side, 2013 has acquired new properties. Formally, it
functions as an adjective, since it combines with so. Semantically, something that is so 2013
is not necessarily happening in 2013, in fact, it is most likely to be 2014 or later. Both facts
could be accommodated in an enrichment analysis by saying that so 2013 means ‘typical of

2013’. However, this is not actually true: the idiomatic meaning of the expression is ‘old-
fashioned, dated’. No clear conceptualization of what is typical of 2013 is required. In fact, it

may not even matter which year is inserted in the construction, as long as it is in the past. The
idiomatic meaning is taken care of by the entire construction, at the expense of the lexical
semantics of the inserted item. In this light, an analysis in terms of override makes the best
sense.
A similar point can be made for (38), introduced as (13) above.
(38) a. This building looks very American.
b. You’re renting your home? That’s really un-Dutch.
The utterances contain a semantic shift: a very American building does not have to be
American or in America, a person behaving in an un-Dutch way does not need to be Dutch. In
fact, un-Dutch cannot be used to mean ‘not Dutch’; witness (39) – the appropriate alternative
would be non-Dutch.
(39) ?”flabbergasted” is an un-Dutch word
In (37) and (38), we see intermediate cases between enrichment and override. On the one
hand, the temporal reference to 2013 and the geographical reference of American and Dutch
are ‘wrapped’ by a function roughly meaning ‘characteristic of’. On the other hand, the
lexical meaning of the words gets jeopardized, and in (38) there is a type shift from relational
to qualifying adjective, which appears to be a consequence of override.
The clearest cases of override, however, are idiomatic constructions that were introduced as
RADICAL COERCIONS in (6) and (21) above. English provides a further example, the
construction V out, meaning ‘go into an unusual mental state’ (Jackendoff 2002). There are
various verbs that appear in this construction; Jackendoff lists the following:
(40) X out

16

X: pass, black, conk, fink, crap, chicken, flake, zonk, zone, bliss, flip, space, phase,
crump, veg, chill, knock, bum, fake, gross, weird, creep, burn, poop, tucker, freak,
wig, stress, mellow
While some of the lexical meanings may contribute to the expression, e.g. in bliss out, creep
out or veg out (‘act like a vegetable’), in many cases the meaning is idiomatic and quite
independent of the inserted items. Here, the constructional semantics is clearly overriding the
meaning of the words. In fact, some of the words may not have any meaning to begin with, as
far as they exist independently at all (zonk? crump? tucker?). Interestingly, the Dutch and
German idioms in (6) and (21) also contain non-existing words. German has auf den Senker
gehen, undoubtedly a variant of auf den Senkel gehen. While neither word makes any sense
compositionally, Senker is a specialist term unlikely to be part of the active vocabulary of
most German speakers. Dutch provides even more striking examples. In the expression REFL
DET X schrikken, we find a variety of items that do not occur outside this and other idioms
and that do not have an independent meaning.

8 Examples are:

(41) Ik schrik me ‘I’m scared to …’

  • de rambam
  • de/het schompes/ schompus
  • het lepalazerus/ leplazarus
  • het apelazerus/ apelazerus
  • het apezuur
  • het habbiebabbie
    The use of such forms indicates that the NP is a mere filler, a meaningless, even if expressive,
    sound sequence (though it still has the syntactic structure of a well-formed NP, and some of
    the fillers may have internal morphological structure). The semantics of the expression is
    entirely constructional and idiomatic. To the extent that the filler items have meaning of their
    own, this meaning – in most cases – does not contribute to the utterance. Therefore, such
    idioms are the clearest examples of constructional meaning overriding lexical meaning. They
    also constitute the most radical instances of coercion: the lexical meaning is ‘coerced away’
    by the constructional context.
    7 Conclusions
    In the above, we discuss a broad variety of coercion phenomena. Hereby, we extend the
    notion of coercion both at the lower and at the upper end. At the lower end, we include cases
    of contextual adjustment. Such instances are argued to be a matter of SELECTION of certain
    aspects of lexical meaning by the context. At the upper end, we discuss idiomatic
    constructions coercing inserted items so radically that their lexical semantics is overridden by
    8 Note also the variable gender of some of the items, reflecting the embarrassment produced
    by a construction requiring a determiner for what is essentially a non-word.

17

the utterance meaning. Between SELECTION and OVERRIDE, we argue for a middle ground
covered by cases of ENRICHMENT, in which the lexical and the phrasal semantics negotiate the
utterance meaning cooperatively. While selection, enrichment and override are usually
presented as separate and even contradictory mechanisms, we argue for a conciliatory
approach in which the three form a continuum. The unifying factor is taken to be the degree
of top-down influence of complex syntactic or morphological structures on lexical meaning.
In this view, selection represents the mildest degree of influence, enrichment an intermediate
level and override the strongest force. By presenting a typology that is compatible with a
variety of formalisms, from Pustejovsky’s Generative Lexicon to Jackendoff’s Parallel
Architecture and various flavors of Construction Grammar, we hope to contribute towards a
discussion that has been sorely fragmented by theoretic differences.
References
Bauer, Laurie, Rochelle Lieber & Ingo Plag 2013. The Oxford Reference Guide to English
morphology. Oxford: Oxford University Press.
Booij, Geert & Jenny Audring. 2007. Uitgezwaaid en aangezwaaid. Participiumconstructies in
het Nederlands. Nederlandse Taalkunde 12, 52-62.
Booij, Geert & Jenny Audring, forthcoming. Coercion and category change in Construction
Morphology.
Briscoe, Ted, Ann Copestake & Bran Boguraev. 1990. Enjoy the paper: Lexical semantics via
lexicology. Proceedings of the 13th international conference on computational linguistics,
Helsinki, Finland, 42–47.
Cappelle, Bert. 2014. Conventional combinations in pockets of productivity: English
resultatives and Dutch ditransitives expressing excess. In Ronny Boogaart, Timothy Colleman
& Gijsbert Rutten (eds.), Extending the scope of Construction Grammar, 251-281.
Amsterdam / Philadelphia: John Benjamins.
Culicover, Peter & Ray Jackendoff. 2005. Simpler syntax. Oxford: Oxford University Press.
De Swart, Henriëtte. 1998. Aspect shift and coercion. Natural Language & Linguistic Theory
16(2), 347-385.
Fauconnier, Gilles. 1985. Mental spaces: Aspects of meaning construction in natural
language. Cambridge: Cambridge University Press.
Fillmore, Charles & Paul Kay.1993. Construction grammar coursebook. Unpublished ms.,
Department of Linguistics, University of California, Berkeley.
Grandi, Nicola, Malvina Nissim & Fabio Tamburini. 2011. Noun-clad adjectives. On the

18

adjectival status of non-head constituents of Italian attributive compounds. Lingue e
Linguaggio 10, 161-76.
Harder, Peter. 2010. Meaning in mind and society. A functional contribution to the social turn
in cognitive sociolinguistics. Berlin & New York: Walter de Gruyter.
Jackendoff, Ray. 1991. Parts and boundaries. Cognition 41(1), 9-45.
Jackendoff, Ray. 1997. The architecture of the language faculty. Cambridge, MA: MIT Press.
Jackendoff, Ray. 2002. Foundations of language: Brain, meaning, grammar, evolution.
Oxford: Oxford University Press.
Jackendoff, Ray. 2013. Constructions in the Parallel Architecture. In Thomas Hoffman, &
Graeme Trousdale (eds.), The Oxford Handbook of Construction Grammar, 70-92. Oxford:
Oxford University Press.
Kuperberg, Gina, Arim Choi, Neil Cohn, Martin Paczynski & Ray Jackendoff. 2010.
Electrophysiological correlates of complement coercion. Journal of Cognitive Neuroscience
22(12), 2685–2701.
Lapata, Mirella, Frank Keller & Christoph Scheepers, C. 2003. Intra-sentential context effects
on the interpretation of logical metonymy. Cognitive Science 27, 649–668.
Lapata, Mirella & Alex Lascarides. 2003. A probabilistic account of logical metonymy.
Computational Linguistics 29, 263-317.
Lauwers, Peter & Dominique Willems. 2011. Coercion: Definition and challenges, current
approaches, and new trends. Linguistics 49 (6), 1219-1235.
Lauwers, Peter. 2014. Between adjective and noun. Category/function mismatch,
constructional overrides and coercion. In Rafaele Simone & Francesca Masini (eds.) Word
classes: Nature, typology and representations, 203-226. Amsterdam/ Philadelphia: John
Benjamins,.
Michaelis, Laura A. 2003. Word meaning, sentence meaning and constructional meaning. In
Hubert Cuyckens, René Dirven & John Taylor (eds.), Cognitive perspectives on lexical
semantics, 163-210. Amsterdam: Mouton de Gruyter.
Michaelis, Laura A. 2004. Type shifting in construction grammar: An integrated approach to
aspectual coercion. Cognitive Linguistics 15-1, 1-67.
Michaelis, Laura A. 2011. Stative by construction. Linguistics 49 (6), 1359-1399.

19

Nunberg, Geoffrey. 1979. The non-uniqueness of semantic solutions: Polysemy. Linguistics
and Philosophy 3, 143-148.
Partee, Barbara & Mats Rooth. 1983. Generalized conjunction and type ambiguity. In Rainer
Bäuerle, Christoph Schwarze & Arnim von Stechow(eds.) Meaning, use and interpretation of
language, 361-383. Berlin: Walter de Gruyter.
Pustejovsky, James. 1991. The generative lexicon. Computational Linguistics 17(4), 409-441.
Pustejovsky, James. 1995. The generative lexicon. Cambridge, MA: MIT Press.
Pustejovsky, James. 2011. Coercion in a general theory of argument selection. Linguistics 49
(6), 1401-1431.
Pustejovsky, James & Elisabetta Jezek. 2008. Semantic coercion in language: Beyond
distributional analysis. Italian Journal of Linguistics 20(1), 181-214.
Suttle, Laura & Adele Goldberg. 2011 The partial productivity of constructions as induction.
Linguistics 49 (6), 1237-1269.
Talmy, Leonard. 1978. The Relation of grammar to cognition. In: Toward a cognitive
semantics, Vol. I: Concept structuring systems. Cambridge, MA: MIT Press.
Verkuyl, Henk. 1972. On the compositional nature of the aspects. Dordrecht: Reidel.
Willems, Klaas. 2013. The linguistic sign at the lexicon-syntax interface: assumptions and
implications of the generative lexicon theory. Semiotica 193, 233-287.

X