Pernilla Hallonsten Halling

Properties deadjectival
nouns
unmarked adjectives
unmarked adverbs
predicate adjectives, copulas

Actions action nominals, participles, converbs complements, relative clauses
infinitives, gerunds
unmarked verbs

In table 3.1, the modification column has been split in two, in order to include both modification within referring expressions and modification within predicating expressions¹ (‘within reference’ and ‘within predication’ for short, cf. discussion of unclear terminology in section 2.3.4). The added column suggests that property words used for modification within predicating expressions takes the form of unmarked adverbs. One such example is English fast, which does not take the -ly ending, unlike most other English property- denoting Adverbs. Object words in the same modifier function are expected to be found as prepositional phrases on verbs, e.g. speak like a child. Finally, action words used as modifiers within predicating expressions can be expected to be found in the form of converbs, e.g. run laughing. When the model is expanded in this way, modification emerges as a two-sided function, applying to reference as well as predication. This includes adverbs in a natural way in the model. In the discussion of Croft’s approach in section 2.3.4, it became clear that modification is a secondary discourse function, in the sense that it cannot occur without the reference or predication which it helps to establish (cf. Croft 2003: 184–185). The expanded table in 2.2 elucidates this characteristic, by indicating that modification is bound to either reference or predication, and that two columns are required in order to account for modification as a whole.
On the level of discourse, modification can be defined as a function that a semantic item takes in use. But such a definition does not spell out what happens syntactically and semantically when modification is used within an expression. In order to do so, I propose the following three-way definition of modification:²

¹ For a further elaboration where also the row for properties has been expanded, see tables 9.3 and 9.4 in chapter 9.

² I would like to thank Östen Dahl for suggesting the basis of the definition in (3.1) to me.

(3.1) Modification takes one expression as its input and yields another expression as its output.

1. 
   Syntactically, the output expression has the same properties as the input expression.
   
2. 
   Semantically, the meaning of the output expression is of the same kind as the meaning of the input expression.
   
3. 
   In discourse, modification is a function that a semantic item takes in use.
   

The definition proposed in (3.1) says that when a modifier is added to an expression, a new expression is formed. The syntactic component in (3.1a) states that the properties of the new expression are the same as the properties of the old one. The semantic component in (3.1b) allows for some change of meaning between the input and output expressions, as long as the meaning remains of the same kind. The discourse component in (3.1c) follows from the expansion of Croft’s model in table 3.1. Let us now apply the syntactic and semantic components to some examples of modifiers within referring and predicating expressions, respectively.

(3.2) (a) the rose

(b) the pink rose

(3.3) (a) Sheila ran.

(b) Sheila ran fast.

Examples (3.3–3.2) have the same syntactic properties after the modifiers pink and fast are added as they have before this addition. In (3.2b), the pink rose is still a noun phrase (cf. the rose), and in (3.3b) ran fast is still a verb phrase (cf. ran). Semantically, the meaning of the modified expression in (3.2b) still belongs to the same domain of roses, which potentially can have a range of different colors. In (3.3b), the modified expression still belongs to the same domain of ways of running, which can be done at different paces. The fact that the modified expression has the same syntactic properties as the expression that it takes as its input implies that the modified expression must also be able to serve as input for further modification. For the purpose of the present discussion, this further modification is termed nested modification. For modifiers within referring expressions, nested modification is unproblematic.

(3.4) (a) the pink rose

(b) the pretty pink rose

In (3.4b), the output is, again, a noun phrase. According to the definition of modification in (3.1), the meaning of this expression is that the rose is pink, and in virtue of this color, it is pretty. Of course, (3.4b) could as such also mean ‘the rose that is pretty and pink’, but then the input expression would simply be the rose, with two modifiers added at the same time. Modifying predicating expressions that already have a modifier is slightly more complex.

(3.5) (a) Sheila ran fast.

(b) Sheila ran fast well.


The sentence in (3.5b) does not come across as the most natural example. But it is neither ungrammatical, nor semantically implausible. The output ran fast well is still a verb phrase, and semantically, the meaning is that Sheila runs fast in a good or skilled way. A potential context where this could be uttered is a situation where there are different runners that are all fast, though some are skilled in running fast and some are not. The expression Sheila runs fast well may then be interpreted as Sheila keeping a steady fast pace and having a good technique of running fast. Other runners, even though they run fast, are perhaps not able to keep up the pace, or maybe they run fast in a way that exhausts them. Accordingly, although (3.5b) is not the most natural example, it can be used in the right context, and illustrates that the definition in (3.1) holds also for the modification of already modified predicating expressions. Moreover, in order to get a coordinated meaning of (3.5b), and must be inserted as in Sheila ran fast and well. Here, the examples of modifiers in predicating expressions and modifiers in referring expressions differ, since both the nested interpretation and the coordinated interpretation are available in the pretty pink rose. There are also examples of modifiers of modified predicating expressions that appear more natural, such as Sheila runs fast in a funny way, where Sheila’s manner of running fast is funny, although her manner of running slow is not expected to be so. Notably, it seems that such expressions are more natural when a multi-word adverbial is added than when a single adverb is used, perhaps because a longer expression can make the modification more explicit. Also, it seems that this type of nested modification is both much more common and easier to interpret in referring expressions than in predicating ones.
So far the three-way definition of modification clearly holds. In what follows, I will discuss other definitions of modification and evaluate them against my own. These defini- tions pertain primarily to syntax and semantics. Traditionally, modification is commonly treated as a syntactic phenomenon, and the key concept to define it is endocentricity, a term introduced by Bloomfield (1935: 194). Lyons (1968: 231–232) states that “[a]n endocentric construction is one whose distribution is identical with that of one or more of its constituents”. Endocentric constructions are argued to be either co-ordinating, where each constituent has the same distribution as the whole construction, or subordinating, where one of the constituents has the same distribution as the whole. This constituent is labeled head, whereas the other is labeled modifier (1968: 233). Endocentricity is a central term also for compounds, which can be either endocentric, e.g. blackbird, where one part of the compound is the head (bird ), or exocentric, e.g. pickpocket where “the true head” is not expressed (Bauer 2006: 723–724). Another fundamental term in the syntactic context of modification is adjunct. According to Lyons (1968: 344), “[a]n adjunct is by definition a ‘modifier’ attached to a ‘head’, upon which it is dependent and from which it can be ‘detached’ without any consequent syntactic change in the sentence”. Adjuncts, which can be syntactically freely added, are often defined as opposed to arguments, as el- ements required by the predicate (e.g. Kroeger 2004: 7, cf. also the discussion of Role and Reference Grammar in section 2.3.1). This points further to the non-obligatory syntactic character of modification. In the various versions of generative grammar, modification is


defined as a relation in the syntactic tree, by which a phrase that modifies a head must also be a sister to the head in question (Carnie 2011: 70). From the perspective of X-bar theory, Haegeman (2006: 691) states that the head of a phrase remains the head after a modifier has been added, and this is what makes the whole constituent endocentric. The definition of modification presented in (3.1) incorporates the essence of endocentricity: the input expression has the same syntactic properties as the output expression.
Semantic definitions of modification are often less explicit than the syntactic ones. In a recent study of modifiers in Romance languages, Valera & Hummel (2017: 1) state that “[m]odification is a linguistic function that refers to a semantic change operated on a primary unit, e.g. a word or a sentence”. Here, it seems that modification is defined solely in terms of semantics. However, the content of the notion ‘semantic change’ is not further discussed. Formal semantic accounts also use the term modifier, but tend to redefine it as an elaborated type of predicate. This is illustrated in Geuder (2000: 1–2, discussed in section 2.3.2) where adverbs are first labeled modifiers, though Geuder later turns to the term predicates of events in order to describe the semantics of manner adverbs. Ramat & Ricca (1994: 290), in their definition of adverbs discussed in section 2.4, state that modifiers “add information”. Irrespective of the semantic theory, it is generally acknowledged that modification implies a semantic change or addition to the expression in which it occurs, although the particulars of this change are usually not explained. The semantic component in (3.1b) makes the semantic change more precise by stating that after the change, the meaning is still of the same kind as before the change. Although this is not as explicit as the syntactic component, it does describe what happens semantically. The proposed definition also holds for a number of instances of modification that are well known for their problematic semantics, such as alleged murderer and fake news (cf.
e.g. Geuder 2000: 6). An alleged murderer may or may not be an actual murderer, and fake news is certainly not real news. Accordingly, the modifier and the modified clash semantically. To account for this, Montague Grammar proposes a number of classes of adjectives, such as intersective, nonsubsective and privative, and plain nonsubsective classes, which relate semantic values in different ways (Partee 2007: 151ff.). Adverbs are dealt with in a very similar fashion (Montague 1974: 213). But alleged and fake can also be accurately defined as modifiers if the meaning of the expression is of the same kind after the modifier has been added, as stated in the definition in (3.1). On a lexical level, the modifier clashes with the modified in these examples, with the result that part of the meaning is weakened. However, not the entire semantic representation disappears: at least part of the concept of murderer or news remains the same. Attributes such as alleged or fake may nonetheless modify the distinction between, e.g., real and suspected or potential candidates. In conclusion, examples such as alleged murderer and fake news are still instances of modification, although they are non-typical instances. Other types of semantically non-typical modification can also be found in the case of epithetic adjectives,
e.g. My nice daughter gave me a present (in the sense where it does not single out a nice daughter amoung a group of daughters, but simply adds the information that it was nice of the daughter to give a present) as well as pleonastic adjectives, e.g. receive a free gift (which is pleonastic because gifts as such are free).
When an expression is altered through linguistic modification, it implies a kind of modification of the concept to which the expression refers. From a cognitive perspective,


several models have been proposed to account for such conceptual modification. Barsalou (2014) discusses accounts of what he terms conceptual combination: how new concepts are constructed from other concepts that are already stored in memory (2014: 168–169). One explanatory model is called intersection, and applies to examples such as pet fish (which is a compound, cf. discussion on endocentricity and compounds above). Pet fish refers to everything that is a pet as well as a fish, i.e. the intersection of the two (2014: 169). Intersection nonetheless implies equal membership in the intersecting categories, but does not hold in all cases. Individual examples vary in terms of how strong a member they are of a specific category (cf. fuzzy set theory following Zadeh 1965, 1996, cited in Barsalou 2014: 169). For instance, guppy is not the most typical member of the category fish, if compared to e.g. trout. Nor is it the most typical member of the category pet. Still, guppy is perceived as a strong pet fish candidate. Intersection, then, is not able to account for examples such as pet fish. Barsalou proposes that instead of operating on sets, concepts are manipulated in the form of selective modification of what is termed frames. A frame (also schema/schemata) is made up of a combination of attributes: the frame for, e.g., car has attributes such as driver, fuel tank, engine, transmission, and wheels (2014: 158). Selective modification of frames draws on implicit information. This is illustrated with the example orange dog, where orange is unexpected. The color attribute of the frame for dog is then argued to be “selectively modified”, meaning that the default color (maybe brown) is replaced with orange (2014: 169). In this example, the color attribute also receives more weight than normally, making color more important in the conceptualization of orange dog (Smith et al. 1988, cited in Barsalou 2014: 169). Another example shows that attributes may constrain each other implicitly: in wooden spoon, the modifier could be expected to only tell us what the spoon is made of (Smith & Osherson 1989, cited in Barsalou 2014: 169). However, as illustrated by Medin & Shoben (1988, cited in Barsalou 2014: 169), wooden in the case of wooden spoon also tells us something about size, exchanging the expected attribute small for large. Barsalou concludes that since “correlations between attribute values pervade human knowledge, the explicit modification of one attribute often produces implicit modification of correlated attributes” (2014: 169–170).
Barsalou (2014: 236) argues that within sentence processing, certain words activate frames, whereas others modify the attributes of frames. The same type of modification is established for adjectives and adverbs, the difference being the frame whose attributes they modify: while adjectives modify the attributes of noun frames, adverbs modify the attributes of verb frames. As in the case of selective modification, information is inferred that is not explicitly stated. One adjective that illustrates this is good, which is instantiated in different ways. A good chair can be good in different ways, such as for relaxing, or for standing on to change a light bulb – these interpretations do not instantiate the same type of chair. Here, we may recall Dixon’s 1982 [1977] discussion of value adjectives from section 2.5.1, pointing to the need of other concepts in order to interpret examples such as good. Likewise, adverbs can be instantiated differently, as illustrated in (3.6) (Barsalou 2014: 241).

(3.6) (a) The tortoise traveled quickly.

(b) The hare traveled quickly.

The tortoise and the hare in examples (3.6a-b) both traveled quickly, but the interpre- tation is still that they traveled at different paces. The tortoise traveled quickly for a tortoise, which our knowledge tells us is actually not that high a speed. Conversely, we know that hares are fast, allowing us to interpret the hare as traveling at high speed in (3.6b).

Barsalou’s conceptual account of modification highlights the fact that modification is complex and dependent on various factors external to the linguistic expression as such. This is in accordance with the definition proposed in (3.1). Although the semantic change achieved by the modifier is complex, the meaning of the expression is still of the same kind as the meaning of the original expression without the modifier. Thus, a good chair is still a chair, and can be used for any of a number of purposes by which it can also be described as good. Also, the traveling performed by an agent (such as a tortoise or a hare) following the characteristics (including typical speed of movement) of this agent remains a way of traveling, regardless of what pace it is performed at.
In conclusion, it is clear that although modification has traditionally been defined syn- tactically (primarily based on endocentricity), semantics is very relevant here. The defi- nition of modification proposed here (see 3.1) elucidates both the syntax and semantics of modification. In terms of discourse, modification is a secondary function, building on an expansion of Croft’s account of parts of speech (see table 3.1). Accordingly, this definition does not go against the definitions found in the literature. Rather, it is the combination of different components that is in focus. In the next section, I will turn to how modification is related to predication.

3.