Pernilla Hallonsten Halling

Parts of speech in general

the presence or absence of structural coding, which is the morphosyntactic marking that indicates the function that a lexical item is used in.

Table 9.1. Encoding of semantic classes and discourse functions (Croft 2001: 88)

Reference Modification Predication

Objects

unmarked nouns
genitive, adjectivilizations, predicate PPs on nouns nominals
copulas

Properties deadjectival
nouns
unmarked adjectives
predicate adjectives, copulas

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

Parts of speech as defined by Croft are not to be confused with language-specific categories. Rather, parts of speech are cross-linguistic prototypes. There is a universal tendency captured in Croft’s structural coding criterion, which holds that a typologically marked item is always encoded by at least as many morphemes as a typologically unmarked item (2001: 90). This was illustrated in detail in section 2.3.4, with the English example red. In its prototypical function as a modifier in, e.g., the red rose, red is unmarked. When used in reference, a derivational morpheme must be added to red, as in, e.g., the red-ness of the rose. Some English object words used for modification have zero structural coding, such as kitchen table or apple basket. This shows that the object words kitchen and apple are encoded by the same amount of structural coding, namely zero, when used in their prototypical function of reference and the non-prototypical function of modification (cf. Croft 2001: 99). Another universal pattern proposed by Croft is that of behavioral potential. Behavioral potential is basically equivalent to the inflection that goes with a specific function. Croft’s criterion says that behavioral potential that is found on marked items (i.e. less prototypical combinations of semantic class and function) are always found on unmarked items (i.e. more prototypical combinations) (2001: 91). In English, this can be illustrated by how tense is found in relative clauses that are finite (action used in modification), just as tense is found in the prototypical combination of action and predication (although note that English is not a very good example in terms of behavioral potential as pointed out by Croft 2003: 188).
While the treatment of parts of speech within Functional Grammar and Croft’s universal tendencies do not clash, the two approaches differ in important ways. Within Functional Grammar, unmarked patterns across functions are seen as constituting flexible parts of speech. For example, in Tongan (Austronesian), there are lexical items that can be used in several functions (see also the in-depth discussion of Tongan by Broschart 1997). The

Tongan lexical item si’i can be used in predication to mean ‘be small’; in reference to mean ‘childhood’; as a modifier within a referring expression to mean ‘little child’; and as a modifier within a predicating expression as, e.g., ‘He studied a little’ (Hengeveld 1992: 66). Thus, Functional Grammar treats Tongan as a language with one flexible part of speech which subsumes the functions of verb, noun, adjective, and adverb. Other languages may have one class of verbs and one flexible class subsuming nouns, adjectives, and adverbs (e.g. Imbabura Quechuan) or different classes of nouns and verbs, in addition to one flexible class subsuming adjectives and adverbs (e.g. Dutch), i.e. what I define as general modifiers (cf. section 6.4.4). Languages that do not have flexible parts of speech are defined as having rigid parts of speech systems, and those among them that have all four major parts of speech are called specialized (e.g. English, Hengeveld 1992: 69).

As pointed out by Croft (2001: 65–75), the flexible part of speech approach appears to overlook one important fact, namely semantic shift. Semantic shift, in this context, refers to meaning shift rather than diachronic shift. It is the meaning shift that takes place when a lexeme is zero-converted to another function than the one with which it is usually associated. Semantic shift is similarly important in the discussion of zero conversion by
e.g. Evans & Osada (2005). The Functional Grammar account allows certain lexemes to be used flexibly in more than one function, but does not explain why and how these items sometimes shift meaning and sometimes do not. Meaning shift over functions is unpredictable, as can be observed from the use of the Tongan si’i. Some consistency can nonetheless be discerned: Croft identifies another universal tendency here, namely that when a lexeme shifts meaning with zero coding, the direction of meaning change goes toward the semantic class that is prototypically associated with the new function (2001: 73). For instance, when a property word is used in reference, the meaning shifts to an object-like meaning. On the other hand, the motivations for allowing semantic shift within the same flexible category differ. Hengeveld & Rijkhoff (2005) argue that flexible lexemes are monosemous, but semantically vague. When flexible lexemes are used in different contexts, they argue that subfields of their general meaning are evoked, explaining the perceived meaning shifts (Rijkhoff & van Lier 2013: 22–23). Don & van Lier (2013) argue that lexemes in flexible word classes have fixed meanings, but that use in specific morphosyntactic contexts leads to shift of meaning. The phenomenon of semantic shift is thus discussed by the proponents of Functional Grammar, but it seems that too little significance is attributed to its consequences. As argued in Croft’s criticism, when the same criteria are applied consistently, English could be regarded as a language with flexible word classes based on examples such as The school was small and We schooled him into proper manners, where flexibility between Nouns and Verbs appears to occur (2001: 69). Since Hengeveld (1992) does not treat English as a flexible language but a specialized one, it seems difficult to apply the criteria of flexibility consistently.
The approach to parts of speech following Croft (1991, 2001, 2003) does not have as its object to divide all items of an individual language into different categories. Rather, it aims at formulating a typologically applicable account of parts of speech, which is consistent in the criteria used in cross-linguistic comparison. Prototypical parts of speech are regarded as categories that can be compared across languages, and should not be confused with language-specific categories. Analyzing parts of speech as prototypical categories further offers a way in which both central and peripheral members can be

integrated. What remains is to determine whether adverbs can be appropriately situated among the other prototypical parts of speech, or what the role of adverbs is otherwise to be.

3.