The origin of prototype effects

An idea that has been extremely influential in the study of grammatical categories and relations over the past three decades is that the various instances of a category (e.g., noun, verb, or adjective) or relation (e.g. subject or direct object) do not all have equal status in a speaker's mental representation of the grammar of their language. Rather, as Hopper & Thompson (1984: 707) put it, speakers discriminate between central, or prototypical, and peripheral, or nonprototypical instances, and this will be reflected in the structural properties of the various instances. This idea has influenced both studies of individual categories and relations carried out within the functional-typological approach, and cognitive theories of grammar in general. For instance, Cognitive Grammar (see, e.g., Langacker 1991 and 1999) assumes that a description of a speaker's mental representation of their language should include some specification of the degree of prototypicality of the various linguistic instances of individual grammatical categories, such as for example noun and verb.

The idea that grammatical categories and relations have a prototype structure makes it possible to account for the fact that, while sharing a number of similarities, the putative instances of individual categories and relations often exhibit several nonoverlapping properties. For example, particular types of lexical roots ('nouns', 'verbs', or 'adjectives') may not always exhibit the same categorial distinctions or combine with the same morphology, both cross-linguistically and across different contexts within individual languages. Also, from one language to another, semantically similar lexical roots may not occur in the same range of syntactic environments. These differences can be accounted for by assuming that they reflect the fact that the relevant constructions represent prototypical vs. nonprototypical instances of the same category or relation.

This view has received several types of criticism over the years (for a critical discussion of applications of the prototype approach to semantic categories, see Lakoff (1987)). For example, Dryer (1997) has criticized the idea that the mental grammar of the speakers of different languages may include prototypical vs. nonprototypical instances of the same grammatical relation, such as prototypical or nonprototypical subjects. This, Dryer argues, implies that speakers have some knowledge both of the grammatical relations of their language and whether or not these relations conform to some cross-linguistically valid prototype, and is unclear how the latter type of knowledge could be acquired.

A more detailed critique has been put forward by Newmeyer (1998: chap. 4). Prototype effects, that is, the phenomena that are usually taken as evidence for the prototype structure of individual grammatical categories, Newmeyer argues, can naturally be accounted for in terms of properties other than prototypicality in itself, for example the fact that the various instances of a grammatical category differ with regard to frequency or markedness, or the fact that certain semantic parameters are incompatible with the semantics of particular instances, and are therefore not encoded for those instances (for example, abstract nouns are incompatible with individuation, so this parameter will not specified grammatically for these nouns). This, Newmeyer concludes, makes it unnecessary to account for the relevant phenomena by postulating that grammatical categories have a prototype structure with 'best-case' members and members that depart from the 'best-case'.

While the issue of the cross-linguistic validity of prototypes will not be addressed here, various data will now be presented suggesting that, diachronically, prototype effects may not originate from differences in the prototypicality status of the various instances of a category. This is in line with the general logic of Newmeyer's (Newmeyer (1998)) arguments. However, the factors that Newmeyer argues to give rise to prototype effects are actually part of most definitions of prototypicality. Proponents of the prototype approach generally agree that the prototypical instances of a category or relation are unmarked or more frequent ones, ones that speakers perceive as more natural or salient, and ones that display specific semantic or pragmatic properties (see, e.g., Hopper & Thompson (1984), Langacker (1999), Croft (2003)). In this respect, both Newmeyer's analysis and prototype analyses imply that prototype effects originate from some perceived asymmetry in the frequency or the conceptual properties of the various instances of a category or relation, and whether or not this asymmetry should be described in terms of prototypicality is essentially a terminological issue. In what follows, it will be shown that various types of prototype effects do not actually originate from any type of asymmetry between the relevant elements, so they provide no evidence that these elements have a different status in a speaker's mental representation.

A well-known application of the prototype approach is a model of parts of speech that has been proposed in various versions by Hopper and Thompson (1984), (1985), Langacker (1987), (1991), and Croft (1991) (2001). The basic idea of this model is that different types of lexical roots will have prototypical and non-prototypical pragmatic functions. For example, Hopper and Thompson (1984), (1985) argue that nouns are prototypically used to denote discourse-manipulable participants and verbs are prototypically used to denote discrete discourse events. Likewise, Croft (1991), (2001) distinguishes between three functions, the referential function (the identification of a referent), the predicative function (the predication of something about a referent), and the modifying function (the attribution of a property to a referent). These are, respectively, the prototypical functions for nouns, verbs, and adjectives (or, in Croft's terms, lexical roots denoting object, actions, and properties.

Individual lexical roots may be used both in prototypical and in nonprototypical functions. When a lexical root is used in a nonprototypical function, however, this is reflected by two major types of grammatical phenomena, or prototype effects. First, since this use is unexpected, or less frequent, it is signaled by special morphology not used for the prototypical function. For example, in English, lexical roots denoting actions, such as play, can be used in modifying function, that is, to construe the action as a property attributed to a particular referent, as is the case in expressions such as those playing the game for the first time. They may also be used in referential function, that is, to denote (rather than predicate) an action, as is the case in expressions such as playing this game can eat up a lot of your time, or I don't like them playing in my yard. When used in these functions, however, these lexical roots bear a special morpheme, -ing, not used when they occur in predicative function, for example in sentences such as I don't play many games. Also, a lexical root used in a nonprototypical pragmatic function may fail to display the full range of categorial distinctions found when it is used in its prototypical function (for example, tense or person distinctions for verbs, or number distinctions for nouns). English -ing forms, for example, do not have the same inflectional potential as the corresponding forms used in predicative function[1].

Similar assumptions are also made in Langacker's (1987), (1991) model of nouns and verbs. Langacker argues that the two categories of noun and verb can be defined both in terms of cognitive prototypes and in terms of abstract cognitive schemas that are common to all of the members of the category. The cognitive prototype for nouns is the conception of a discrete physical object, while the cognitive prototype for verbs is the conception of participants interacting energetically. In terms of abstract cognitive schemas, nouns profile things, that is, regions in some domain, while verbs profile processes, that is, sets of states scanned sequentially through conceived time. Nouns and verbs may undergo shifts in profiling. For example, some types of nominalization shift the profile of a verb from a process to a region consisting of the component states of that process. Such shifts are indicated by the use of special morphemes, for example -ing, as shown by the contrast between verbs such as complain and walk and their nominalized forms complaining and walking.

When the possible diachronic origin of the relevant constructions is considered, however, a number of cases emerge suggesting that at least some prototype effects are unrelated to the fact that particular lexical roots are being used in a nonprototypical (unexpected, less frequent) function, or that speakers perceive an asymmetry between the various uses of these roots anyway.

In what follows, this will be illustrated by discussing some of the possible origins of the special morphemes that characterize a number of forms variously referred to in the literature as participles, gerunds, action nominalizations, infinitives, converbs, dependent verb forms, and the like (see on this point Koptjevskaja-Tamm(1993a) and Comrie & Thompson (2007), among others). These forms are used when lexical roots denoting actions ('verbs') occur in a number of functions that have been indicated as nonprototypical for these roots in the literature. This includes various types of modifying functions, as associated for example with relative clauses, cases where the actions being described are conceived as components of a larger macroevent (as is the case, for example, with serial verbs and similar constructions), and cases where the relevant lexical roots are used in referential function, that is, to denote a particular action rather than predicate that action of a referent, as is for example the case in complement clauses (clausal arguments of a main predicate). It is important to note in this connection that the various forms may be inflectionally reduced, as is the case with the forms in examples (15) and (16) below, or they may display the full range of inflectional distinctions effects are not necessarily correlated with the frequency, hence the prototypicality of the various uses. For this reason, his model does not represent an application of the prototype approach discussed in this section. available to the relevant lexical roots in the language, as is the case with the forms in examples (11b), (12b), and (14).[2]

The type of morphemes that will be taken into account are morphemes such as English -ing, which are not used when the relevant lexical roots occur in their assumed prototypical function, but are also not used when other types of lexical roots occur in their assumed prototypical functions. This is an important point, because in many languages, when lexical roots denoting actions are used in non-prototypical functions, they may take the same morphology as the lexical roots for which those functions are assumed to be prototypical. For example, when they are used in referential function, they may take case markers, just like lexical roots denoting objects ('nouns'). This is illustrated in (10), which shows that, in Krongo, the locative case marker can be used both on nouns and on infinitival complements of a main verb. These phenomena suggest, basically, that different types of lexical roots may be treated in the same way for grammatical purposes, but this does not imply that speakers perceive an asymmetry between different uses of particular types of roots. Therefore, these cases will be disregarded.

Krongo (Nilo-Saharan)

(10) n-dtdasd d?dyu?y k-daldndd?dy ki-niind 1/2-iMPFV-want I you LOC-iNF:teach I-LOC-language 'I want you to teach me Krongo' (Reh (1985): 337)

In some cases, when used in modifying or in referential function, lexical roots denoting actions display morphemes also used in coordinate clauses, though not in independent clauses taken in isolation. This is the case with the narrative prefix n-in Maasai, which can be used both in coordinate clauses, as in (11a) and in certain types of complement clauses, as in (11b). A similar example is provided by Krongo, where the connective prefix y-can be used in coordinate clauses (12a) and in relative clauses (12b).

Maasai (Nilo-Saharan)

(11) a. E-rrit-a nkishu n-e-rrip

3-herd-PAST cattle NARR-3-guard

'He herded the cattle and guarded them' (Tucker & Mpaayey (1955): 61) b. a-idim n-a-as

1-repeat NARR-1-dance

'I'll repeat dancing' (Heine & Claudi (1986): 120)

Krongo (Nilo-Saharan)

(12) a. i)-aan dttumdntdard adiyd

conn-cop jackal iNF.iMPFV.come

'and the jackal comes' (Reh (1985): 194)

b. n-6oni d?dy bilyaata ^-afdrd

1/2-iMPFV.know.TR I boy CONN-iMPFV.weep

'I know the boy who is weeping' (Reh (1985): 253)

For Maasai, Heine & Claudi (1986: 118-21) demonstrate that the narrative prefix n-originated from the affixation of a conjunction naa that is still used as such in the language, as is illustrated by (13) below. Similar processes have been reconstructed for other languages. For example, Mithun (2003)) shows that the Yupik subordinative suffix -llu, used to link actions or states that are part of a larger event or episode, originated from the affixation to the verb of a former enclitic =llu 'and, and also, too'. This is illustrated in example (14) where the first verb shows the original use of the suffix, and the second verb shows its use as a marker of the subordinative mood.


(13) Naa k-e-puonu and PTCL-3-come

And they will come' (Tucker & Mpaayey (1955): 104)

Yupik (Eskimo-Aleut)

(14) nange-nga-ata=Mu ellait uci-lir-lu-teng used.up-CONSEQ-3PL=and they. ERG cargo-make-SUBORD-3PL

And when they were finished, they packed up' (Mithun (2003): 556)

The use of the same prefix in coordinate clauses and complement or relative clauses is naturally accounted for by assuming that the latter originate from coordinate structures, e.g., for (11b) and (12b), 'I repeat and I dance', or 'I saw the boy, and he was weeping'. If the prefix is a former conjunction, however, this means that its use in complement and relative clauses cannot possibly reflect any perceived asymmetry between the use of the relevant lexical roots in these clause types and their use in other (prototypical) contexts. Rather, the prefix is there because it was part of the source construction, where it provided a specific contribution to the overall meaning of the construction by encoding the particular semantic relationship established between the linked events.

In some languages, when used in modifying or in referential function, lexical roots denoting actions take the same form used in so-called agentive or objective nominalizations. These are constructions used to denote an entity that performs the action described by the root, or is the object or result of this action

(Comrie & Thompson (2007)). Agentive and objective nominalizations are characterized by special morphemes, often referred to as nominalizers. An example of these morphemes is represented by the proto-Carib prefix * n-discussed by Gildea (1998: 128-33). In a number of Cariban languages, the reflexes of this morpheme are found in various types of objective and sometimes agentive nominalizations, illustrated in (15) for Apalai. These forms can also be used to modify a referent, as in the Bakairi examples in (16).

Apalai (Carib)

(15) o-nii-mero-'piri 2-O.NOMLZR-write-NOMLZR

'the thing you have written' (lit. 'your [past] writee [written thing]'

(Gildea (1998): 130, quoted from Koehn & Koehn (1991): 92)

Bakairi (Carib)

(16) a. dtd [maria y-igoke-ibe] ti/ike-ba

clothes Maria OBJ.NOMLZR-wash-PAST.NOMLZR dirty-NEG

'The clothes that Maria washed got clean' (Gildea (1998): 130)

b. ugdndo dgdw n-i-aki [maria y-e-tibe]

man snake 3/3-kill-TAM maria OBj.NOMLZR-see-PAST.NOMLZR

'The man killed the snake that Maria saw' (Gildea (1998): 130)

The origins of nominalizers are often obscure. In a number of cases, however, there is evidence that they originate from the grammaticalization of expressions such as 'thing', 'matter', 'one', 'person', and the like. For example, Munro (1976: 229) argues that the Mojave prefix 2c-, used in agentive nominalizations and illustrated in (17a), is related to the indefinite pronoun 2c, 'something' ((17b)). In Shina, the suffix -k is likely to have ultimately originated from the numeral ek, 'one'. In addition to being used in a variety of contexts related to the semantics of 'one', such as indefinite NPs ((18a)), this suffix is used in agentive and objective nominalizations, as in ((18b-d), as well as in cleft-constructions, as in ((18e)). In Classical and Lhasa Tibetan, a suffix -rgyu, found in objective nominalizations ((19a)), modifying expressions ((19b-c)), and referential expressions ((20)), is related to the noun rgyu 'matter, substance' (Beyer (1992): 296, DeLancey (2003): 276). In Qiang, the suffix -m found on relative clause verbs ((21)) is derived from a noun mi 'person' (LaPolla (2003): 584). Mojave (Hokan)


a. li-iyer


'bird' (Munro (1976): 229)

b. ?c isva:r

something sing

'sing something' (Munro (1976): 229)

Shina (Indo-European)

(18) a. ék müsha-k-ai

one man-one-GEN

'of one man, of a man' (Bayley (1924): 83)

b. Vatü-k came-NOMLZR

'the thing that came' (Bayley (1924): 82)

c. -k


'a striker' (Bayley (1924): 84)

d. mal málüs rai 'tü-k my father-by say-NOMLZR give 'give me the thing that my father said' (Bayley (1924): 83)

e. asá mi zá-s-o a-ií gyá-o-k mo that my brother-was-3M.SG come-CP went-3M.SG-NOMLZR I pató beé-t-o-s-os behind sit-PERF-M-AUX.PAST-1M.SG

'That was my brother who passed by, I had stayed behind'

(Schmidt & Kohistani (2008): 77)

Classical Tibetan (Sino-Tibetan)

(19) a. ña-la dgos-rgyu


'(something) for me to need' (Beyer (1992): 297)

b. bla-ma oü-rgyu lama come-NOMLZR

'the lama to come' (Beyer (1992): 296)

c. kha-tshems biag-rgyu last.testament leave-NOMLZR

'a last testament to leave' (Beyer (1992): 296)

Lhasa Tibetan (Sino-Tibetan)

(20) 'di'i skad=cha dris=rgyu gus=zhabs med-pa red

this-GEN question ask-NOMLZR polite

'It's not polite to ask about this' (DeLancey (2003): 284)

Qiang (Sino-Tibetan)

(21) qa-ha-Bdzp-m khud-le


'the dog which just bit me' (LaPolla (2003): 584)

In all of these cases, there is a relatively transparent connection between the original meaning of the nominalizer and the corresponding agentive or objective nominalizations, in that the latter can be assumed to have originated from patterns such as 'thing that Verbs/is Verbed, 'one/person who Verbs', and the like. This is also consistent with the use of these forms in modifying function, in that expressions of the type 'the snake that Maria saw', 'a last testament to leave', or 'the dog which bite me' may originate from patterns such as 'the snake, Maria's seen thing' 'a testament, a thing to leave', 'the dog, one who bite me', where the agentive or objective nominalization is used in apposition to a noun. This analysis has been proposed, for example, by Gildea (1998)) for Cariban languages, and by a number of scholars (DeLancey (1986), Noonan (1997), among others) for Sino-Tibetan languages.

Comparatively less attention has been devoted to the possible origin of the use of these expressions in referential function, as for example in (20), where the nominalization in -rgyu denotes the action described by the root rather than the entity performing this action or the object of this action. In this case too, however, it is in principle possible to postulate a connection with the original meaning of individual nominalizers. For example, when this meaning is 'thing', 'matter', or the like, as in (17), (19), and (20), it is plausible that the use of the nominalizer in expressions such as 'it is not polite to Verb' originates from patterns such as 'the thing/matter to be Verbed is not polite', or possibly 'the thing/matter of Verbing is not polite', and the like.

Such hypotheses, of course, have to be refined and verified for each language depending on the original semantics and distribution of the relevant nominalizers (see, for example, Yap & Wang (2011) for a recent discussion of the evolution of some agentive and objective nominalizers in Chinese). The general point, however, is that, to the extent that the use of the nominalizers in particular constructions can be related, directly or indirectly, to their original semantics, they cannot be argued to reflect the fact that the relevant lexical roots are being used in a particular, nonprototypical function. In constructions such as 'person who Verbs', 'thing that is Verbed', and the like, the elements meaning 'person', 'thing' etc. are there because they provide a specific contribution to the overall meaning of the construction, and they are maintained in the construction even when their original meaning is no longer transparent. In this sense, their use is a result of the rather general principle whereby complex meanings are conveyed by a combination of different elements, and there is no evidence that it reflects a distinction between the prototypical and nonprototypical functions of individual lexical roots that is somehow part of a speaker's mental representation.

The prototype approach originated from a wide body of studies carried out within cognitive psychology, which revealed significant asymmetries in a subject's judgements of category membership (see Lakoff (1987) and Taylor (1989) for a detailed history of the field, as well as Lakoff (1987): chap. 3 for a critical discussion of the implications of the relevant findings for hypotheses about mental representation). In most cases, however, prototype effects for grammatical categories have not been defined based on subjects' judgments. Rather, it has been assumed that the distribution of certain grammatical phenomena across different contexts (for example, the presence vs. absence of overt morphology) reflects an asymmetry between these contexts which is somehow relevant for speakers. The observed distributions, however, are a result of specific diachronic processes. The data discussed in this section show that these processes may be unrelated to any postulated asymmetry between different contexts, for example because they are standard grammaticalization processes based on the meaning relationship between a source and a target construction. In this respect, the relevant grammatical phenomena cannot be taken as evidence that the postulated asymmetry is part of a speaker's mental representation of individual grammatical categories.

  • [1] This model is similar to the model of parts of speech proposed in Hengeveld (1992). However, contrary to Hopper and Thompson (1984), (1985) and Croft (1991), (2001), Hengeveld does not relate the distribution of particular lexical classes to the semantics of their referents. Also, Hengeveld (1992: 59) identifies prototypicality with frequency, and assumes that, while the use of a lexical root in different pragmatic functions may have different grammatical effects, these
  • [2] In both of these cases, the relevant forms cannot usually occur in an independent declarative clause taken in isolation, and hence they count as instances of what Cristofaro (2003) defines as a deranked verb form (see Stassen (1985) for a first, though slightly different application of this notion).
< Prev   CONTENTS   Next >