1.1 Inflection vs derivation

1.1.3 Ambiguities in the supercategory of quantity

1.1.3.1 Evaluative morphology

This chapter opens with a quotation from Katamba’s (1993: 212) discussion of the sometimes inflectional and sometimes derivational nature of diminutives. Katamba notes that, in many African languages, augmentatives and diminutives are marked by affixes that are at the heart of the inflectional system. Thus, in Luganda the prefixes ka-/bu may function as unmarked inflectional singular/plural markers as in ka-solya ‘roof’ vs bu-solya ‘roofs’, but also as derivational diminutive prefixes as in mu-kazi ‘woman’ vs ka-kazi ‘little woman’ and ba-kazi ‘women’ vs bu-kazi ‘little women’. Similarly, bi- and ki- are unmarked inflectional singular/plural markers in ki-sero ‘basket’ vs bi-sero ‘baskets’, but derivational augmentative prefixes in ki-kazi ‘large woman’ and bi-kazi ‘large women’ (Katamba Reference Katamba1993: 211).

As a confirmation of Katamba, Anderson (Reference Anderson1992: 80ff.), in reference to diminutive affixes in Fula, a Niger-Congo language spoken in Africa, argues that, while diminutive formation is derivational in many languages, diminutives in Fula are integrated into the paradigmatic system as the marking of number, and that this paradigmatic system takes part in the syntactic relation of agreement, which is a major inflectional feature.

By contrast, Brown and Dryer (unpublished manuscript) give evidence that the category of diminutiveness in Walman, a language of the Torricelli family spoken in Australia-New Guinea, is inflectional. Similarly, Contini-Morava (pers. com.) shows that diminutives in Swahili are formed by replacing the noun class prefix with ki- (Class 7) and vi- (plural, Class 8), as in (17), and augmentatives are formed by replacing the noun class prefix with zero (Class 5) and ma- (plural, Class 6), as in (18):

(17)

(17)

Swahili kitoto < mtoto ‘small child, infant’ ‘child’

(18)

(18)

Swahili toto < mtoto ‘large, ungainly child’ ‘child’

Halfway between these two positions, Scalise (Reference Scalise1984) describes evaluative morphology as a third level of morphology different from both inflectional and derivational morphology. Based on examples from Italian, Scalise shows that evaluative suffixes:

1. (a) violate the Unitary Output Hypothesis,

2. (b) are transparent with regard to the lexical category (the category is always that of the base) and to the syntactic features,

3. (c) may be attached to more than one lexical category,

4. (d) change the semantics of the base,

5. (e) allow repeated application, and

6. (f) are always peripheral with regard to genuine derivational suffixes and internal with regard to inflectional morphemes.

However, Scalise’s claims are not universal. Thus, diminutive suffixes may change the lexical category of the base in Slovak: the masculine gender of medveď ‘bear’ changes to neuter in one of its diminutive forms, medvieďa, and the feminine gender of žena ‘woman’ changes to neuter in the diminutive žieňa. Similar cases abound. While the language-specific nature of Scalise’s account of evaluative morphology was demonstrated by Stump (Reference Stump1993), augmentatives and diminutives are intriguing for their relation to inflection and derivation (cf. Dressler Reference Dressler, Hickey and Puppel1997: 1403, 1406).

1.1.3.2 Aspect/Aktionsart

Aspect/Aktionsart is further evidence of the vague borderline between inflection and derivation, and has been discussed in the literature accordingly (Spencer Reference Spencer, Orgun and Sells2004: 1–2):

there are numerous cases in which we wish to say that a semantic predicate is added to a word without changing the word class, for example, Aktionsart markers. These cases are notoriously difficult to classify as inflection or derivation and I argue that they represent the final type of category mixing, asyntactic derivation.

The problem concerning the nature of aspect/Aktionsart has a terminological, a semantic and a formal side. Terminologically, there does not seem to be a unified use of the term aspect. In the English-written literature there is usually no distinction between aspect and Aktionsart. It is only recently that this distinction came to be drawn, in particular, between more or less grammatical aspect (covering notions like perfectivity) and more or less lexical (which presumably implies derivational) Aktionsart (covering various aspects of action, such as iterativity, durativity, inchoativity, habituality, telicity, etc.). According to Dressler (Reference Dressler1968: 40), this distinction dates back to Agrell (Reference Agrell1908).

Boogaart (Reference Boogaart, Booij, Lehmann and Mugdan2000: 1166) distinguishes between lexical aspect and grammatical aspect, thus reflecting the difference between Aktionsart as a lexical–derivational property of verbs and aspect per se expressed grammatically. Boogaart (Reference Boogaart, Booij, Lehmann and Mugdan2000: 1167) points out problems with this distinction, like the fact that Aktionsart is in many languages expressed by grammatical means. A case in point is Jaqaru. Hardman (Reference Hardman2000: 59) reports that the iterative suffix -q“a ‘is considered a part of the inflectional system because its use is obligatory, because it does not enter into lexicalizations (its meaning is always predictable) and morphologically it forms an integral part of the marking within the person/tense system’. However, the fact that the suffix in question is integrated in the stem seems to argue against its treatment as an inflectional suffix.

But the crux of the matter here is, again, semantic: What actually is a new word, i.e. when can we speak about a new word? Boogaart (Reference Boogaart, Booij, Lehmann and Mugdan2000: 1167) tackles the issue in the discussion of the perfective verb in Slavic languages, which are reputedly true aspectual languages: ‘the distinction is obligatory and it is independent of tense. However, it is a matter of some debate to what extent the perfectivizing prefixes of Russian in fact derive perfective verbs with a different lexical content, i.e. with a different Aktionsart.’ As grammatical markers usually develop from lexical units, it is not always possible to classify aspectual forms as belonging either to the lexicon or to the grammar. Dressler (Reference Dressler1968: 49) maintains that a number of Aktionsarten feature close semantic affinity to aspect and they usually incline to either perfective or imperfective aspect. Specifically, Dressler notes that this interconnection of aspect and Aktionsart is so strong in Slavic languages that ‘most of the Aktionsarten are exclusively bound to a single aspect’.⁸

Table 2.1 shows the extensive possibilities of prefixation in German, Latin and Slovak. Many of these prefixes combine the aspect-related and the Aktionsart-related effects. Thus, when the prefix pre- is attached to the imperfective Slovak verb písať ‘write’, its meaning acquires a perfective nuance and, at the same time, changes into ‘rewrite’. The Slovak verb niesť ‘carry’ is imperfective but, when the prefix pri- is attached, it becomes perfective and its meaning changes to ‘bring’ and, when the prefix za- is attached, the same base takes on the (perfective) meaning ‘take it to (somebody)/(somewhere)’. This type of modification is extremely common in Slavic languages and turns imperfective into perfective verbs, usually altering the original meaning of the base and, thereby, producing new words.

An even more problematic side of the category of aspect/Aktionsart is habituality and iterativity. These are further subcategories of the supercategory of quantity and it is disputable whether affixation of this type keeps the verb within the original framework of the base or whether it produces a new verb.

Various positions can be found in the literature in this respect: Anderson (Reference Anderson and Shopen1985: 7–8) describes features of aspect in Russian in terms of derivation rather than of inflection and Hardman (Reference Hardman2000: 59) points out the ambiguous nature of the category of iterativity in Jaqaru. Štasni (pers. comm.) treats the Serbian-Croatian verb do-pis-iva-ti ‘add (in writing)’, where -iva expresses iterativity, as a case of derivation, and so do Horecký and Ološtiak (pers. comm.) with regard to its Slovak counterpart dopisovať.⁹ Similarly, Fortescue (Reference Fortescue1984) considers the West Greenlandic aspectual affix -sar of repetition as a derivational one and Nikolaeva and Tolskaya (Reference Nikolaeva and Tolskaya2001: 308–9) also view deverbal aspectual affixes in Udihe as derivational.

Other examples of derivation expressed by aspectual markers are the momentaneous (semelfactive) aspect in Ket and other Yeniseian languages, which may refer to a single, unrepeated action, and the iterative aspect which refers to an ongoing action. According to Werner (Reference Werner1998: 146), the two aspects¹⁰may differ from each other by referring to either one or several objects. With verbs of motion, the aspectual forms also refer to the length of action. Finally, in Malayalam, the first constituent of verb + verb compounds always takes the form of an adverbial participle. Asher and Kumari (Reference Asher and Kumari1997: 401) make explicit mention of the borderline nature of this structure, as in (19):

(19)

(19) Malayalam

1. a. koɳʈu-varika

   take.hold.of-come

   ‘bring’

2. b. nookkikkaa ·uka

   look-see

   ‘look at admiringly’

3. c. kaɻtupitikkuka

   see-grasp

   ‘find out’

The above can be interpreted as evidence that at least those subcategories of Aktionsart which fall under the supercategory of quantity are, by analogy to plurality, more derivation-like than other subcategories of aspect and Aktionsart, because they contribute to the lexical meaning of the motivating words in one or the other way. This is also in line with Kiefer’s (1999) view of Aktionsart as ‘the meaning of a morphologically complex verb introduced by morphological means which adds a semantic feature to the meaning of the base verb’.¹¹

1.1.3.3 Plurality

Arguments have been raised in the literature concerning the derivational status of what has traditionally been considered an integral part of the inflectional system of language, as is number (cf., among others, Plank Reference Plank and Asher1994: 1672; van Marle Reference van Marle1995: 74). Beard (Reference Beard1982) lists several similarities between pluralization and derivational processes:

1. (a) formal and semantic irregularities (e.g. English deer, salmon, etc., and English skies, heavens, etc., respectively). Similarly, Russian gender distinction of singular paradigms is lost in the plural and Slovak collective nouns derived from a singular noun yield neuter gender from the masculine and, as the new noun behaves as a singular, a new inflectional paradigm,¹²

2. (b) lexical constraints, as numerous nouns cannot be pluralized or cannot take the singular form,

3. (c) preservation of number markings even where inflection completely disappears in the historical development of a language (e.g. in Bulgarian, English and Hindi), and

4. (d) use of borrowings and agreement with lexical gender rather than with inflectional case in the relative pronoun of inflectional languages like Russian.

Similarly, Kießling (pers. comm.) assumes that plural formation in Datooga and other Southern Nilotic languages is derivational because it belongs to a larger number reference system in nouns where plural and singulative are just two categories: many nouns show a tripartite distinction of the basic form with collective or generic reference, a singular reference form (singulative) and a multiple reference form (plural). All three forms could be morphologically derived, but the capacity to derive two or three forms and the selection of plural and singulative markers largely depends on the lexical root and is not predictable by phonological or morphological criteria. Therefore, plurals and singulatives should be viewed as part of the lexicon and, thus, of word-formation.

The derivational pole of the category of plurality may also be illustrated by the so-called pluractional verbs in Hausa as discussed by Newman (Reference Newman2000: 423–9). Pluractional verbs in Hausa indicate the plurality of action and are derived productively from corresponding non-pluractional verb stems by reduplication. Their meaning can concern a multiple, iterative, frequentative, distributive or extensive action. (20a) and (20b) are two of Newman’s examples, both based on preposing reduplication:

(20)

(20) Hausa

1. a. tun~tùnā

   red~tunàa

   pl/ite~remind

   ‘remind many or often’

2. b. hàhhaifàa

   red~hàifā

   pl/ite~give.birth

   ‘give birth many times or to many children’

A similar point can be made of Kwakw’ala, where plurality is a derivational category. According to Boas (Reference Boas, Boas Yampolsky and Harris1947: 246), this category may express repetition (21a), several subjects (21b) or simultaneous action (21c) in different parts of a unit:

(21)

(21) Kwakw’ala

1. a. mede´lxu~mede´lqwela

   ‘It is boiling repeatedly’

2. b. me^-ɛ~mEdE´lqwela

   ‘Many are boiling’

3. c. ma^ɛe´~mdElqwEla

   ‘It is boiling in all of its parts’

The three different plural formations in Kwakw’ala resemble the plural system in Hausa. According to Anderson (Reference Anderson and Shopen1985: 30), the derivational status of at least the temporal, aspectual and plural groups is indicated by their being ‘optional, and present only where necessary for emphasis or disambiguation; and . . . equally applicable to words of any syntactic function or word class’.

While these facts argue for a case of pluractional markers as derivational in nature, this is not so in every language. A case in point is Karitiâna, an Arikem language spoken in South America. Sánchez-Mendes and Müller (Reference Sánchez-Mendes, Müller and Deal2007) provide conclusive evidence that pluractionality excludes all singular, atomic events from the denotation of their verbs and thus behaves like the plural affix for nouns: its basic meaning is ‘more than one event’. Consequently, pluractionality in Karitiâna cannot, among other things, express distributive meaning (the distributive meaning postulates the distribution of singular events, which are, however, not in the denotation of pluractional predicates). This point has been raised in reference to another language, Ket, where an inflectional suffix with plural meaning becomes a derivational one. Werner (Reference Werner1998: 41ff.) shows that the derivational suffix -η goes back to the plural suffix -η. (22) and (23) are examples of the acquired derivational status of -η:

(22)

(22)

Ket al’eη < al + -η ‘trousers’ ‘half’ + pl

(23)

(23)

Ket ²ti^?η < ¹ti.k+ -η ‘hoarfrost’ ‘snow’ + pl

In many cases the lexicalized plural forms are collective nouns which have developed a secondary plural form, such as those in (24):

(24)

(24)

Ket ¹qə.n qənten qənteneη primary pl secondary pl ‘ant’

However, as noted by Werner (Reference Werner1998: 46), the derivational status of the original plural suffixes -η and -s^’in may be called into question, because they are preceded by a plural suffix:

(25)

(25)

Ket ⁴qä-s’, > qe-η-s’in ‘superior.n’ ‘superiors.n’

Werner refers to a similar observation by Bloomfield about German, where the plural of the diminutive Kindchen ‘small child’ is formed from the plural form of the non-diminutive base, i.e. Kinder yielding Kinderchen. As noted by Bloomfield (Reference Bloomfield1933: 226), ‘if a language contained too many cases of this sort, we should simply say that it did not distinguish such morphologic layers as are denoted by the terms “inflection” and “word-formation”.’

A different piece of evidence is the participation of derivational affixes in inflectional paradigms, as in Breton (Stump Reference Stump, Štekauer and Lieber2005: 61–4), where the highly productive suffix -enn attaches to collectives. The resulting word is in the singular:

(26)

(26)

Breton a. melien > melienenn ‘ants.col’ ‘ant.sgt’ b. gwez > gwezenn ‘trees.col’ ‘tree.sgt’ c. sivi > sivienn ‘strawberries.col’ ‘strawberry.sgt’

This suffix can derive countable nouns from mass nouns and also be attached to a count noun or an adjective to produce a semantically related noun. All the derivations are of a feminine gender regardless of the properties of the base. The ability to change word-class and to determine the gender of the resulting formation and also other facts (e.g. -enn can be followed by another derivational suffix) suggest a derivational status for -enn, whereas the fact that the choice of the form (singulative or collective) is determined by the syntactic context¹³ suggests inflectional status of -enn, because selection of inflectional forms for the sake of grammatical agreement is a typically inflectional property.

A final argument for the derivational nature of various aspects of plurality is its additional semantic content beyond the basic meaning of plural ‘more than one’. When simple plural, as in prototypical inflectional paradigms, is compared with various meanings of plurality, a substantial difference between plural and plurality follows from the inclusion in the latter of additional semantic information represented by categories like collectivity, discontinuity, distribution, iterativity, quantitativeness, simultaneity, totality, etc.¹⁴The word-formation meaning of plurality thus follows from the additional semantic content added to the simple concept of ‘more than one item or action’ by the various plurality categories, as shown in Table 1.1:

Table 1.1. A range of semantic categories within plurality

Additional examples can be found in several other languages. Campbell (Reference Campbell1985: 46) states that some plural forms in Pipil take the suffix -wan when they are possessed. The suffix -wan expresses inalienable possession, usually with kinship terms or other closely related terms:

(37)

(37)

Pipil

-manuh-wan

‘brothers’

(38)

(38)

Pipil

-miyak-wan

‘family’²⁰

Cowan (Reference Cowan1969: 108) characterizes the Tzotzil suffix -an with the meaning ‘multiple action of the same kind’ as a derivational suffix:

(39)

(39)

Tzotzil

mĭl-an

kill.it-pl

‘do a lot of killing to them’

Finally, Rubino (Reference Rubino and Haspelmath2005a: 115) points out that in Luiseño, a Uto-Aztecan language spoken in North America, different variants of reduplication can denote various plural actions: lawi means ‘make a hole’ while law-lawi means ‘make two holes, make a hole twice’ and lawa-láwi ‘make many holes, more than two’.

The evidence and the arguments cited above do not conclude for or against the derivational nature of the traditionally inflectional category of number. However, they suggest that there seem to be enough grounds to separate, in cross-linguistic characterizations, the category of plural as an inflectional category from the conceptually/semantically more complex category of plurality, and that the latter of these would fall within the scope of word-formation.

2.1 Different notions of word

It is not easy to identify what language users intuitively have in mind when we talk about words. This can be seen in the discussion of this issue by Di Sciullo and Williams (Reference Di Sciullo and Williams1987) and also in Dixon and Aikhenvald (Reference Dixon and Aikhenvald2002). Several theoretical distinctions have been introduced to clarify the issue, like the separation of different words, i.e. words from the orthographical, phonological, lexical, morphological, grammatical and syntactic points of view.

This distinction does not solve the problem of word-formation and the units identified as words on the basis of one of these criteria need not necessarily coincide with what are words according to another (Julien Reference Julien and Brown2006: 617–18). This can be illustrated with Jaqaru, in which it is convenient to distinguish between morphological words, i.e. plain roots and roots with verbal and/or nominal suffixes (including thematic morphemes, (1) and (2), and syntactic words, which also contain ‘suffixes which tie them into the sentence structure’, (3) and (4); Hardman Reference Hardman2000: 8):

A number of replacement terms for word have been made available in the literature, such as lexical unit (Cruse Reference Cruse1986), listeme (Di Sciullo and Williams Reference Di Sciullo and Williams1987: 3) or naming unit (Štekauer Reference Štekauer1998, drawing on Mathesius Reference Mathesius1975). Each of these and others which are not mentioned here, have far-reaching theoretical implications. Thus, the listeme is defined as a unit listed in the lexicon, i.e. stored in the mental lexicon of a language user. Importantly, Di Sciullo and Williams (Reference Di Sciullo and Williams1987) explicitly emphasize the difference between listedness (listing a unit in the lexicon) and wordhood: not only words, but also morphemes, idioms, phrases and even sentences like English How do you do? are listed, because they do not have any internal structure to predict their semantic or grammatical behaviour and, therefore, must be memorized and stored in the lexicon. The lexicon is thus presented as a set of irregularities: ‘a collection of the lawless’ (Di Sciullo and Williams Reference Di Sciullo and Williams1987: 4). Yet, a great number of complex words are not listed, specifically those based on regular and productive word-formation rules. In other terms, it is not the word but the particular word-formation rule that is stored.

Listedness raises the issue of whether the various forms of words, the word-forms which are produced for categories like case, number, gender, person, tense and a number of other categories in inflectional languages are stored in the lexicon. It could be argued that they are stored and that, therefore, they should be considered listemes and, consequently, words. However, it can also be said that, as with productively coined words, a paradigm is identified with a word stored in the lexicon and the individual forms are generated for the purpose of syntax. It may even be the case that different inflected forms are dealt with differently. Julien (Reference Julien and Brown2006: 618) gives an example from Northern Saami, a Uralic language spoken in Eurasia: while it may be assumed that beana ‘dog.nom’ and beatnaga ‘dog.acc’ are both listed, the listedness of beatnagiidisctguin ‘with their (own) dogs’ is questionable.

Listedness is also difficult to apply to idioms: while they are semantically idiosyncratic and therefore should be listed (by the criteria given by Di Sciullo and Williams Reference Di Sciullo and Williams1987), they are also analyzable and therefore should not be listed (by the criteria given by Lieber Reference Lieber1981). The confusion is aggravated by the scalar nature of the compound vs idiom relationship, where compounds and idioms form a continuum of different degrees of transparency/opacity (cf. Kavka Reference Kavka2003, Reference Kavka, Lieber and Štekauer2009; Fernández-Domínguez Reference Fernández-Domínguez2009: 68). This discussion thus leads to the question of whether compounds should be listed or not and, if so, which compounds should be, which should not be and why.

Since, however, the linguistic signs which fall within the scope of word-formation are of a highly diverse nature, both cross-linguistically and not infrequently within one and the same language, linguists came to be rather uncertain as to what actually is a word and what is not. The following examples from several languages of a number of language families show that what is a word in one language can correspond to a sentence in another, in this case in English:

Fortescue (Reference Fortescue1980: 260) demonstrates for West Greenlandic that the sentence words of (5) to (10)¹ should be viewed as real polysynthetic words rather than as strings of analytic words (cf. in this respect Anderson Reference Anderson and Shopen1985: 32–3, in reference to the polysynthetic language Kwakw’ala). Thus, the status of some sequences can be other than words or sentences by falling between the two in some ways. Schiffman (Reference Schiffman1996) notes that Tamil can combine nouns with verbs and make compound verbs, i.e. verbs that are equivalent to a verb in another language. The nominal part of this compound is not marked for case, though it may in fact be the semantic object of the verbal action.

The notion of compound makes the point at issue even more difficult to handle. Specifically, the separation of compounds from phrases has been described as ‘one of the more vexed problems in morphological theory, particularly in languages which are relatively poor in agreement morphology’ (Spencer Reference Spencer, Booij, Lehmann and Mugdan2000: 315; cf. also Giegerich Reference Giegerich2004, Reference Giegerich2009 on English; Booij Reference Booij2009 on Greek and Dutch; Schäfer Reference Schäfer2009 on Mandarin Chinese; van Goethem Reference van Goethem2009 on Romance and Germanic languages). This is the result not only of the vagueness of the notions under discussion and of the different theoretical positions that can be taken with respect to them, but also of the fact that a number of compounds develop from phrases.

The concept word is thus a fuzzy notion covering a range of various referents. In addition, the issue of wordhood is further obscured by the status of some bound elements which semantically behave as free morphemes and therefore remind us of Beard’s (Reference Beard1981) forty-four cognitive categories which can be potentially expressed by both inflection and derivation. From this it follows that there is a large number of elements which have identical cognitive bases and unequal status in terms of wordhood. Malkiel (Reference Malkiel and Greenberg1978: 127) gives examples of this well-known Indo-European phenomenon from German and Latin, but this is also common in Slavic languages as seen in Table 2.1.

Table 2.1. Examples of bound elements in German, Latin and Slovak

Similar cases exist beyond the Indo-European linguistic area too. As noted by Dixit (pers. comm.), attachment of postpositions or adverbs to the stem is the crucial process of word-formation in Marathi. It is applicable to nouns, pronouns, adjectives, postpositions, adverbs and verbs, and produces adverbs, adjectives and conjunctions. Laca (Reference Laca, Haspelmath, König, Oesterreicher and Raible2001: 1219) also mentions formatives that have freely occurring particles but do not qualify automatically as lexemes:

1. (a) adpositional or adverbial particles that play a role both in phrase and in lexeme formation, mostly as prefixes, e.g. German gegen ‘against’, gegen mich ‘against me’ vs Gegenteil ‘opposite’ or vor ‘before, in front of’, vor ihm ‘before him’ vs vorkommen ‘happen’,

2. (b) lexemes belonging to major lexical categories that are being grammaticalized as derivational affixes by a well documented diachronic process in which derivational patterns emerge from series of compounds (or of set phrases that have been subject to univerbation). Thus, the status of German frei ‘free’, voll ‘full’ and arm ‘poor’, in risikofrei ‘free from risk’, angstvoll ‘fearful’ and gefühlsarm ‘showing little feeling, insensitive’ is doubtful. In such cases, affixal status only becomes certain when the corresponding lexeme disappears (as in the case of German -heit; cf. also English suffixes developed from lexemes) or when a clear semantic and/or formal split intervenes that disrupts the etymological connection (English full vs reduced and unstressed -ful, Spanish mente ‘mind’ vs the adverbial suffix -mente).

As in other issues, a continuum is assumed to exist between affixes (bound units) and words (free units), with different sorts of units between them, like various types of clitics (Zwicky Reference Zwicky1977). Thus, in describing Chalcatongo Mixtec, an Oto-Manguean language spoken in North America, Macaulay (Reference Macaulay1996: 14) refers to the category of phrasal affixes (i.e. affixes attached to phrases) and maintains that they are affix-like in terms of dependence and word-like in terms of placement.

2.2 The status of compounds

A number of criteria have been proposed in the literature for the identification of compounds. These criteria may have to be applied differently in different languages, as Plank (Reference Plank and Asher1994: 1672) proposed for the opposition inflection vs derivation, or each language may have additional criteria bound to its specific system: besides the difficulty inherent in the concept of compounding, the cross-linguistic study of compounding is further complicated by the fact that some languages may accept as compounds units which would not be accepted as such in other languages,² or by the fact that they just do not exist in all languages (L. Bauer Reference Bauer and Brown2006: 721–2).³ A similar picture is obtained for noun incorporation, where opposing views and varied cross-linguistic evidence make it difficult to go beyond a rather general description.

Leaving aside the sentence-like words illustrated in (5) to (10), one of the central problems of word-formation is the position of compounds and their relation to syntactic phrases. The special position of compounds may be seen from their treatment or, rather, lack of any treatment, in early lexicalist works. Neither Halle (Reference Halle1973), in the first programmatic paper on generative word-formation, nor Aronoff in his breakthrough monograph of generative word-formation (1976), deal with compounds. Halle speaks of two types of word-formation rules, those combining stems and affixes, and those combining words and affixes. Aronoff’s word-formation rule is a phonological operation on the base which consists in attaching an affix.

In the transformationalist tradition, compounds are generated from kernel sentences and are treated as syntactic structures (Lees Reference Lees1960, Reference Lees, Bierwisch and Heidolph1970; Marchand Reference Marchand1965a, Reference Marchand1965b; Kastovsky Reference Kastovsky1969; Li 1971;⁴ Levi Reference Levi1978; Roeper and Siegel Reference Roeper and Siegel1978).

Compounds are assigned a special status by, among others, Strauss (Reference Strauss1982) and Anderson (Reference Anderson1992), who identify three components of morphology (inflection, derivation and compounding), and also by Jackendoff (Reference Jackendoff, Lieber and Štekauer2009: 113) in his conception of protolanguage as cognitive scaffolding on which modern language is built:

No matter which of the numerous positions we adhere to, it is necessary to admit the diversity of views reflecting the diversity of the phenomenon of compounding, a word-formation process that is far from homogeneous. A case in point is Tzotzil, in which compounding admits four different types of stems (Cowan Reference Cowan1969: 89ff.):

1. (a) root (i.e. one morpheme),

2. (b) radical (root + one or more derivational affixes),

3. (c) compound itself, and

4. (d) compound radical (compound + one or more derivational affixes).

This gives rise to a number of combinations which are summarized in Table 2.2.

Table 2.2. Some combinations of the internal structure of compounds in Tzotzil

A number of references can be found in the literature on the identification of compounds (Marchand Reference Marchand1969; Adams Reference Adams1973, Reference Adams2001; L. Bauer Reference Bauer2001; Donalies Reference Donalies2004). One of the most recent ones, Lieber and Štekauer (Reference Štekauer2009), discusses various aspects of compoundhood (wordhood) and maps the state of the art in the field from the perspective of various theoretical approaches as well as by describing the process of compounding in various languages of the world. In the following we confine ourselves to the illustration of the word-formation/syntax problem by discussing the status of noun incorporation.

2.3 Compounding and noun incorporation

Like compounding, noun incorporation appears as one of the most problematic areas at the word-formation/syntax interface. Noun incorporation is here understood as a verb-forming process whereby a nominal stem is fused with a verbal stem to yield a larger, derived verbal stem. In noun incorporation, the incorporated noun functions as an argument (usually object) of the predicative verb. This is a definition of prototypical cases of noun incorporation, but this definition is not an all-encompassing one.⁵ Sadock (Reference Sadock and Brown2006: 585) shows that the scope of the concept of incorporation varies considerably, ranging from the original, narrowly defined noun incorporation up to any juxtaposition of two functional or lexical categories (Baker Reference Baker, Everaert, Evers, Huybregts and Trommelen1988).

Thus, while incorporation usually makes use of full words, there are also cases in which incorporation is based on an affix, like nominal stems fused with an affix. Recognizing an affix-based incorporation can be, according to de Reuse (Reference de Reuse1994: 200), traced back to Kroeber (Reference Kroeber and Heger1910). In West Greenlandic there are many bound verbal suffixes, such as -qar ‘have’, -si ‘get’, -lisaar ‘wear’, -tur ‘eat/drink’, -liur ‘make’ and -ssaaliqi ‘lack’ (Fortescue Reference Fortescue1984: 82):

Fortescue (Reference Fortescue1984: 84) wonders ‘whether “incorporation” is the best term to use here, since all verbalizing affixes like qar are bound forms, never stems’. In fact, as argued by Malouf (Reference Malouf, Webelhuth, Kathol and Koenig1999), while the resulting denominal verb has the full syntactic and morphological distribution of any verb in West Greenlandic and retains some of the properties of the incorporated nominal, it has little in common with noun incorporation constructions in languages like Mohawk, an Iroquoian language spoken in North America, or in Southern Tiwa, a Kiowa-Tanoan language also spoken in North America.

This case is not unique: Mithun (Reference Mithun1984: 885) reports on languages that ‘contain affixes which function much like I[ncorporated]N[oun]’s, often highly productively. Others contain affixes which function much like incorporating V[erb]’s, again often very productively.’ Similarly, Gerdts (Reference Gerdts, Spencer and Zwicky1998: 94–7) refers to the abundance of the so-called lexical suffixes in Salishan, Wakashan and other northwestern Amerindian languages which carry a nominal meaning denoting body parts, environmental concepts, cultural items and human terms.⁶ These suffixes undergo the same kind of incorporation as nouns and establish the same kind of relations as stem-based noun incorporation. In Gerdts’ view, ‘lexical suffixes can be regarded as incorporated nouns that have lost their status as free-standing nominals’ (Reference Gerdts, Spencer and Zwicky1998: 97).

A different case is offered by noun-stripping-based incorporation in Lakhota,⁷ in which a noun is followed by an adjectival modifier:

The usual objection is that a noun cannot be considered stripped if followed by an adjectival modifier. De Reuse argues, however, that the noun and the adjective are not a phrase. Instead they form a syntactic compound with its characteristic stress (i.e. main stress on the second syllable of the first constituent and secondary stress on the second syllable of the second constituent).

Mithun (Reference Mithun1984: 847) points out that ‘noun incorporation is perhaps the most nearly syntactic of all morphological processes’ and Katamba (Reference Katamba1993: 287) adds that it ‘obscures the boundary between syntax and morphology’. This is clear in borderline cases like the following from Udihe, where the bond between the verb b’a- ‘get’ and the incorporated noun in accusative is rather loose (Nikolaeva and Tolskaya Reference Nikolaeva and Tolskaya2001: 326ff.), and the latter may be modified by an adjective (24) and may even be separated from the governing verb (25) (Nikolaeva and Tolskaya Reference Nikolaeva and Tolskaya2001: 327):

In principle, there are two major positions on the status of (noun) incorporation: the syntactic position, which explains noun incorporation by a movement operation in syntax, and the lexical/word-formation position. In Leza’s (Reference Leza, Haspelmath, König, Oesterreicher and Raible2001: 714) view, neither the proponents of a syntactic approach nor those of a morphological (i.e. lexical) approach have argued unambiguously and compellingly in favour of their respective positions.

The syntactic approach to noun incorporation goes back to Mardirussian (Reference Mardirussian, Grossman, San and Vance1975) but was primarily developed by Sadock (Reference Sadock1980, Reference Sadock1985, Reference Sadock1986) and Baker (1983, Reference Baker1985, Reference Baker, Everaert, Evers, Huybregts and Trommelen1988).⁹ The argument is that noun incorporation is basically a syntactic process. In Katamba’s (Reference Katamba1993: 283) words it is ‘a productive syntactic rule that builds a special type of compound verb’. Baker makes use of his Mirror Principle to suggest that the processes employed in generating a complex form by noun incorporation mirror the syntactic processes. In his view, noun incorporation is based on the movement of a word from its position in a noun phrase of a sentence to a position inside a verb. Thus, Kageyama (Reference Kageyama1982: 243) treats noun incorporation like Japanese 腰掛ける (kosi-kakeru) ‘sit down’ (lit. ‘put one’s buttocks’),手間取る (tema-doru) ‘take time’, 物言う (mono-yuu) ‘say (something)’, etc. as nominalization compounds ‘derived from their paraphrasing sentential constructions’ by movement transformation. Baker argues in favour of a syntactic nature of noun incorporation by assuming that:

1. (a) incorporation can leave behind the non-head portions of the noun phrase,

2. (b) a copy may be left behind,

3. (c) only objects and subjects of intransitives can be incorporated, this being a syntactic specification, and

4. (d) incorporated nouns may introduce new discourse referents

   (cf. Di Sciullo and Williams Reference Di Sciullo and Williams1987).

Syntactic arguments were called into question by Mithun (Reference Mithun1984), Di Sciullo and Williams (Reference Di Sciullo and Williams1987: 63–9) and Spencer (Reference Spencer1995, Reference Spencer, Booij, Lehmann and Mugdan2000). Mithun (Reference Mithun1984, Reference Mithun1986), Rosen (Reference Rosen1989), Anderson (Reference Anderson1992: 267–9) and Gerdts (Reference Gerdts, Spencer and Zwicky1998) propose a lexical approach instead, and treat noun incorporation as a lexical process, specifically as a special case of compounding. Thus, Gerdts (Reference Gerdts, Spencer and Zwicky1998: 84) defines noun incorporation as ‘the compounding of a noun stem and a verb (or adjective) to yield a complex form that serves as the predicate of a clause’.¹⁰

The proponents of the lexical/word-formation approach to noun incorporation agree on its general compounding principle, but the details and foci of their respective treatments of noun incorporation differ. Thus, Mithun (Reference Mithun1984) distinguishes four noun incorporation types primarily on the basis of various (discourse) functions and Rosen (Reference Rosen1989) classifies noun incorporations according to their syntactic features as two groups:

1. (a) those which preserve the argument structure of the incorporating verb intact (‘classifier N[oun] I[ncorporation]’),¹¹ and

2. (b) those in which one argument of the verb is satisfied with the complex verb which results in turning a transitive verb into an intransitive one (‘compound N[oun] I[ncorporation]’).

Noun incorporation is, in the vast majority of cases, connected with incorporating a single noun. Yet, a rare case of two incorporated nouns, mostly combined with locative and instrumental prefixes, can be found in Lakhota. One such example, whose structure is [N+[Loc+[N+[Loc+[Loc+V]]]]], is given in de Reuse (Reference de Reuse1994: 216):

De Reuse (Reference de Reuse1994: 202) also shows that, in addition to an incorporated noun stem, Lakhota stative verbs can incorporate a pronominal prefix too. Thus, the stative verb chóla ‘be missing’ in (27) below has two arguments after noun incorporation: the stative pronominal prefix ma- ‘first person’ and the incorporated noun sí ‘foot’:

The inclusion or not of noun incorporation in the field of word-formation is just one example of how relevant theoretical frameworks can prove in typological research. It also illustrates that, in some cases, theory-independent approaches in typological research are just not possible.

Another point concerns the internal relations in word-formation itself. If we admit that noun incorporation is a word-formation process, we are then probably expected to decide on such borderline cases as represented by this phenomenon in English. It has long been assumed by the majority of morphologists that examples like English to chain-smoke, to sleep-walk, etc. are back-formations from the corresponding verbal compounds. The arguments are based mainly on the diachronic data indicating that the longer forms appeared historically before the shorter ones.

Not only does this sort of argument not seem to be sufficiently reliable to justify the claim concerning the nature of a word-formation process, but it may also be argued that the original process of back-formation in English gave rise to numerous analogical formations which did not rely on longer counterparts any more. The growing number of such formations¹² may have triggered a synchronically productive process of noun incorporation. This account is favoured in Kiparsky’s approach (Reference Kiparsky and Yang1982a), which explains the generation of this class of expressions as compounding, based on the rule [Y Z]_X (where X stands for a verb). An even stronger claim in this line was made by Kastovsky (Reference Kastovsky, Kastovsky and Szwedek1986: 419), when he predicted the gradual development of a productive process of noun incorporation. Kastovsky (Reference Kastovsky, Lieber and Štekauer2009: 338) recently strengthened this claim by arguing that ‘this type has become rather popular in Modern English as a result of the re-interpretation of denominal back-formations’. The position presented here, as in the opening lines of this section, is in accordance with E. Sapir (Reference Sapir1921), who also treats this kind of complex words as cases of noun incorporation: noun incorporation is thus viewed as compounding, i.e. a process controlled by word-formation rules (cf. Štekauer Reference Štekauer1998 and Reference Štekauer, Štekauer and Lieber2005).

2.4 Summary

Compounding and noun incorporation lie at the interface of word-formation and syntax and show that similar problems to the ones discussed in chapter 1 on the limit between inflection and derivation can be found between word-formation and syntax. In this respect, word-formation is no exception to the general fuzzy nature of linguistics. Partly for this reason, derivation, like inflection, is best described in terms of prototypical characteristics or, maybe, it can be described only in terms of prototypes. Unlike what happens in inflection and derivation, where the separation is presented as a gradient both by the literature (among others, Bybee Reference Bybee1985: 5; Plank Reference Plank and Asher1994: 1672) and by a review of language systems, the separation between compounding or noun incorporation and syntactic structures is much more difficult to express even in a gradient.

The implications for word-formation typology and universals are that decisions concerning the selection of the categories considered to be word-formation may be called into question. The range of ambiguous categories is relatively large. What seems therefore necessary is to focus on the prototypical and most productive instances of the individual word-formation processes. With such an approach (cf. also Booij Reference Booij and Brown2006: 654), there may still be a substantial space of uniqueness pertaining to both formal and semantic characteristics of the individual word-formation processes and categories and, first and foremost, to their respective functions.