University Press.

Uni-directional flexibility and the noun–verb distinction in Lushootseed. In Jan Rijkhoff and Eva van Lier (eds.), Flexible word classes: A typological study of underspecified parts-of-speech. Oxford: Oxford University Press. Recent work on the typology of parts-of-speech systems has shown that a significant parameter of variation in the organization of the lexicon concerns the number of open or major word classes that are recognized in a language. While languages with the familiar Indo-European system distinguish four major " contentive " classes (noun, verb, adjective, and adverb), it is not uncommon for languages to distinguish fewer. In many such cases, a language with a reduced parts-of-speech inventory conflates two or more major classes, creating a flexible part of speech that fills a variety of syntactic roles. One of the most contentious issues that falls out from this observation is whether or not it is possible for a language to conflate all of the major lexical classes, grouping all of its contentive lexical items into a single, maximally flexible class of words (opposed only by the minor, grammatical classes) and thereby neutralizing the distinction between nouns and verbs. Claims for the absence of a noun–verb distinction have been advanced for a number of languages and are discussed most extensively for languages from the Salishan, Polynesian, and Munda families. Examining these cases reveals that they fall into two general types which I will refer to in this paper, loosely following Evans and Osada (2005), as precategorial and omnipredi-cative. The precategorial type of language, as represented by languages of the Munda and Polynesian families, has received the most attention in the recent literature (e. the omnipredicative type has not been discussed to the same extent, although languages of this kind, particularly those belonging to the Salishan family, are frequently offered uncritically as examples of languages that lack a distinction between nouns and verbs. In this paper, I will present data from the Salishan language Lushootseed 1 which demonstrates that, while the noun–verb distinction is neutralized in syntactic predicate position, it is still relevant for words used as syntactic arguments, giving us a pattern that will be referred to here as unidirectional flexibility. Unidirectional flexibility as the term is used here is intended to complement the notion of " bidirectional flexibility " put forward by Evans and Osada (2005) as a criterion for determining whether or not a language has genuinely neutralized a part-of-speech distinction. For …

typical of two (or more) parts of speech, thereby conforming to the definitions of both lexical classes.Unidirectional flexibility, on the other hand, entails that for a particular pair of lexical classes, X and Y, X can appear in the syntactic roles criterial for Y, but Y can not appear in the roles criterial for X.While unidirectional flexibility entails the neutralization of a parts-of-speech distinction in a particular syntactic environment, it can not be equated with the complete absence of the distinction.Unidirectional flexibility and methods for establishing it will be discussed in Section 1 of this paper, following which Lushootseed data illustrating this pattern will be presented in Section 2. The facts in Lushootseed closely parallel those of other Salishan languages, which in turn seem substantially the same as the patterns seen in other languages that have been claimed to follow the omnipredicative pattern of noun-verb neutralization, implying that omnipredicative languages in general show only unidirectional, rather than genuine bidirectional, flexibility between nouns and verbs.Since precategorial languages are also argued (for different reasons) by Evans and Osada (2005) not to represent a genuine example of noun-verb flexibility, it seems probable that a distinction between nouns and verbs is indeed a universal of human language (Croft 2003).Evans and Osada's position, some counter-arguments to it, and some of the more general implications of this discussion for typological approaches to parts-of-speech systems will be discussed in Section 3.

Flexibility in parts-of-speech systems
The notion of flexibility in parts-of-speech systems is first articulated in the context of a full typology of lexical classes by Hengeveld (1992a, b), who uses the term "flexible" to refer to a part of speech that meets two or more of the definitions for lexical classes given in (1), these definitions hinging crucially on the morphosyntactic properties of classes of lexical items appearing in certain criterial syntactic environments:2 (1) verb-a lexical item which, without further measures being taken, has predicative use only noun-a lexical item which, without further measures being taken, can be used as a syntactic argument adjective-a lexical item which, without further measures being taken, can be used as the modifier of a noun adverb-a lexical item which, without further measures being taken, can be used as the modifier of a syntactic predicate3 (adapted from Hengeveld 1992b: 58) Thus, a class of lexical items is said to be flexible if it meets, say, both the definition of an adjective and of an adverb simultaneously.Hengeveld further proposes, based on a moderately large sample of languages, that the patterns of flexibility thus defined are not unconstrained, but follow the implicational hierarchy shown in (2): (2) Parts of Speech Hierarchy Syntactic predicate > Syntactic argument > Adnominal modifier > Adverbal modifier (adapted from Hengeveld, Rijkhoff and Siewierska 2004) According to (2), a part-of-speech system that has a class of words used both as unmarked syntactic predicates and as unmarked syntactic arguments will also use the same class of words for adnominal and adverbal modification; a flexible class of words that is used as a syntactic argument and adnominal modifier must also be flexible with respect to adverbal modification; and so on.The resulting taxonomy of flexible parts-of-speech systems is shown in Fig. 1:  Since its inception, this taxonomy has been influential and controversial, both in terms of the typological predictions it makes and in terms of the methodological implications it has for the investigation of lexical class systems.One methodological question that is of importance to us in the context of the present discussion is the notion of "without further measures," which is defined by Hengeveld in rather vague terms and seems to correspond roughly to additional morphological or syntactic means required for the use of a particular lexical item in a non-canonical syntactic role (referred to by Tesnière 1934Tesnière , 1959 as "transfer") as "transfer").Some of the implications of this are discussed in Beck (2002), where it is proposed that "further measures" be re-defined in terms of the relative markedness of particular lexical classes of item in specific syntactic roles.Of the criteria for determining relative markedness, the most relevant for this paper is the notion of Structural Complexity: (3) Structural Complexity: An element X is marked with respect to another element Y if X is more complex, morphologically or syntactically, than Y Applying this measure to parts-of-speech typology, establishing the markedness of a lexical class X relative to lexical class Y requires showing that members of Class X are relatively more structurally complex than those of Class Y in a criterial syntactic environment A. This is in essence the equivalent of the descriptive claim that words of Class X are the target of morphosyntactic rules (aimed specifically at Class X, which must therefore be recognized in the lexicon) allowing for their use in environment A. Re-formulating this in terms of markedness allows the analyst a principled way to establish language-specific diagnostics and criteria for structural complexity (thereby avoiding what Croft 2005: 434 refers to as "methodological opportunism").When a part-of-speech distinction exists between two word classes, each with its own distinct unmarked syntactic role, the comparison of the properties of Classes X and Y in two of the criterial syntactic roles identified in (1), A and B, would give us the pattern shown in Fig. 2: Here, Class X (say, for English, nouns) is relatively more complex in terms of derivational or syntactic means, and therefore marked, in Role A (syntactic predicate) than Class Y (verbs), while in Role B (syntactic argument) Class Y is relatively more complex than Class X, giving us a clear bidirectional lexical class distinction. 4n the case of a truly flexible part of speech, we would expect that for an established class of words, any bipartition of that class into two putative sub-classes X and Y (at random or based on semantic criteria) would show the pattern in Fig. 3: In this situation, words belonging to Class X are unmarked in both criterial Roles A and B, whereas words in Class Y are unmarked only in Role A, but are marked in Role B -in effect, the distinction between Classes X and Y is present in the language but is "neutralized" for Role A. Situations such as this are not uncommon in languages, but do not constitute genuine flexibility: it is still possible to define distinct word-classes based on the contrastive properties of the two classes in Role B. Thus, for instance, in a language where words with substantive meanings (Class X) are both unmarked as predicates (Role A) and as syntactic arguments (Role B), but words expressing events (Class Y) are unmarked predicates but marked arguments, Class Y still conforms to the definition of "verb" given in (1) but does not conform to the definition of "noun."Class X, on the other hand, conforms to both (or would, without the stipulation that verbs be "only" syntactic predicates), but, given the contrast with Class Y, can be classified as a flexible class of nouns.As will be shown in Section 2 below, the apparent neutralization of the noun-verb distinction in Salishan languages constitutes a clear case of this type of unidirectional flexibility; because Salishan provides a typical case of what Evans and Osada (2005) refer to as an "omnipredicative" language (borrowing the term from Launey 1994) the discussion below strongly suggests that languages in this category do not constitute a genuine case of noun-verb flexibility.
Another controversial aspect of the definitions of parts of speech in (1) is the absence of semantic criteria associated with any of the word classes (Beck 2002).This seems unfortunate from a theoretical point of view, given the well-known and quite robust clusterings of certain meaning-types around certain parts of speech, shown in Fig. 5   While it is well-known that semantic category membership is (at best) problematic for establishing lexical-class membership, it is nevertheless true that accounting for these patterns is a desirable feature for a parts-of-speech typology.One of the unintended consequences of this focus on syntactic over semantic criteria is that it often leads to a tacit methodological bias towards strictly morphosyntactic comparisons of related wordforms in different syntactic environments without attention to concomitant differences in their meanings (a similar point with respect to the Salishan noun/verb issues is made in Van Eijk and Hess 1986: 328).In cases where the two wordforms being compared are phonologically identical, however, inattention to semantics -particularly changes in the meaning of one of the forms associated with its appearance in a particular syntactic role -leaves the door open to a (mis)analysis wherein two words are judged to be morphosyntactically equivalent in spite of a significant semantic difference between them.This approach begs the question of whether or not the two items being compared are, in fact, the same word with a flexible distribution, rather than two homophonous forms, each with its own meaning and syntactic distribution.While such cases generally pass without comment in languages like English, which has a number of such pairs of homophonous forms (e.g., hammer N vs. hammer V , cook N vs. cook V ), there appear to be languages (the most frequently cited examples being languages from the Munda and Polynesian families) where such pairs are very common, perhaps to the point of constituting the bulk of the lexicon.As Evans and Osada (2005) note, languages of this type, often referred to as precategorial languages, constitute a second language-type that is often analyzed as having a flexible class of words fitting the definition of both nouns and verbs. 5For the case of Mundari, Evans and Osada argue (on different grounds) that this situation is, like the omnipredicative case, not an example of true flexibility.Thus, with both putative types of noun-verb flexibility in doubt, it would seem that the case for the typology in Figure 1 is considerably weakened, at least insofar as the possibility of having a language with a single major part of speech is concerned.I will return to this point in the conclusion to this paper.

Unidirectional flexibility: Noun and verb in Lushootseed
One of the most frequently-cited cases of a language family that is flexible with respect to the noun-verb distinction is that of Salishan languages.These claims were put forth initially in the specialist literature (e.g., Kuipers 1968;Kinkade 1983;Jelinek and Demers 1994) and then adopted by typologists interested in variation in parts of speech systems (e.g., Broschart 1991;Sasse 1993;Bhat 1994;Hengeveld and Rijkhoff 2005), although the current consensus in the Salishanist community seems to be against this position (e.g., van Eijk and Hess 1986;Demirdache and Matthewson 1995;Matthewson and Davis 1995;Davis andMatthewson 1998, 1999;Kroeber 1999;Beck 2002).The primary argument for the absence of a noun-verb distinction in Salishan is data such as that from Lushootseed shown in (4): 6 dicative and precategorial type, they mention the "Broschartian" language and languages with "rampant" conversion.The thrust of their article, however, is to show (I believe correctly) that the precategorial and Broschartian types of language are, in fact, better analyzed as languages with rampant conversion -and that this last category does not in fact represent a true example of noun-verb flexibility.Since the term "rampant conversion" language presupposes the outcome of this discussion, I have chosen to use "precategorial" for the moment as a more neutral cover term. 6The abbreviations used in this paper are as follows: -= morpheme boundary; = = clitic boundary;  (Hess 2006: 49, 180) Sentences such as (4a) represent a fairly common type of construction where the syntactic predicate is a word with a substantive meaning, sbiaw 'Coyote', whose subject appears to be a word expressing an event, ʔuxw 'go'.As in all sentences with substantive syntactic predicates, the meaning of the construction here is equative.Likewise, in (4b) the syntactic predicate is the substantive p'q'adᶻ 'rotten log' and has as a subject what appears to be the translation equivalent of a verb, dxʷxqabac 'be wrapped up inside', inflected for the stative aspect and preceded by a determiner.As shown in (4c), words corresponding to English verbs are not confined to argument-position in constructions with substantive predicates: in this sentence, the predicate is t'əq'ʷ 'snap', but the subject is apparently the expression of an event xwaqʷabac 'be wrapped around body'.Likewise, (4d) shows that event-words can be direct objects of transitive predicates.Such words can also appear in other syntactic argument roles such as agentive complement of a passive (5a), as well as being found in (non-criterial) roles such as complement of a preposition (5b) which are cross-linguistically most typical of nouns: ( 'and they were thrown aboard' (Hess 2006: 58, line 399) In fact, if judged on superficial distributional criteria, there appear to be no syntactic roles that are open to words with substantive meanings (i.e., words we would expect to be nouns) that are not also open to words that express events (words we would expect to be verbs).
On the basis of evidence such as that presented in (4) and ( 5), it would seem that a prima facie case for the absence of a noun and verb distinction in Salishan languages can be made.However, closer examination of data from Lushootseed reveals that all in fact is not as it seems: while it may be true that the noun-verb distinction is neutralized in syntactic predicate position (Section 2.1), it can be shown to persist for words from these two classes in syntactic argument position (2.2).The primary piece of evidence for this persistence is that argument-phrases like ti ʔuxw 'the one who goes' in (4a) are, as their translation implies, headless relative clauses (2.2.1).Furthermore, there are constructions in which words expressing events appearing in argument position show clear morphological and syntactic evidence of recategorization as nominals (2.2.2), as well as constructions in which the treatment of particular words depends crucially on whether they belong to a nominal or a verbal lexical class (2.2.3).Thus, Lushootseed (and most likely Salishan languages in general) do not meet Evans and Osada's (2005) criteria of bidirectionality, and so do not constitute a true case of noun-verb flexibility, but instead correspond to a case of unidirectional flexibility.

Neutralization in predicate position
As shown in the previous section, word classes in Lushootseed do show flexibility in that the noun-verb distinction in the lexicon is neutralized in syntactic predicate position, as illustrated by the sentences in (4a) and (b), which have as syntactic predicates words with substantive meanings ('coyote' and 'rotten log', respectively).Lushootseed is a predominantly predicate-initial language, and so in the matrix clause the syntactic predicate is the first word belonging to a major word class: ( 'I will be a big man when I grow old' (Bates, Hess and Hilbert 1994: 109) As the data in (9) show, migration applies equally to the subjects of event-word and substantive predicates.

FOC PAST=just PAST=ADD=arrive
'it was he who had just kept arriving there' (Hess 2006: 21, line 244) These clitics can appear on event-word predicates, as shown in (10a), or on substantive arguments, as in (10b) and (c).( 10b) also shows the past-tense clitic appearing on a substantive predicate, q'iyaƛ'əd 'slug'. 7The phrase-level clitics may appear more than once in a clause, (10b) and (c), or even be iterated within a single phrase, (10d).The fact that all of these clitics are found attached to all types of words in both predicate and argument phrases has been in some measure responsible for the misapprehension that lexemes of all classes take word-level inflections for tense, aspect, and mood categories.8Just as subject inflection is the same for substantive and event-word predicates in matrix clauses, it is also the same in subordinate subjunctive clauses whose syntactic predicates are words of either type.These clauses take a special series of subject enclitics, illustrated in ( 11 'finally an even bigger river was made for them, if it was a river' (Hess 2006: 36, line 354) The full set of subjunctive person-markers is given in Fig. 6: 'if you folks always chase me' (Hess 1967: 52) Here, the subjunctive person marker appears in a clause introduced by an adverb, ck'ʷaqid 'always'; because the adverb is clause-initial, the person-marker is encliticized to this word rather than to the syntactic predicate, maintaining its clause-second position.Once again, the fact that these person inflections are sentence-second clitics is often overlooked, and these morphemes have been used as evidence for the claim that words of any class (including adverbs) can be inflected for person and number of their subjects in subjunctive clauses.
As noted by Kinkade (1983) for Salishan in general, the patterns shown by words expressing events and substantives in predicate position also apply to other types of word.Thus, Lushootseed has clauses predicated on a variety of word classes such as lexical pronouns (13), adverbs (14), numerals (15), and interrogative words ( 16 PR NSPEC meat 'into the fire falls the meat' (lit.'the fall of the meat is into the fire') (Kroeber 1999: 381) b. tul'ʔal čəd sqaǰət tul'-ʔal čəd sqaǰət CNTRFG-at 1SG.SUB Skagit 'I am from Skagit' (Bates et al. 1994: 6) Note that in (17b) the subject marker immediately follows the preposition, separating it from its complement in order to maintain sentence-second position.
Predicates such as those in ( 13)-( 16) also take the subjunctive person-markers in subjunctive subordinate clauses: (18) a. gʷudaʔatəb dᶻəɬ gʷəcədiɬəs kʷi gʷəuʔatəbəd gʷə=ʔu-daʔa-t-b dᶻəɬ gʷə=cədiɬ=əs kʷi gʷə=ʔu-ʔatəbəd Based on the data presented here, then, it does seem that Lushootseed (like most of the languages in the family) shows a great deal of flexibility with respect to what class or classes of word are unmarked syntactic predicates.Nevertheless, the neutralization of lexical class distinctions in predicate position is not enough to demonstrate that the noun-verb distinction is non-existent in the Lushootseed lexicon or is irrelevant to Lushootseed syntax: in addition to showing that all words are unmarked in the criterial syntactic role for verbs, it is also necessary to show that all words are unmarked in the criterial syntactic role for nouns.As will be seen in the following section, this is clearly not the case.

Contrastive behaviour in argument position
As demonstrated in the preceding section, the syntactic parallels between substantive and event-word predicates in Lushootseed are exact, and the similarities between the two are heightened by the behaviour of person-markers and phrase-level inflectional clitics, which contribute to the illusion that the inflections pertaining to the predicate phrase are morphological categories of a conflated noun-verb lexical class.9This pattern of neutralization in predicate position, however, is not all that uncommon, and is found in a variety of languages (e.g., Buriat, Arabic, Nanay, Beja) where the noun-verb distinction is not in question (Beck 2002: 107-108) On the surface, the subjects of the sentences in (19) all appear to have a similar syntactic structure: a determiner followed by a lexical item with either a substantive meaning (19a) or a meaning expressing an event (19b-c).However, closer examination of the syntactic properties of argument-phrases based on event-words reveals that constructions of the type shown in (19b) are, in fact, headless relative clauses (Section 2.2.1), while those in (19c) are non-finite clauses resembling English gerunds (2.2.2): in other words, when eventwords are used in argument position, they are marked in terms of Structural Complexity, requiring us to make a distinction between these words (marked syntactic arguments -i.e., verbs) and substantives (unmarked syntactic arguments -i.e., nouns).Further evidence for the necessity of maintaining this distinction can be found in other non-criterial syntactic environments such as negative constructions (2.2.3),where the interpretation and syntactic treatment of complements of the negative predicate depends on whether the complement is a word with a substantive meaning (a noun) or a word expressing an event (a verb).This evidence strongly supports the conclusion that any flexibility in the Lushootseed lexicon that neutralizes the distinction between noun and verb applies only to the syntactic role of predicate, but is unidirectional and does not apply to the role of syntactic argument.

Headless relative clauses
In spite of the frequent appearance of event-words in argument position in sentences such as (19b) above, these constructions can be shown to be marked in the sense of being syntactically more complex than ordinary nominal arguments: they are, in fact, headless relative clauses (Beck 2002: 113-122 č'ač'as child 'I see the dog that was clubbed by the boy' (Hess and Hilbert 1976: II, 124-125) (20a) gives an example of a modifying relative clause with a third-person subject and a third-person object; in this case, the only interpretation of the sentence possible is that of a subject-centred relative clause.When the object-centred reading is desired, it is necessary to passivize the embedded clause as in (20b).The same holds for headless relative clauses such as those shown in ( 21): 'he gave us the deer which he had butchered' (Hukari 1977: 53) ii.taxʷčəɬəb sʔəɬəd ʔə tiʔəʔ diʔəʔ stawixʷʔɬ tasčəbaʔəd tul'ʔal tudiʔ čaʔkʷ tu=ʔas-dxʷ-čəɬ-əb sʔəɬəd ʔə tiʔəʔ diʔəʔ stawixʷʔɬ PAST=STAT-CTD-make-DSD food PR PROX here children tu=ʔas-čəbaʔ-əd tul'ʔal tudiʔ čaʔkʷ PAST=STAT-backpack-ICS PR over.therewaterward 'she wanted to make food of the children she carried up from the water' [DM Basket Ogress,line 73] Even in such cases, object-centred relatives are unusual, the more common pattern being for the embedded verb to be used in the passive voice.
( These headless relative clauses are obligatorily subject-centred.When expression of the PATIENT or ENDPOINT of the event is the sentence predicate, the subject-phrase appears in the passive voice, as in (23b) and (c).
The syntactic properties of event-words in argument position in Lushootseed point quite clearly to their analysis as embedded predicates contained within a relative construction, something which is quite widely accepted as evidence of structural markedness in other contexts (for instance, in the discussion of property-concept verbs in languages with re-duced classes of adjectives -e.g., Dixon 1982;Hengeveld, Rijkhoff and Siewierska 2004).Even though in Lushootseed there are no overt markers of this embedding such as special inflections or unambiguous complementizers, the added syntactic complexity of these embedded structures is manifest in the patterns of accessibility and voice restrictions discussed in the sections above.All of these properties of event-words in argument position point to their being significantly different (and more complex) constructions than a simple English noun phrase like the boy.Kinkade (1983) addresses this point by arguing, in effect, that substantive syntactic predicates like those discussed in Section 2.1 are evidence that a word like Lushootseed sbiaw 'coyote' in (4a) is the expression of the underlying semantic predicate 'be a coyote'.12Thus, according to Kinkade, all the translation-equivalents of nouns in languages like English are, in Salishan languages, the expressions of semantic predicates based on 'be'.If this is the case, then it must be true that not only are substantives predicative in constructions such as (4a), but they must also be predicates in sentences where they are syntactic arguments, as in ( 24 (Nicodemus 1975, cited in Kinkade 1983: 34) A better literal gloss for Kinkade's purposes might be 'the ones who are deer are good meat', the deictic xʷe introducing a headless relative clause formed from c'iʔ 'be a deer'.Presumably, Nicodemus would also gloss the Lushootseed sentence in (24) as 'the one who is a girl fed on the one who is a crab'.The facts that subjects are gapped in subject-centred relative clauses and intransitive verbs show no overt agreement with third-person subjects lend a semblance of credibility to Kinkade's position in that if tsi č'ač'as in ( 24) were a subject-centred relative clause formed on a predicate 'be a child' with a zero subject, this is the form that it would be expected to take (a determiner followed by a bare intransitive predicate-cf.ti ʔuxw 'the one who goes' in 4a) (a similar point is made in Van Eijk and Hess 1986: 324-325).Although Kinkade's interpretation of Nicodemus has had a certain intuitive appeal, it is difficult to know how seriously to take such considerations.What is needed before accepting such a radical claim -that all NPs are, in fact, syntactically relative clauses -is hard syntactic evidence, evidence which captures aspects of Salishan syntax and differentiates it from languages like English where such a position is clearly undesirable (although it has been argued for in the past -Bach 1968).So far none has been forthcoming, or at least none that can not be handled in other ways such as a DP-analysis of noun phrases (Matthewson and Davis 1995;Beck 1997), which nonetheless maintains the noun-verb distinction.In the absence of such evidence, the more parsimonious analysis is to treat argument-phrases such as tsi č'ač'as in (24) as a simple substantive preceded by a determiner, a structurally less complex (and therefore unmarked) construction than that required for event-words in the same position, which -for subject-centred and some object-centred constructions -are clearly relative clauses.

Oblique-centred constructions
Further evidence for the markedness of event-words in argument position, and the formal identity of these constructions with relative constructions, comes from the consideration of oblique-centred modifying and argument phrases.Since Lushootseed does not allow the relativization of oblique objects or adjuncts, it creates the structural equivalent of obliqueand adjunct-centred relative clauses through the formation of gerund-or participle-like constructions.Such constructions are used both as adnominal modifiers and as syntactic arguments, and have essentially the same internal syntax as matrix clauses in terms of the valency and transitivity of the embedded predicate.However, an important difference between the two clause types is that these non-finite clauses mark their subjects with the possessive series of subject-marking clitics, as shown in (26) for oblique-centred constructions formed with the proclitic s=, generally analyzed by Salishanists as a nominalizer: (26) a. xwul' čəd ɬuləʔuxwtxʷ tiʔəʔ ɬadsʔəɬtxʷ xwul' čəd ɬu=lə-ʔuxw-txʷ tiʔəʔ ɬu=ad=s=ʔəɬ-txʷ just 1SG.SUB IRR=PROG-go-ECS PROX IRR=2SG.PO=NM=eat-ECS 'I will just be taking [them] what you will feed [them] with' (lit.'I will just be taking them your future-feeding them') (Hess 1998: 58 In this example, the adverbial ti ƛ'ubəstiləbsəxʷ ƛ'ubəšəq 'its habitually suddenly being high again' contains an adverb tiləb 'suddenly' which precedes the clausal predicate šəq 'be high', and it is the adverb (rather than the clausal predicate) that bears both the possessive subject clitic and the nominalizing proclitic.Possessive affixes, on the other hand, are not mobile and remain affixed to the possessed, even in the presence of a pre-posed modifier (see the phrase siʔab dsyaʔyaʔ 'my noble friend' in 26b above, where the first-person possessive remains on syaʔyaʔ 'friend' rather than migrating to siʔab 'noble'). 14s the example in (28) shows, it is not only the possessive subject clitics that are mobile, it is also the nominalizing proclitic itself.This property differentiates it from the homophonous (and certainly cognate) nominalizing prefix swhich forms a part of a great many nouns whose etymology is transparently that of a verbal radical plus this prefix, as well as a great many more where the etymology is no longer transparent.For many nouns formed with s-, the meaning of the derived form is fairly predictable: for intransitive verbs, the s-form refers to the subject of the verbal radical (e.g., q'axʷ 'be frozen' > sq'axʷ 'ice'), while for bivalent verbs it refers to the object (xʷiʔxʷiʔ 'hunt something' > sxʷiʔxʷiʔ 'game').However, it is also very common for s-forms to have unpredictable, lexicalized meanings (e.g., xəɬ 'be sick' > sxəɬ 'sickness', xǎʔxǎʔ 'be forbidden' > sxǎʔxǎʔ 'in-laws').Even more significantly, the sprefix-unlike the s= proclitic-is not mobile and can never separate from the radical to which it is attached: (29) laʔbəxʷ haʔɬ stalxəxʷ laʔb=əxʷ haʔɬ s-talx=əxʷ really=now good NP-be.able=now'now he is really a very capable one' (Hess 2006: 40, line 461) Note also that s-forms do not require an expression of a possessor, whereas nominalizations with the s= proclitic always appear with a possessive clitic expressing their subject.
In addition to the proclitic s=, Lushootseed also has a second proclitic, dəxʷ=, which forms non-finite clauses with essentially the same syntactic properties as those formed with s=, including the use of possessive subject clitics to express their subjects:  (Hess 2006: 66, line 592) shows the first-person singular proclitic d= marking the subject of the non-finite clause cəxʷuʔibəš 'where I travel'.The next example in (30b) contains a non-finite clause with a second-person plural subject, which is in turn contained within an prepositional phrase acting as a locative adverbial modifier.The distinction between s= and dəxʷ= is, roughly, that s= forms the equivalent of oblique-centred relative clauses, whereas dəxʷ= is used with adjunct-centred expressions referring (among other things) to locations, motives, and instruments.
When the subject of either type of non-finite clause is third-person, it shows the same patterns as the expression of the third-person possessor, using the possessive subject enclitic =s if there is no overt subject NP, otherwise making use of a periphrastic possessive construction: (31) a. xwul ' p'aƛ'aƛ' tiʔiɬ  Again, here we see the use of the subject enclitic =s when there is no overt subject NP present (32a), and the periphrastic construction with ʔə used with an overt NP (32b).
A second function of the s=nominalizer is to form sentential nominals (Beck 2000), non-finite clauses whose reference is the event rather than a particular event-participant.Compare the non-finite clauses in (31) with those in ( 33 The non-finite clauses in these examples refer to entire events -the coming of the water in (33a) and the making of a noise in (33b).In neither case is the reference of the non-finite clause an argument of the verb in the nominalized clause.
As it turns out, the interaction of the s=proclitic and words with a substantive meaning offers some evidence against the analysis of NPs as relative clauses and, as such, helps to establish the distinction between verbs and nouns.Consider the sentences in ( 34): (34) a. ɬustitčulbixʷ čəxʷ ɬu=s=titčulbixʷ čəxʷ IRR=NM=small.animal2SG.SUB 'you are the one who will be a small animal' (Hess 2006: 8, line  Recall that Kinkade (1983) claims that all phrases in Salishan languages are in fact relative clauses, making an expression like tiʔiɬ titčulbixʷ 'the small animal' in (34a) more literally 'the one who is a small animal' -that is, a subject-centred relative construction based on a monovalent predicate 'be a small animal'.Similarly, the form skikəwič in (34b) (based on the proper noun kikəwič 'Little Hunchback') would also seem to correspond to a subject-centred relative clause.If were the case, however, then the occurrence of the proclitic s= with such words should, it does with words expressing events, result in a sentential nominalization in which the subject is expressed as a possessor (meaning something along the lines of 'his being a small animal' or 'his being Little Hunchback').Yet in the constructions in (34) the subject of the putative s=nominal is in fact not expressed at all; instead, such constructions seem to be interpreted as subject-centred relative clauses, which for event-words do not require the proclitic s=.Since s= is usually reserved for the "relativization" of arguments that are not part of a predicate's core valency (i.e., not the subject or direct object), the obvious conclusion is that the subjects of substantive predicates are not in fact part of their core valency, which is consistent with the idea that substantives have a semantic valency of zero.This is fairly good evidence against the proposal that NPs are underlying relative clauses formed on 'be an X' type semantic predicates, given that such an analysis predicts that substantives should pattern in the same way as intransitive expressions of events and form full non-finite clauseswhen affixed with the s= proclitic. 15 Although s= and dəxʷ=clauses are not relative clauses per se, they represent the same type of structural markedness that headless relatives do -they are phrasal or clausal syntactic units and so count as being structurally complex.Further evidence for the markedness of these constructions of a different kind can adduced from the fact that they undergo a certain degree of recategorization (Bhat 1994) (also "recategorialization" -Hopper and 15 An alternative analysis is that the /s/ here is not the nominalizing proclitic but rather the nominalizing prefix, s-, in which case a more accurate gloss of the forms in (34) might be something along the lines of 'the small-animal being' (34a) and 'the Little-Hunchback being' (34b).Even if this proves to be the case, the substance of the argument remains the same: the formation of sentential nominals is blocked for substantive predicates but allowed for words expressing events.Thompson 1984) in that, by expressing their subjects as possessors, they take on some of the inflectional properties of the part of speech that is unmarked in the same syntactic role -that is, they become more like nouns (cf.Van Eijk and Hess 1986).Recategorization (and its counterpart, decategorization) falls under the heading of Contextual Markedness (Beck 2002: 23), and constitutes a clear example of what Hengeveld (1992aHengeveld ( , 1992b) ) would classify as a "further measure."The fact that the morphosyntactic properties of event-words in Lushootseed in argument position more like those of substantives seems to be strong evidence of the link between this syntactic role and the latter class of words -that is, evidence of the existence of a class of nouns in the Lushootseed lexicon.

Negative constructions
Once the existence of a lexical class distinction has been established by the examination of criterial syntactic environments, it is almost always the case that the distinction can also be shown to be present in other constructions as well.In Lushootseed, for instance, nouns and verbs can be shown to be clearly distinct in the context of negative constructions headed by the impersonal negative predicate xʷiʔ 'it is not, there is no'.This predicate is used to negate the existence or reality of an object or the realization of an event, and can take either a noun or a verb as its complement. 16When the complement is a noun, the expression has the reading 'there is no' and the nominal complement appears with the subjunctive proclitic, gʷə=:17  When the complement is a verb, the subjunctive proclitic also appears, but the verbal predicate is obligatorily nominalized with the proclitic s=: (36) a. xʷiʔ uʔxʷ gʷəsɬaʔ ʔə tiʔəʔ čaləs xʷiʔ uʔxʷ gʷə=s=ɬaʔ ʔə tiʔəʔ čaləs-s

NEG PTCL
SBJ=NM=arrive PR PROX hand-3PO 'his hand still can not reach it' (lit.'there is no his hand's reaching it') (Hilbert and Hess 1977 As with all of the other distinctions discussed in this section, this difference in the treatment of the two classes of words is categorical and requires reference in the syntax to a wordclass designation that maps a set of syntactic behaviours onto specific items in the lexicon -in other words, a designation which constitutes a part-of-speech distinction.The fact that the semantic makeup of one of the classes corresponds almost exactly with the semantic category of substantives and the other contains those meanings belonging to the semantic category of events points us squarely to the conclusion that the distinction is one that any typologically-responsible analyst would characterize as one between nouns and verbs.

Unidirectional flexibility in the Salishan lexicon
As seen in the preceding sections, Lushootseed shows some flexibility as to the treatment of substantives versus event-words in its syntax, but this flexibility is unidirectional and only applies in predicate position.In argument position, event-words are either contained within headless relative clauses or are recategorized as non-finite argument phrases with some of the morphological properties of nouns (specifically, that they realize their subjects as possessors).The same type of recategorization applies when substantives and eventwords are found in certain types of complementizing constructions such as negatives.In these cases, as in the case of non-substantive arguments, the syntax of Lushootseed makes reference to the lexical class of a word in determining its syntactic treatment: words that express substantive meanings appear in syntactic argument position without further measures, whereas words expressing events must be contained in some sort of embedded (headless relative or non-finite) clause when used as syntactic arguments.The fact that this syntactic distinction groups words into the semantic classes that it does (substantive versus event-word) shows quite clearly that Lushootseed makes a robust lexical class distinction between noun and verb.Of course, this is not to say that the Lushootseed part-of-speech system is precisely the same as the traditional Indo-European system, nor that nouns and verbs behave in Lushoot-seed exactly as they do in English or Latin.Like other Salishan languages, Lushootseed quite freely allows nouns and other non-verbal elements to serve as syntactic predicates without requiring the use of a copula, thereby neutralizing lexical class distinctions between verbs and nouns in predicate position.The situation can be summarized as in Fig. 8 Thus, the flexibility displayed by Lushootseed is particular only to one of the relevant criterial syntactic positions, but the noun-verb distinction can by no means be said to be absent from the Lushootseed lexicon or irrelevant to Lushootseed syntax.Rather, the distinction between verbs and nouns is not relevant to the behaviour of either class of word in predicate position -but it is relevant in argument position, as it is in languages like English and others that are said to have a "rigid" distinction between noun and verb.This means, of course, that Lushootseed is like English in that it does distinguish between nouns and verbs as they are defined by Hengeveld (1992aHengeveld ( , 1992b)), but that it differs from English in that it shows flexibility with respect to the behaviour of nouns in predicate position.This is the type of unidirectional flexibility that Evans and Osada (2005) point to as a frequent source of claims for the absolute neutralization of the noun-verb distinction.As they correctly observe, such flexibility is a necessary but not a sufficient condition for the claim that a particular language does not distinguish the two lexical classes.

Omnipredicative and precategorial languages
The type of unidirectional flexibility shown by Lushootseed (and, to the best of my knowledge, other Salishan languages as well) is a typical case of what Evans and Osada (2005) refer to as the "omnipredicative" pattern of putative noun-verb flexibility -that is, languages where all open-class lexical items are claimed to be semantic predicates with a minimum syntactic valency of one, those words with substantive meanings expressing predications of the type 'be an X'.As argued in the preceding sections, there is no positive empirical evidence for such a claim in Lushootseed, and some syntactic evidence against it.Taking the opposite point of view, that Lushootseed does have a noun-verb distinction and argument phrases headed by substantives are not syntactic predications (that is, they are not headless relative clauses), allows for a satisfactory account of the behaviour of open-class words in criterial and non-criterial syntactic positions without recourse to needless exoticisms in the syntax or the structure of the lexicon.Given that the facts in other languages where similar claims of omnipredicativity have been made (e.g., Nootkan -Swadesh 1939, Nahuatl -Launey 1994) look, to the non-specialist at any rate, to be substantially the same, it would seem that the case for this type of language representing a true case of bidirectional noun-verb flexibility does not stand up to careful scrutiny.
There is, of course, a second logically-possible type of language that neutralizes the syntactic distinction between words with substantive meanings and those that express events.In these languages, rather than substantive interpretations of 'be an X' semantic predicates being forced by syntactic context, it is the substantive interpretations of eventwords that are context-dependent.Languages that follow this pattern fall under Evans and Osada (2005)'s heading of precategorial languages.Rather than subdividing lexical items into classes based on their syntactic behaviours and distributions, precategorial languages organize their lexicon around roots that are not specified for any particular syntactic distribution, and thus do not conform to any of the definitions of parts of speech offered in (1) above.The meanings of these roots are often claimed to be "vague" (Hengeveld et al. 2004;Hengeveld and Rijkhoff 2005) (Tchekoff 1981: 4, cited in Hengeveld 1992b: 66) This data shows the root siʔi in a variety of contexts -as a syntactic predicate (37a), as the complement of a preposition (37b), as an adnominal modifier (37c), and as an adverbial (37d).These represent all four of the criterial syntactic contexts listed in (1) and so, given the absence of obvious morphosyntactic differences between the instances of siʔi in the various contexts, it is claimed that this root conforms to the definition of all four lexical classes. 18n obvious objection to this interpretation of the data is, of course, that siʔi does not mean the same thing in each of these four contexts (Croft 2000;Vonen 2000;Beck 2002).In (37a), (37c), and (37d), siʔi expresses a semantic predicate, something like 'be small' or 'be of reduced scale or intensity '; however, in (37b), siʔi expresses a more substantive concept, 'childhood' -that is, 'stage of human development during which a person is mentally and physically immature (and therefore small)'.While clearly not random, the exact semantic relationship between the two meanings of the root is not transparent or predictable, but rather is reminiscent of the relationship between homophonous pairs of English words such as hammer N vs. hammer V or cook N vs. cook V , which are generally held to be distinct lexical items related by a process of conversion (Mel'čuk 2006).Conversion posits the existence in the lexicon of homophonous but semantically-related words whereby each has a particular meaning associated with its use in a particular syntactic role or roles. 19The semantic relationship between members of a conversive pair is not arbitrary but at the same time is not entirely predictable and is established by linguistic convention, requiring the speaker to learn and enter into the mental lexicon the particular meaning of a root associated with its use in a particular semantic role.Evans and Osada (2005) make a similar observation for a large number of noun-verb pairs in Mundari and suggest that this pattern is evidence that precategorial languages are those that make extensive use of conversion in the lexicon, the distinct meanings of words like siʔi in (37) in fact constituting different lexical items (see also Vonen 2000).Hengeveld and Rijkhoff (2005) counter this argument by claiming that data like that in (37) are evidence that roots of this type have "vague" meanings which become specified only once the root appears in a given syntactic context.Rijkhoff (2008) illustrates this idea using Fig. 9 (Mosel andHovdhaugen 1992: 80, 73, 74, cited in Rijkhoff 2008: 729) 19 It should be pointed out that the claim of conversion is not (as suggested by some common misnomers for "conversion"-"zero conversion" and, worse, "zero-derivation") a claim that there is some sort of classchanging zero affixation.Zero affixation requires an explicit formal contrast between a zero and a non-zero exponent of the same category, and it is precisely the nature of conversion that there is none.Furthermore, the existence of zero derivational elements is, under any viable theory of morphological  According to Rijkhoff's proposal, the meaning of the root lā consists of a set of components or features which comprise the sum total of the components of the meanings of its contextualized use.Speakers learn to associate certain subsets of the semantic components of lā with its appearance in particular semantic slots, giving rise to the more specific meanings that are lexified as different words in languages such as English.
In a sense it is certainly true, even under the conversion analysis, that the different context-bound meanings of precategorial roots share certain semantic elements and could, possibly, be shown to be extensions or subsets of a single, abstract schematic meaning; however, it is not clear how this impacts on the question of parts of speech.Hengeveld and Rijkhoff (2005) imply that a schematic relationship between the meanings of two signs with homophonous signifiers is sufficient for their treatment as the same lexical item.Nevertheless, the fact remains that the context-bound meanings of precategorial roots are quite specific, and that speakers must learn and memorize which specific sub-schematic meaning of the precategorial root (or which subset of its semantic components) is associated with which particular syntactic context. 20Even for a carefully chosen example like lā, where the context-specific meanings of the root seem almost to be predictable (assuming that these meanings are exactly the same as their English glosses), it remains the case that speakers must learn and memorize the fact that, for instance, in the "head of 'NP'" slot lā expresses the meaning components B D E (= 'sun') and not A D E (say, 'sunshine').
The issue of predictability of the meanings of roots in particular semantic slots is discussed extensively by Evans and Osada (2005: 367-375) for another putative precategorial language, Mundari, under their heading of "compositionality."For Evans and Osada, a pair of words such as lā 'be sunny' and lā 'sun' can only be considered instances of the same lexeme if the difference in their meaning as it is manifest in two different syntactic slots is predictable as a function of that syntactic environment.Even a small shift such as 'sun' to 'be sunny' (that is, 'environment is illuminated brightly by the sun') seems not to be entirely predictable (why does the predicative use of lā not mean 'be a sun'?).However, the problem becomes even more acute for a root like the Tonga siʔi in (37) which in its predi-cative and modificative function means 'small' and in its argument function means 'childhood' (i.e., 'stage of human development during which a person is mentally and physically immature').In this case, the meaning of the precategorial root would have to include not only a component meaning 'small' but also all of the components that make up the meaning of the much more complex notion of 'childhood'.Furthermore, a speaker would have to learn (and store) the information that only the meaning component 'small' is associated with its use in predicative and modificative syntactic roles, and that the meaning 'childhood' (and not 'child' or 'short person' or any other conceivable recombination of the semantic component-set comprising the union of the components of 'small' and 'childhood') is associated with its use in syntactic argument roles.
Speakers may (or may not) be aware of the semantic relationship between the two meanings of the precategorial root, but the fact of the matter is that the speaker's knowledge of siʔi must include a learned pairing between a particular meaning (sub-schematic or not) and a syntactic distribution.And this pairing of meaning and unmarked distribution is what is generally understood by "part-of-speech."The abstract schemas or vague meanings underlying the networks of related words may constitute part of the lexicon of a precategorial language (as may the schemas shared by non-homophonous forms related by overt derivation and other word-formation processes); however, it is not the precategorial roots that meet the definitions of lexical classes shown in (1), it is the context-bound uses of the roots.Like most approaches to parts of speech, the definitions in (1) are based on the premise that parts of speech define high-level taxonomic groupings of lexical items (i.e., lexical signs that pair a signifier and a particular signified) according to their (unmarked) syntactic distribution.For a language like Tongan, such definitions fit quite naturally if we recognize that the term "lexical item" applies separately to each member of a pair like siʔi 'be small'/siʔi 'childhood', rather than to an under-specified, abstract root siʔi, which seems more than anything else to define the semantic domain of the word-family.Like the analysis of omnipredicative languages suggested above, this approach effectively removes precategorial languages from the putative group of languages that show bidirectional nounverb flexibility, and avoids treating the lexicon in such languages as being overly exotic, instead showing them to be extreme examples of a lexicon built on the cross-linguistically well-attested process of lexical conversion.
Seen from this perspective, the type of noun-verb flexibility manifested by omnipredicative languages and that seen in precategorial languages represent markedly different phenomena.In the former case, words with specific meanings fall into clear categories based on their morphosyntactic behaviour, but rather than showing bidirectional patterns of markedness in both of the criterial syntactic roles for nouns and verbs, the class of verbs shows itself to be marked in the role of syntactic argument, while both nouns and verbs are unmarked syntactic predicates.In precategorial languages, homophonous lexical items appear in various syntactic roles, but have divergent (albeit not always unrelated) meanings associated with their different uses.The homophonous lexical items undoubtedly form related sets and so may be linked to an abstract precategorial schema, but the sets are in a sense "pre-lexical," representing a more abstract level of the lexicon than has traditionally been of interest to syntacticians and lexicographers.Irrespective of the internal structure of these sets, it is the individual members (that is, the distributionally-specified meaning-form pairs) that constitute the genuine parts of speech in such languages, making the precate-gorial language -like the omnipredicative language -a false case of noun-verb flexibility.
Thus, both possible types of noun-verb flexibility proposed by Evans and Osada (2005) seem (as they suggest) not to stand up to careful consideration.This is an unsurprising result, given the widely-held position among typologists that the noun-verb distinction is one of the best candidates we have for a genuine universal of human language (Croft 2003).Naturally, this raises a question with respect to the status of the typology of parts-of-speech systems in Fig. 1 and the implicational hierarchy in (2), given that both predict, or at any rate allow for, the existence of languages that do not distinguish verbs and nouns.While this may not be a terribly serious shortcoming of the typology, as it could simply be stipulated that all languages make at least the first-order distinction on the hierarchy and the Type 1 language could then be removed from the typology, the fact that we need to make such a stipulation at all seems significant, and lays the groundwork for future investigation into the origins and motivations of this restriction on the logically-possible range of variation in human languages.
A second, and perhaps more serious, objection to the proposed typology that seems to fall out from the analysis of the data presented here has to do with the relative rankings of the different word-classes in the hierarchy, in particular with respect to the rankings of noun and verb.Even if it is the case that no language fails to distinguish between nouns and verbs, the way that the hierarchy is presently constructed characterizes languages that distinguish only two lexical classes as making a two-way distinction between verbs and everything else.However, the distributional evidence from both Salishan and Tongan seem to point us in the opposite direction: that languages with only two lexical classes distinguish between nouns and everything else.In Salishan, this is manifest in the pattern of distributional markedness.Nouns are syntactically privileged in that they have a syntactic role (syntactic argument) that is not open without further measures to other lexical classes, effectively subdividing the lexicon between words which are marked and unmarked syntactic arguments, rather than between words that are marked and unmarked syntactic predicates, as predicted by the typology in Fig. 1. 21In the Tongan case, assuming that the pattern shown by siʔi is typical of all precategorial roots in the language, the two interpretations of the root are divided along similar lines -the substantive interpretation is restricted to "nominal" syntactic roles and the predicative interpretation is found elsewhere, once again dividing the lexicon between words (or interpretations of precategorial roots) which are without-further-measures syntactic arguments and words which are found without further measures in the other criterial syntactic environments.Thus, it seems that the crosslinguistically attested patterns of flexibility in lexical classes favour a typology, like those argued for in Dixon (1982) and Beck (2002), that allows for the flexible grouping of nouns against verbs, adjectives, and adverbs rather than verbs against a potentially-conflated class of nouns, adjectives, and adverbs.From a semantic perspective, this makes a great deal of sense, given that verbs, adjectives, and adverbs are expressions of semantic predicates and share the property of having non-zero syntactic valency, as opposed to nouns which in Langacker's (1987) terms are "conceptually autonomous" and generally have a syntactic (and semantic) valency of zero.If this pattern turns out to be cross-linguistically robust, it constitutes an important finding, as it offers us an example of the influence of the semantic structure (as opposed to the content) of meanings on the organization of their expressions in lexicon.
A final implication of this study for the approach to parts-of-speech typology being discussed here concerns the issue of the directionality of lexical flexibility.Although the principle of bidirectionality was proposed by Evans and Osada (2005) as a test for determining whether or not a part-of-speech distinction is truly absent in a language, it also sheds light on an important potential difference in types of lexical flexibility, highlighted by the contrast between Fig. 3  While flexibility between lexical classes has often been assumed to entail bidirectionality, languages like Lushootseed provide us with a clear example of a different, unidirectional type of flexibility.Unidirectional flexibility does not equate with the absence of a lexical class distinction, but it does correspond to the neutralization of that distinction in one (or more) of the criterial syntactic positions.This type of neutralization is a relative commonplace across languages, and seems like a good candidate for inclusion as a parameter for a comprehensive typology of parts of speech systems.The patterns shown by unidirectional systems and the ways in which they parallel and depart from the patterns observed for bidirectionally-flexible systems seem sure to inform parts-of-speech typology and will advance the cause of understanding the parameters of variation open to human languages.

Figure 5 :
Figure 5: Proto-typical associations of meaning-type and lexical classes

(
21) a. wiw'su tiʔəʔ ʔučalad tiʔəʔ sqʷəbayʔ wiw'su tiʔəʔ ʔu-čala-d tiʔəʔ sqʷəbayʔ children PROX PFV-chased-ICS PROX dog 'the ones who chased the dog are the children' *'the ones who the dog chased are the children' b. sqʷəbayʔ ti ʔučalatəb ʔə tiʔiɬ wiw'su sqʷəbayʔ ti ʔu-čala-t-əb ʔə tiʔiɬ wiw'su dog SPEC PFV-chased-ICS-PASS PR DIST children 'the one who is chased by the children is the dog' (Hess 1995: 99) In (21a), the only interpretation open to the sentence is the one where the headless relative clause identifies the subject of the embedded verb, in spite of the fact that the opposite interpretation, where the dog chases the children, is semantically and pragmatically quite plausible.Again, where this is the desired interpretation, the embedded clause appears in the passive, as in (21b). 11When the subject of the embedded clause is first-or second-person, only object-centred relative clauses are possible: 11 When discourse context leaves no room for ambiguity as to the syntactic roles of the third-person arguments of the verb in the embedded clause, object-centred relatives and headless relatives are possible: i. tuɬiltubuɬ ʔə ti sqigʷəc tuq'ʷəxwəd tu=ɬil-txʷ-buɬ ʔə ti sqigʷəc tu=q'ʷəxw-əd PAST=give.food-ECS-1PL.OBJ PR DEF deer PAST=butchered-ICS

Figure 8 :
Figure 8: Unidirectional flexibility between nouns and verbs

Figure 10 :
Figure 10: Types of flexibility (cf.Croft 1991): ) by =əs '3 subjunctive': ): 2SG.SUB stone 'you will bring two stones' (lit.'thestones that you will bring will be two') [AW Basket Ogress, line 80] (16) a. tučadəxʷ čəxʷ tu=čad=əxʷ čəxʷ PAST=where=now 2SG.SUB 'where have you been?' b. tul'čad čəxʷ tul'-čad čəxʷ CNTRFG-where2SG.SUB 'where are you coming from?'/'where are you from?'(Bates et al. 1994: 59)Kroeber (1999) also notes that many Salishan languages, including Lushootseed, allow prepositions to head clauses, as in (17):(17) a. dxʷʔal tə hud tə sxʷit'il ʔə tə biac . Although bare nominal predicates are allowed in such languages, the distinction between nouns and verbs is rarely in doubt once other types of evidence are taken into consideration.In Lushootseed, the crucial evidence is found by examining the syntax of argument phrases: of the open lexical classes, only words with substantive meanings are unmarked syntactic arguments.Consider the following: 'and so the one who (can) carry that will be (my) wife'[MW Star Child, line 77]

sqʷəbayʔ ʔuč'axʷatəb ʔə tiʔiɬ č'ač'as
Keenan and Comrie 1977)or this comes from the restrictions on accessibility to relativization (in the sense ofKeenan and Comrie 1977)that hold for both nominally-headed and headless relative clauses: all else being equal, in clauses with a thirdperson subject and a third-person object, only the subject can be relativized: *'we saw the boy that the man hit' b. ʔušudxʷ čəd ti

ti ʔut'uc'utəb ʔə tiʔiɬ šəbad
This is almost certainly a syntactic restriction, as the subject-markers are not themselves nominals and so cannot head an NP or be modified by a relative clause.There are no examples of first-or second-person pronouns heading a relative clause construction, but there are numerous examples of pronouns functioning as predicates of sentences with headless relative clause subjects.In these cases, the pronoun and the headless relative express the same event-participant, and the predicate of the embedded clause is in the third-person: Crucial to this line of reasoning is the fact that in Lushootseed, as in most Salishan languages, nominal arguments are almost invariably introduced by determiners, as are headless relative clauses.13Thus,under Kinkade's analysis, each of the argument phrases in (24) would actually be the equivalent of a relative clause, the determiner in reality being a complementizer.According to Kinkade, Lawrence Nicodemus, a native speaker of Coeur D'Alene Salish with some linguistic training, regularly glosses argument phrases as relative clauses, as in: (32) a. ƛ'al' bədiɬ dəxʷʔa ʔə tiʔiɬ dəxʷʔəy'dubs ʔə tiʔiɬ sgʷəlub ƛ'al' bə=diɬ dəxʷ=ʔa ʔə tiʔiɬ dəxʷ=ʔəy'-dxʷ-b=s also ADD=FOC ADNM=be.therePR DIST ADNM=find-DC-PASS=3PO (31b) the subject is an overt NP, tə wiw'su 'the children', and so the periphrastic possessive construction with ʔə is used.The same two patterns are also observed with dəxʷ=constructions: and to remain indeterminate between substantive or eventreadings until they appear in a particular syntactic context.This situation is illustrated by the following by-now-familiar examples from Tongan: , which is based on the Samoan data in (38):