Semantic transparency and Oneida morphological parts of speech

1 Introduction

North American languages have figured prominently in discussions about parts of speech. Northern Iroquoian is no exception and much of the discussion has focused on whether those languages have a noun/verb distinction (Mithun 2000; Sasse 1988, 1993) or whether it has an adjective category (Baker 2003; Chafe 2012). In this paper, we examine more closely the evidence for morphological parts of speech in one Northern Iroquoian language, Oneida, and tackle the issue from a slightly different perspective by asking what semantic properties underlie its morphological parts of speech (see Evans 2000b for the distinction between morphological and syntactic parts of speech). We will make two claims. First, inflected parts of speech in Oneida are semantically transparent in that a part of speech can be determined based solely on well-established semantic distinctions. Oneida parts of speech are, in this respect, canonical in the sense of Spencer (2005) and Corbett (2012). Second, Oneida morphological parts of speech are organized along two orthogonal dimensions, an ontological dimension, i.e., the sort of entity that members of the category describe, and a semantic type dimension, i.e., the kind of semantic relation conveyed by members of the category. We show that Oneida inflection is sometimes sensitive to distinctions along the ontological semantic dimension, sometimes to distinctions along the semantic type dimension, and sometimes a single inflectional process is sensitive to distinctions along both dimensions at the same time. Put in Discourse Representation Theory terms (Kamp and Reyle 1993), Oneida morphological processes sometimes only apply to lexical items that are used to introduce discourse referents of particular sorts and sometimes only to lexical items whose predicative conditions are of a particular semantic type.

The paper is organized as follows. Section 2 provides some general information on Iroquoian languages and Oneida. Section 3 briefly introduces some of the traditional analyses of parts of speech in Iroquoian. Section 4 reports our study of the ontological distinction that underlies parts of speech in the dictionary of Michelson and Doxtator (2002). Sections 5 and 6 show that Oneida inflection can be sensitive to either ontological sort or semantic type. Section 7 shows that the two dimensions are indeed orthogonal. Section 8 considers more general implications of the semantic transparency of the distinction between nouns and verbs in Oneida and shows that the bi-dimensional semantic analysis of part-of-speech distinctions explains the mixed inflectional behavior of kinship terms that Koenig and Michelson (2010) had to stipulate.

2 Background

The Northern Iroquoian languages Oneida, Mohawk, Onondaga, Cayuga, Seneca, and Tuscarora are spoken by the Haudenosaunee (People of the Longhouse), a term that refers to the nations that comprise the Haudenosaunee Confederacy or League of the Five [Six] Nations. Other Northern Iroquoian languages include Huron-Wyandot and a few lesser known languages such as Laurentian and Susquehannock or Andaste. Cherokee is the sole Southern Iroquoian language. The Northern Iroquoian languages were spoken mostly in New York State and in the provinces of Quebec and Ontario in Canada. Oneida is presently spoken at three locations, by first and second language speakers at the Oneida Nation of the Thames in southwestern Ontario, and only by second language speakers at the Oneida Nation near Syracuse, New York and at the Oneida Nation of Wisconsin. At recent count there are only 38 fluent speakers at the Oneida Nation of the Thames who learned Oneida as a first language.

Nonetheless, Oneida is very well described, especially for a First Nations or Native American language, beginning, most importantly, with Lounsbury’s (1953) influential description of the Oneida verb, the most complex of the three traditional parts of speech (the others being nouns and uninflected particles).

There are comprehensive dictionaries of both the Ontario (Michelson and Doxtator 2002) and Wisconsin (Abbott 1996) varieties. Grammatical descriptions are Abbott (2000) and Part III of Michelson et al. (2016), which concentrates on collocations that involve more than a single word. The bulk of Michelson et al. (2016) is a substantial collection of 52 transcribed and translated texts recorded over the last 35 years by fluent speakers at the Oneida Nation of the Thames in Ontario. A large corpus of texts was collected from speakers and written down by other speakers at the Oneida Nation of Wisconsin in 1939–40 during the Works Progress Administration. Many of these were edited by Clifford Abbott and reproduced in several volumes by the Oneida Nation of Wisconsin (Abbott 1982–1983).

Most of the examples in this paper were gleaned from the text collections just mentioned. Extracts are reproduced exactly as in the original, including any punctuation (or lack of it) when the extract is only part of an utterance. For extracts from Michelson et al. (2016), the speaker’s name is identified after the extract, followed by the page number where the extract can be found, and the utterance number. Extracts from the Wisconsin texts are identified by the speaker, the volume title, and the page number.

3 Traditional analyses of parts of speech in Iroquoian

Words in Oneida are traditionally assigned to three categories, nouns, verbs, and particles (see, for example, Lounsbury 1953). Baker (2003) suggests Mohawk includes also a class of adjectives but see Chafe (2012) for cogent arguments against this view. Morphological nouns, morphological verbs, and particles are exemplified in (1). Particles are in normal font; verbs are bolded; nouns are italicized; kinship terms are both bolded and italicized (we distinguish kinship terms, as Koenig and Michelson 2010 suggests that kinship terms is a fourth kind of morphological part of speech that is both nominal and verbal using the theory of mixed categories proposed in Malouf 2000).^[1]

We summarize below the traditional morphological inflectional templates that is standardly used to justify this tripartite classification across Iroquoian languages (Chafe 1967; Koenig and Michelson 2010; Lounsbury 1953; Mithun 2000), abtracting away from any details for now. A detailed analysis of these criteria will be provided in the following sections.

(2)

Verb stems are inflected with aspect suffixes such as the punctual (pnc) in (3), and with pronominal prefixes, which we say more about in Section 5. Basically, portmanteau-like prefixes reference two animate arguments, and Agent and Patient prefixes reference one (or zero) animate arguments. Verb bases can also include derivational affixes such as the benefactive (ben) or the reflexive (refl).^[2] Finally, verb stems can bear prepronominal prefixes. Examples of words based on verb roots are given in (3).

The structure of nouns is represented in (4), following Koenig and Michelson (2020a). Noun stems are inflected with noun prefixes (3z/n.sg.p; see (5a)) or possessive prefixes – which are a subclass of Patient prefixes – (poss, as in (5b)), and (most of the time) with noun suffixes (nsf; see (5a) and (5b)). Some noun stems can have locative endings instead of a noun suffix (loc; see (6)).

(4)

Particles are morphologically inert in that no derivational or inflectional process can apply to particles; some examples of particles are given in (7).

The preceding discussion suggests that there is a robust noun/verb morphological distinction in Oneida as the distribution and properties of affixes seem to converge on two classes typical of lexical classes labeled noun and verb across languages. But, the schematic templates in (2) and (4) are simplified, and possibly misleading, in that they miss important aspects of lexical classes in Oneida and how they differ from better known examples of verb and noun classes.

We demonstrate this by approaching the organization of lexical items in Oneida not from (supposed) convergence of derivational potential and inflectional categories (and their particular exponents) but from two dimensions that involve meaning: ontological sort and semantic type. By ontological sort we mean the kind of entity a lexical item describes. By semantic type we mean the kind of relation its meaning includes, in particular whether a lexical item is semantically polyadic and if it is, whether or not it involves a relation of possession or the name of a body part. These two semantic dimensions are orthogonal in that each targets a set of derivational and inflectional processes that do not converge on the two classes labelled traditionally verb and noun. Thus the unidimensional classification of morphologically active lexical items into verb and noun misses the unique (non-overlapping) roles that these two semantic dimensions play in Oneida morphology. It also misses the fact that whether an inflectional lexical item is a noun or a verb can be inferred from its semantic properties, or, as we argue, that Oneida morphological parts of speech are semantically transparent.

4 The ontology of morphologically active lexical items in Oneida

That parts of speech are canonically associated with semantic categories is hardly news. It dates back to Aristotle (On Interpretation, 16a, 2–3); it is part and parcel of traditional grammars and is taught to students in many countries. It is also part and parcel of many approaches to parts of speech as argued by canonical typologists (Corbett 2012; Spencer 2005). It has also been suggested that canonicity might help children learn lexical categories (Gentner 1982; Pinker 1984). But most linguists, including those just cited, assume that typical part-of-speech categories are not canonical. Canonicity might aid learning part-of-speech categories, but in order to attain the adult grammars of most languages, children need to be disabused of canonicity. On the other hand, some linguists have argued even for languages like English (see, for example, Langacker 1987 or Chaves 2013) that part-of-speech categories are reducible to semantic categories. But it is of note that these scholars have appealed to rather subtle and non-standard semantic distinctions. Langacker (1987), for example, puts forward a notion of region in space, which does not correspond to any independently justified and agreed upon semantic category. Thus, whatever one thinks of the semantic underpinning of English parts of speech, Oneida’s parts of speech are remarkable for their much simpler semantic transparency. It is this transparent and canonical semantic underpinning, and the inflectional processes that are sensitive to it, that this and the next section are devoted to. Typologically oriented studies such as Croft (2001) or Haspelmath (2012) associate with parts of speech more standard semantic categories, but parts of speech in their analyses are comparative concepts (see Dryer 1997: 116–117 for the distinction between word classes “in particular languages” and “in a crosslinguistic sense”). In such typological work, semantic categories are correlates of crosslinguistic or comparative uses of the labels noun and verb, not correlates of particular languages’ parts of speech. What we are arguing for in this paper is that the semantic correlates of morphological part-of-speech distinctions in Oneida – a particular language – are standard semantic distinctions.

The first semantic correlate of morphological parts of speech in Oneida concerns the sorts of entities they describe. We assume that individuals in the universe of discourse are sorted into a richer set of sorts than customary in traditional logical approaches to the semantics of natural languages, a stance that is shared with much work in semantics since the early 80s (see among others, Asher 1993; Bach 1986b; Carlson 1977; Chierchia 1984).^[3] Each sort corresponds to a category of entities, some of which are assumed to be part of the world, and some may be abstract in that they are objects of thought and discourse but not necessarily distinct categories of entities in the world (Asher 1993).

Our ontology remains agnostic as to which sorts of individuals belong to the world and which only to our thinking and talking about the world, i.e., they may only be part of natural language metaphysics to borrow an apt term from Bach (1986a). To assess the ontological categories that correspond to the label noun and verb in Oneida (i.e., lexical items that fit the traditional schematic templates in (2) and (4)), we retrieved from a comprehensive dictionary of Oneida (Michelson and Doxtator 2002) the list of underived lexical items that fit the patterns (2) and (4). These are entries in the dictionary labeled n. and entries labeled v.a., v.s., or v.m. (see the next section for a definition of these terms). We then retrieved the list of derived entries in the dictionary, i.e., those labeled v.>n. and V > N. Lower case n. stands for noun roots and noun stems in Michelson and Doxtator’s dictionary. Similarly, v.a., v.s., or v.m. stands for verb roots or verb stems. Michelson and Doxtator (2002) also includes two kinds of derived noun entries labeled v.>n. and V > N, respectively, depending on whether the derived nouns are derived from verb roots or stems and can take some nominal inflection or are derived from fully inflected verbs and carry no nominal inflection. We classified each set of entries according to the sort of individuals they describe. The sorts associated with underived and derived entries are discussed in turn.

5 Ontological sorts and morphological parts of speech

We provided in (2) and (4) the internal structure of verbs and nouns in the traditional morphological approach to lexical parts of speech in Oneida. In this section we analyze in more detail the morphological processes underlying these traditional structures. We show that the morphological processes involved in building words that fit (2) and (4) match the classification of sorts of entities Section 4 focused on, i.e., either situations or objects.

Table 1 lists inflectional properties that are sensitive to the sort of entity or description associated with the meaning of entries they target.

The five inflectional properties in Table 1 target lexical items exclusively on the basis of the ontological sort of what they describe. By target we mean that the processes constrain or apply to a set of lexical items. Some of these properties target lexical items that describe situations, and some of these properties target lexical items that describe objects. Before discussing these, we describe in a little more detail pronominal prefixes in Oneida, as their properties are relevant to the inflectional constraints we discuss in this paper.

All inflecting lexical items, be they nouns or verbs, bear pronominal prefixes in Oneida. There are three kinds of pronominal prefixes: so-called Transitive prefixes and so-called Agent and Patient (Intransitive) prefixes. We capitalize the labels Transitive and Intransitive to indicate they refer to paradigm classes. Transitive prefixes are portmanteau-like and reference two animate semantic arguments. Agent and Patient prefixes reference a single animate argument (the feminine-zoic singular prefix acts as a default pronominal prefix, abbreviated z/n, when the verb describes an entity that is inanimate or does not have an animate participant). Transitive prefixes and the two classes of Intransitive prefixes form paradigm classes. As Koenig and Michelson (2015a, 2015b) show, selection of Agent versus Patient prefixes, although to a large extent motivated semantically, can be unpredictable and must be stipulated in lexical entries. (See also Mithun 1991 and Mithun 2000.) For example, as shown in (16) and (17), different weather verbs select Agent or Patient prefixes with no apparent semantic motivation since weather verbs are not taken to take animate participants (and thus the prefix is the default feminine-zoic prefix).^[5]

The distinction between Agent-selecting and Patient-selecting lexical items is also relevant to entries that describe a relation between participants when just one of the members of that relation is animate, as the lexical item then behaves like an entry with a single semantic argument. Which semantically dyadic lexical items select Agent or Patient prefixes is most often predictable from the meaning of the verb and which of its arguments is (in)animate – the proto-agent or proto-patient (Dowty 1991). But the selection of Agent or Patient prefixes must sometimes be stipulated, as shown in (18), where the root -aty-/-uty-/-ati-/-uti- ‘drop in, add, leave, throw’ exceptionally takes a Patient prefix despite the fact that what is referenced on the prefix is the lexical item’s proto-agent argument, the thrower.

Stems that describe objects (at least, if they do not contribute non-relational semantic content, see below for relational nouns) can, rather arbitrarily, select Agent or Patient prefixes. Thus, in (6), oska·wáku ‘in the bush’ is based on a base (-skaw- ‘bush’) that selects the Patient prefix o-. In contrast, kanuhsáktaʔ ‘near the house’ is based on a base (-nuhs- ‘house) that selects the Agent prefix ka-.

Having explained the paradigm classes for pronominal prefixes, we now turn to the first inflectional property (S1 in Table 1) that targets lexical items on the basis of the sort of entities they describe, situations versus objects. This property pertain to the form of pronominal prefixes. Certain pronominal prefixes that are part of words that describe objects lack an initial glide; whereas pronominal prefixes that are part of words that describe situations include a glide.^[6] This is illustrated in (19) (the prefixes are bolded; as mentioned in footnote 1, underlining indicates segments which are devoiced due to a set of phonological changes that take place at the ends of utterances).

The (default) third person singular zoic/neuter pronominal prefix for Oneida nouns meaning ‘feather’ (ostó·sliʔ) and ‘blanket’ (okʌ´haʔ) is o-, but the corresponding prefix on the verb meaning ‘it is warm’ (yoʔtalíhʌ) is yo-.

The next two properties pertain to negation (properties S2 and S3 in Table 1). Negation for objects describing words is different from negation for situation describing words. Typical sentential negation in Oneida is marked by a particle yah, glossed negation in the examples, and either the negative prepronominal prefix teʔ- or the contrastive prefix th-. An example is given in (20b).

Negative equative predications, though, follow a different pattern. They have the particle té·kʌ rather than a verb with the negative or contrastive prefix (21a). Etymologically this particle is probably a verb, composed of the negative prefix teʔ-, the default feminine-zoic singular pronominal prefix ka-, and the verb root -i- ‘make up the total of’ (cf. Woodbury 2018: 214 for Onondaga). An example is given in (21b).

Examples (20) illustrate the fact that a subset of inflecting lexical items can take prefixes before pronominal prefixes, those that describe situations (property S5). They are called prepronominal prefixes.^[7] These prefixes include the future prefix ʌ´-, the optative prefix aa-, but also the negative and the contrastive (see Diaz et al. 2019 for a comprehensive analysis of prepronominal prefixes).

The final inflectional property that targets lexical items on the basis of ontological sort concerns lexical items that describe situations. Some lexical items that describe situations can occur in all three aspects (habitual, punctual, and stative). They are labelled v.a. (for “active verb”) in Michelson and Doxtator (2002) and verb in Abbott et al. (1996) and the on-line Oneida dictionary (available at www.uwgb.edu/Oneida/Dictionary.html). Examples are provided in (22) and (23). Some lexical items that describe situations, in contrast, only occur in the stative aspect. They are labelled v.s. (for “stative verb”) in Michelson and Doxtator (2002) or state in Abbott et al. (1996) and the on-line version of the dictionary (available at www.uwgb.edu/Oneida/Dictionary.html). Examples are provided in (22) and (23). Examples (24)–(25) are repeated from (8)–(10).

As was the case for the distinction between stems selecting Agent versus Patient prefixes, the distinction between active and stative bases that describe situations, while motivated, must sometimes be stipulated, as the contrast between the active roots in (26) and the semantically quite similar stative roots in (27) shows.

We have just discussed five inflectional properties that target lexical items based on the ontological sort of what they describe. But, as we mentioned in the introduction, one remarkable and underappreciated aspect of Iroquoian morphological parts of speech is that they are sensitive to two orthogonal semantic classifications. We now turn to inflectional properties that are sensitive to a second kind of semantic information included in the meaning of inflecting lexical items, namely the kind of relation that it includes in its semantic content. Section 7 will then provide evidence of the orthogonality of these two dimensions of classification.

6 Semantic content and morphological parts of speech

This section focuses on Oneida inflectional processes or constraints that are sensitive to whether a lexical item includes in its semantic content a polyadic relation or not and, more generally, what kind of relation a lexical item includes in its semantic content. The main property is polyadicity or degree of a concept (Quine 1936: 2); the other two are possession and body parts. As polyadicity is a well-known concept and its application to particular lexical items’ meaning is rather uncontroversial, there is no need for us to describe it further. We turn to the description of the inflectional properties that target lexical items on the basis of their relational semantic content. Two of these processes target lexical items that denote polyadic semantic relations (except possession and body parts) and four of these processes target lexical items based on the kind of denoted polyadic semantic relation, namely whether it is the relation of possession or body parts. We start with the inflectional constraints that target polyadic semantic relations to the exception of possession and body parts.

The first constraint (R1 in Table 2) targets lexical items that contribute semantic relations that have two animate arguments. Such lexical items bear so-called Transitive prefixes. As mentioned, Transitive prefixes in Oneida reference two animate semantic arguments of verbs through a single portmanteau-like prefix, as shown in (3a) and (3b). Importantly, Transitive prefixes are not restricted to lexical items that describe situations; they can attach to kinship terms, as shown in (28) and discussed in detail in Koenig and Michelson (2010).

As both arguments of a kinship relation are animate, kinship terms bear Transitive prefixes that reference ϕ properties of both arguments (a few kinship terms, named dyadic kinship terms by Koenig and Michelson after Evans 2000a bear an Intransitive prefix that references both kins). The fact that lexical items that describe situations in (3) and those that describe objects in (28) can similarly take Transitive prefixes shows that the target of Transitive prefix exponence rules is defined in terms of properties of the semantic relation contributed by lexical items, not in terms of what these lexical items describe. (As Koenig and Michelson 2021 show and discussed below, Transitive prefixes cannot attach to verbs that denote a possession relation even when both arguments are animate.) What is critical for the selection of Transitive prefixes is the polyadicity of the relation that helps the description (what is shared between situation descriptions and kinship terms) not what kind of entity is being described (a situation or an object). In Discourse Representation Theory terms (Kamp and Reyle 1993), what matters is not the sort of the discourse referents contributed by lexical items, but the kind of predicative conditions they include.

The second constraint (R2 in Table 2) pertains to reflexive prefixes. Reflexivization affects inflection in Oneida, since pronominal prefixes mark up to two distinct animate semantic arguments. The pronominal prefix of the relevant verb is Intransitive in (29) because there is only one distinct animate argument.

Now, interestingly, as Koenig and Michelson (2010) discusses, reflexivization also applies to lexical items that describe objects, namely kinship terms. Compare (30) to (31). In (30), the prefix on the kinship term liyʌ´ha ‘my son’ is Transitive and thus mark both members of the kinship relation, even though the word describes the younger member of the kinship relation. In contrast, the kinship term in (31), onatatyʌ´ha ‘mother and daughter’ includes the reflexive prefix -atat- and it describes the two persons that are in a kinship relation.

Reflexivization is thus a process that does not target just lexical items that describe a situation, but one that targets lexical items that share a certain semantic content, even if they differ in terms of the ontological sort of what they describe. Despite the fact that kinship terms describe persons (those that stand in a particular kinship relation to another person), as Koenig and Michelson (2010) argue, they can undergo reflexivization, because their semantic content is a dyadic relation between kin.

Transitive prefixes and reflexivization are two processes that target a class of lexical items on the basis of their polyadicity. But there are also processes that target lexical items on the basis of the particular relation that is part of the semantic content of lexical items. Three processes target relations of possession (R3, R4, and R5 in Table 2). The first one pertains to the form of pronominal prefixes (R3 in Table 2). Nouns in Oneida can be possessed and when they are, it is the possessor that is referenced by the pronominal prefix (rather than what is possessed). Interestingly, there is a special set of prefix forms in Oneida (as well as in Mohawk) for Patient prefixes that mark the possessor (see Koenig and Michelson 2021 for details). For example, the prefix lao- in (32) references whose fur it is and not the fur itself, and the prefix la- in (33) references whose nose it is and not the nose. In contrast, the pronominal prefix on non-possessed nouns like ostó·sliʔ ‘feather’ and okʌ´haʔ ‘blanket’ in (19) is the default feminine-zoic singular prefix o- since the only argument of the predicate denoted by these nouns, their referential argument (Higginbotham 1985), is inanimate.

More generally, Patient prefixes of lexical items that describe possessed entities have different forms than Patient prefixes on lexical items that describe entities that are not possessed or on lexical items that describe situations. This is illustrated in Table 3 which provides Patient possessive and verbal prefixes for stems that begin in a consonant or with a, the two most frequent classes of stems.

The second inflectional property that singles out lexical items that include a possession relation in their semantic content (R4 in Table 2) is the assignment of lexical items to the Agent and Patient paradigm classes. As noted above, the assignment, while motivated, is not entirely predictable except when a lexical item includes a relation of possession in its semantic content. In that case, it is entirely predictable: lexical items that include an inalienable possession relation in their semantic content select the Agent paradigm and lexical items that include an alienable possession relation in their semantic content select the Patient paradigm.

The third and final inflectional property (R5 in Table 2) that singles out lexical items that include a possession relation in their semantic content is the fact that these lexical items select Intransitive prefixes irrespective of the animacy of what is possessed, as shown in (34).

So, whereas other polyadic lexical items with two animate semantic arguments take Transitive prefixes, lexical items that include a possession relation where both the possessor and possessed are animate (as in kin relations) still take Intransitive prefixes, just as they do when what is possessed is inanimate.

We just saw three ways in which lexical items that include a possession relation in their semantic content behave differently. Lexical items that describe body parts are also exceptional in that rather than taking the usual so-called noun suffix, they necessarily take a locative suffix, even when no locative relation is involved, as shown in (35) (R6 in Table 2). See Koenig and Michelson (2019) for details on this issue.

We now have discussed several properties that hold of lexical items that describe entities of a particular ontological sort (situations vs. objects), or of lexical items that include in their semantic content a relation of particular kind (a polyadic relation, a possession relation, or a body part). In the next section, we show that these two dimensions of classification are, for the most part, orthogonal.

7 The orthogonality of the sort and relation classifications

In this section, we discuss ways in which a single lexical item can instantiate properties characteristic of lexical items contributing a polyadic or possession relation and properties characteristic of lexical items describing an entity of the sort object or situation. Table 4 summarizes the three inflectional properties we discuss that illustrate the orthogonality of the two semantic classifications.

The first piece of evidence for the orthogonality of the sortal and relational dimensions (B1 in Table 4) is the assignment of lexical items to the Agent and Patient paradigm classes. As we saw, the assignment is predictable in this case and depends on the (in)alienability of the possession relation. Koenig and Michelson (2021) discuss an interesting construction where possessed nouns are incorporated into a small set of stative verb roots. What is important for our purposes is that when possessed nouns so incorporate, pronominal prefix selection is determined by the ϕ properties of the possessor and, as a result, (in)alienability determines the Intransitive prefix paradigm class. Sentences (36)–(37) illustrate with inalienably possessed and alienably possessed incorporated nouns: the pronominal prefixes on the verb reference the possessor argument of the possessed noun rather than the verbs’ arguments (the verbs’ arguments would be ‘eye’ and ‘house’).

Koenig and Michelson call the fact that the possessor is referenced in examples like (36), rather than the noun’s referent, Possession Dominance. The possession relation is the dominant relation when it comes to pronominal prefix selection. Because the relation determining pronominal prefix selection is a relation of possession, Agent/Patient paradigm class assignment rules follow the (in)alienable pattern and can thus override the paradigm class of the verb the possessed noun is incorporated into. Furthermore, because the possessed argument is part of the meaning of a lexical item that describes a situation, the form of the pronominal prefix is that appropriate for lexical items describing situations. Assignment to Agent or Patient paradigm class and form of the prefix are thus sensitive to different dimensions of semantic classifications: kind of semantic relation (possession) versus ontological sort of what is described (situation).

The second piece of evidence for the orthogonality of the two dimensions of classification (B2 in Table 4) pertains to Patient possessive prefixes. As shown in Table 3, forms of third person Patient prefixes that are specific to so-called possessed nouns include an additional a vowel. But, as just noted, these forms also are glideless as befits lexical items that describe objects. So the forms of possessive prefixes obey both inflectional constraints on lexical items that describe objects and constraints on lexical items that include a possession relation in their semantic content.

The third piece of evidence for the orthogonality of the two dimensions of classification (B3 in Table 4) comes from a derivational rule that feeds into negation marking. We saw above that there are two patterns for negation in Oneida, one for situations that are not equative predications and one for equative predications where the second term of the equative is a word describing an object. Words based on roots that describe objects and include in their semantic content a polyadic relation, though, can participate in both patterns, as illustrated in (38a) and (38b) below. Sentence (38a) uses té·kʌ, which is expected for negated equatives. Sentence (38b) is unexpected. Koenig and Michelson (2021: 653–656) discuss this issue in detail and argue that it is the result of a lexical construction or rule that takes as input a lexical item that describes objects and includes a polyadic semantic relation in its semantic content (a possessed noun or a kinship term) and outputs a lexical item that describes a time-conditioned negative equality relation. Because the output of the lexical rule is a lexical item that describes a time-conditioned equality relation, negation is expressed as it is for all other situation descriptions (i.e., with the negative prefix teʔ-) and the form of the pronominal prefix is that appropriate for lexical items describing situations. The fact that so-called relational nouns in Oneida can negate either like other nouns or, via the lexical rule targeting relational nouns, like verbs further demonstrates the orthogonality of the two classifications of lexical items we have discussed in Sections 5 and 6.

8 General discussion: Whither nouns and verbs?

The last three sections have shown that morphological processes in Oneida target lexical items either on the basis of the ontological sort of the individual they describe or in terms of properties of the relation (or one of the relations) that is part of their semantic content (polyadicity, possession, or body parts). Note that in the description of the selection of lexical items no mention of noun and verb was needed. Membership in these classes reduces to semantic properties. As we mentioned in the introduction, that the class of lexical items linguists would label noun and verb include mostly lexical items that describe objects and situations, respectively, is not news. It is part of traditional pedagogical grammars and is what Spencer (2005) and Corbett (2012) call canonical nouns and verbs. But what is interesting is that Oneida lexical classes are canonical nouns and verbs (semantically, at least). We say that lexical classes in Oneida are semantically transparent in that lexical classes targeted by constraints or rules are defined intensionally by the semantic properties members of the class share. But such semantic transparency means that the labels noun and verb have a different status in the grammar of Oneida and the grammar of many other languages in that the labels noun and verb are dispensable in the former but not the latter. In Oneida, any constraint or rule which linguists think involves these labels can be replaced by the appropriate semantic property or conjunction of semantic properties. In this final section, we illustrate the consequence of eliminating labels such as noun and verb from the grammar of Oneida for the analysis of kinship terms.

Koenig and Michelson (2010) show that kinship terms have some of the properties traditionally associated with Oneida nouns (glideless pronominal prefixes, dimininutive clitics rather than aspect suffixes, negation) and some of the properties of verbs (reflexive marking in some cases, Transitive prefixes instead of pronominal prefixes marking the possessor, impossibility to incorporate without being nominalized). Koenig and Michelson conclude that kinship terms are both verbal and nominal, following the theory of mixed categories proposed in Malouf (2000). Such a conclusion is necessary if the grammar of Oneida includes labels such as noun and verb and kinship terms have some of the properties characteristic of nouns and some of the properties characteristic of verbs. They cannot be mere nouns or mere verbs, they must be both verbal and nominal. Such an analysis fails to explain why kinship terms have the nominal properties they have or the verbal properties they have. Any other distribution of nominal or verbal properties would, in principle, be possible. Maybe they could have aspect suffixes rather than diminutive suffixes but pronominal prefixes without glides. The reduction of the labels noun and verb to semantic properties explains why they have the nominal and verbal properties they have. Pronominal prefixes are without glides because kinship terms describe objects (a person that stands in a certain kin relation to another person) and the absence of glides is a property of lexical items that describe objects; the same is true of diminutives. And kinship terms may take reflexive prefixes and Transitive pronominal prefixes as these are properties of lexical items that include in their semantic content a polyadic relation (that is not a possession relation). Dispensing with the lexical classes noun and verb in favor of the lexical classes contributes a polyadic relation and describe objects (for kinship terms), thus allows us to predict which so-called nominal or verbal properties kinship terms have.

More generally, eliminating noun and verb as labels of classes of lexical items in the grammar of Oneida avoids the need to stipulate which properties are characteristic of nouns and verbs in Oneida, when what matters is ontological properties (for all monadic object describing lexical items and all situation describing lexical items aside from possession) or semantic relation properties (polyadicity irrespective of ontological sort or possession and body part), and sometimes both (for kinship terms and words denoting possession). A simple noun versus verb classification would miss the relevance of two dimensions of classification and their orthogonality.

More broadly, typological discussions of parts of speech should consider the epistemological status of lexical distinctions: Are they convenient labels for linguists – as noun and verb are in Oneida, in our analysis – or are they irreducible parts of the grammar of the language? Because the distinction is rarely if ever made, it is impossible to determine whether the Oneida situation is an exception, something relative rare, but attested elsewhere, or something quite frequent in languages of the world. Additionally, as a reviewer points out (p.c.), the semantic transparency of Oneida morphological parts of speech dovetails with what one could call the discourse transparency of Iroquoian word order argued for in Mithun (1987). Iroquoian could thus be called “semantically (or pragmatically) transparent”, as the reviewer suggests. And, indeed Koenig and Michelson (2014) claims that there is no formal syntactic features in Oneida and that syntactic combinatorics is direct, i.e., not mediated by semantically or pragmatically opaque formal features.^[8]

Finally, the cross-cutting semantic classification of inflecting lexical items in Oneida stresses yet again the importance of distinguishing referential and attributive (Donnellan 1966) or predicative (Kamp and Reyle 1993) properties contributed by linguistic expressions. Koenig (1999), Koenig and Mauner (1999), and Farkas and de Swart (2003) show that some noun phrases can introduce predicative properties or conditions without introducing discourse referents (in Discourse Representation Theory terms). We have shown in this paper that inflectional morphology can be sensitive to one kind of semantic properties or the other, or both, within a single language: distinguishing between (discourse) referential and predicative properties (conditions) matters to morphology too.