The Wiru Noun-Modifying Clause Construction

We present an analysis of the general noun-modifying clause construction in Wiru (Trans–New Guinea). In this construction, a subordinate clause modifies a head noun, whether or not the head noun plays a syntactic role in the subordinate clause. We present several structural features that are of typological interest. Subordinate clauses inflect for a restricted set of tense–aspect–mood categories: they neutralize the distinction between future and optative, and they exhibit a strong preference for anterior aspect. They can contain resumptive pronouns as well as full nouns that are coreferential with the head noun. Moreover, the head noun itself can be omitted, although other noun phrase constituents (such as adjectives or determiners) may still be present. This construction also exhibits special tonal properties, in that the lexical tone of the head noun is overridden and replaced with an HL falling contour. Finally, verb agreement with subjects that contain subordinate clauses is semantically governed, not structurally. We conclude with some areal comparisons.

1. INTRODUCTION.¹

In this paper, we discuss a subordination construction in Wiru, a language from the highlands of Papua New Guinea. In this construction a subordinate clause modifies a head noun attributively, but we argue that the construction should not be considered a relative clause but rather a general noun-modifying clause construction (GNMCC) (Matsumoto et al. 2017). GNMCCs allow for a greater range of semantic relationships between the modifying clause and the head noun than relative clauses do. We present evidence for the GNMCC analysis and then describe the properties of Wiru GNMCCs. The discussion highlights several features of the construction that are of typological interest, including agreement behavior in which semantic considerations override structural ones, tonal behavior that is associated exclusively with the GNMCC construction, and the permissibility of GNMCCs without an overt nominal head.

We present the necessary background on GNMCCs in the following subsection and introduce Wiru after that. In section 2, we describe the semantic properties of the GNMCC construction, detailing the kinds of relationships that can exist between the subordinate clause and its head. We then present various formal properties of the construction: section 3 on the structural properties of the subordinate clause, section 4 on the structural properties of the head noun, and section 5 on properties of the construction as a whole. In section 6, we describe the external syntactic behavior of the GNMCC, focusing on agreement, and in section 7, we present some comparisons and re-examine some older data in light of our Wiru analysis. We conclude in section 8.

1.1. GNMCCs

The term GNMCC was coined rather recently, after a series of workshops focused on noun-modifying clauses in Eurasian languages (Matsumoto et al. 2017). But the analytical insight dates at least to the 1980's and Yoshiko Matsumoto's work on Japanese (1988, 1997). She observed that the standard relative clause analysis, which posited a gap in the subordinate clause that was coreferential with the head noun, encountered difficulties with some Japanese data. Some constructions could be analyzed as relative clauses, such as (1).

(1) Japanese

 

Here, the subordinate clause gakusei ga katta 'the student bought (it)' modifies hon 'book', and there is no overt object in the relative clause. But even in examples such as this one, positing a gap is not necessary due to the frequency of zero anaphora in Japanese. More problematically, positing a gap is not possible for many other so-called relative clauses, such as (2).

(2) Japanese

 

Here, the subordinate clause is atama ga yokunaru 'the head improves'. This modifies hon 'book', but there is no gap in the subordinate clause that corresponds to the head noun—yokunaru is an intransitive verb.

Matsumoto's analysis is that (1) and (2) are the same construction, in which a subordinate clause modifies a head noun attributively, but in which the syntactic role that the head noun plays in the subordinate clause, if any, is not marked overtly. The "gap" in examples like (1) is, in this analysis, no longer considered a gap that is referentially bound to the head noun, but rather a normal occurrence of zero anaphora of a highly accessible discourse participant.

The term GNMCC, then, refers to this type of noun-modifying clause construction (NMCC): one that has both relative-clause-like interpretations, in which the head noun functions as an argument or adjunct in the subordinate clause, and interpretations in which the head is neither an argument nor adjunct of the modifying clause. Since Matsumoto's original work, GNMCCs have been identified in a variety of languages besides Japanese (Comrie 1998; Matsumoto et al. 2017), primarily in Eurasia.

This paper is, to our knowledge, the first analysis of a GNMCC in a Papuan language (but see some mention of GNMCCs in Daniels 2015). Previous work on subordinate clauses in Papuan languages has focused primarily on the fact that they often employ nominal morphology as subordinating morphology and conflate the distinction between complement, adverbial, and relative clauses into a single structural type (Foley 1986, 2018). The Wiru construction at issue here is, as we will see, different.

1.2. THE WIRU LANGUAGE

Wiru is spoken in the Ialibu-Pangia district of Southern Highlands province, Papua New Guinea. The population was listed as 15,300 in 1981 (Wurm and Hattori 1981), and that figure is still cited in the current Ethnologue (Eberhard et al. 2019), although the current population is surely much higher.²

The first classificatory statement that we are aware of involving Wiru was made by Wurm (1960:127), who placed it in his Enga-Huli-Pole-Wiru family, one of five branches of his East New Guinea Highlands Stock. Wurm maintained essentially this classification in subsequent work (1961, 1964, 1975), although he elevated Wiru to "the status of a family-level isolate in the stock" (1975:466) and judged that the problem of its genetic affiliation could not "as yet be regarded as fully settled" (1975:472).

In the same 1975 volume, more evidence for the genetic affiliation between Wiru and other East New Guinea Highlands Stock languages is provided by Kerr. His primary evidence is typological, and he discusses topics such as the prevalence of 2/3pl syncretism in both groups, the existence of medial verbs, and the existence of a benefactive construction. This is no longer considered convincing evidence for genetic affiliation, since we now know that typological features can be spread through contact (Ross 2007). But Kerr also mentions more promising facts, most of which are apparent cognates between Wiru and the Engan language Kewa (Franklin 1971). Among these are a samesubject suffix -ma that attaches to medial verbs; a noun-phrase enclitic =me that marks agents and instruments; a comitative enclitic or postposition that is pala in Wiru and para in Kewa; and a sentence-final interrogative particle pe (Kerr 1975:280). He also lists a great deal of lexical cognates. Many of these are cultural vocabulary that is easily borrowed, but some of the more convincing meanings include 'name', 'eye', 'mouth'/'lip', 'dry', 'heavy', 'painful', 'sick', 'skin', 'food', 'mountain', 'urine', 'good', 'big', 'eat', 'shake', 'yesterday', 'breast', and 'flying fox'.

The East New Guinea Highlands Stock is no longer considered a valid genetic grouping and is now regarded as a collection of disparate subgroups of Trans–New Guinea (TNG), which have not been shown to form a clade of their own within the family (Ross 1995; Pawley 2005). Our provisional evaluation of Kerr's evidence, then, is that it supports the claim that Wiru belongs to the TNG family. In this, we agree with Pawley and Hammarström, who consider Wiru to be one of the groups with "relatively strong claims to membership in TNG" (2018:31), citing in particular 1sg, 2sg, and 2pl pronouns, the negative auxiliary, and forms for 'name', 'louse', 'ashes', 'moon', 'instructions, incantations', and 'heavy' (2018:71), all of which appear to be reflexes of Pawley's Proto-TNG reconstructions.

The data for this paper come from a corpus of elicited and naturalistic speech by two native speakers of Wiru, Susan Yakip and Thompson Mange. It was collected during and after a two-semester field methods sequence at The Australian National University in 2017. For naturalistic examples, we cite the line of the recording they come from next to the translation; for elicited examples, we indicate that they are elicited.

In the remainder of this section, we provide a typological overview of Wiru grammar, as context for our treatment of GNMCCs in the following sections. Prior description of the language was done by Harland Kerr. He produced an M.A. thesis which focused primarily on the verb, and also treated some other topics (1967), as well as two dictionary manuscripts that contain sometimes extensive notes about the functions of individual morphemes (Kerr n.d. a, b).

The typical order of constituents in a Wiru transitive clause is subject–object–verb (3).³

(3)

 

The basic order of constituents in the noun phrase is outlined in (4); all constituents can be omitted. Determiners include demonstrative forms and possessors. (Other constituents, such as possessors and postpositional phrases, can also be present, but we omit these for the sake of simplicity.)

(4) Det [S] Adj N Quant

We discuss the position of the subordinate clause below, so for now we only exemplify the relative order of determiner, adjective, and noun in ena tu aroa 'that young woman' (5).

(5)

 

In many ways, Wiru neatly fits the general typological profile of a highlands TNG language. It possesses clause chaining constructions (Roberts 1997; Foley 2018), in which a series of "medial" clauses are chained together and end in a "final" clause. Verbs in medial clauses are marked for switch reference, which is to say that their morphology indicates whether their subject is the same as, or different from, the subject of the next verb in the chain. For example, in (6), the first clause is yapu wirakoli 'they arrived at the house', and the subject, a group of friends, is different from the subject of the following clause, so the verb is marked with different-subject ('ds') morphology.

(6)

 

In (7), on the other hand, the dog is the subject of both clauses, so the verb in the first clause tue pome 'the dog went and' is marked with same-subject ('ss') morphology.

(7)

 

Wiru also possesses a large inventory of serial verb constructions, not all of which share the same morphosyntactic properties (Chen 2018). These express a wide variety of meanings, including tense–aspect–mood marking, directionality, causation, sequential action, and the modification of an action by a stative predicate. For example, in (8), yawa 'walk' is interpreted lexically; kawa 'stand' contributes progressive aspect; and pome 'go' signals directionality.

(8)

 

A final construction that bears mention is one that has received a variety of labels, including verb adjunct construction (Pawley and Bulmer 2011), adjunct plus light verb construction (Foley 2018), adjunct nominal construction (Donohue 2005), and coverb construction (Evans 2014). Many Papuan languages combine a light verb with another word—the "adjunct" or "coverb"—to form a complex predicate. The other word is usually a noun, an adjective, or a member of a word class of "adjuncts" that only occur in this construction. In Wiru, we have encountered nominal and adjectival verb adjuncts, but none of the third type. An example is the adjective kõ 'cross' in (9); this word combines with o- 'say' to mean 'be cross (at)' in the first clause of this example. In the second clause, o- has its normal lexical function and means 'say'.

(9)

 

2. THE WIRU GNMCC: SEMANTIC PROPERTIES

In this section, we present a brief outline of the structure of Wiru noun-modifying clauses and describe the different semantic relationships that can be realized between the subordinate clause and the head noun. This discussion will make it clear that NMCCs in Wiru really are GNMCCs. We then describe the formal properties of this construction in more detail in section 3.

Wiru NMCCs precede the head noun and take the form of a well-formed clause (or clause chain). For example, the matrix clause in (10) is modifying the head noun mairakoma 'children' in (11) without any change to its morphosyntactic properties.

(10)

 

(11)

 

NMCCs can have a relative clause interpretation, in which case arguments and adjuncts of the subordinate clause serve as the head noun—in other words, NMCCs can relativize on all syntactic positions. In example (11), the recipient of the subordinate clause is coreferential with the head noun. Other subordinate grammatical roles that can serve as head noun include subjects, as in (12); locations (13); comitatives (14); and possessors (15).

(12)

 

(13)

 

(14)

 

(15)

 

Sometimes it is not clear exactly how to characterize the relationship between subordinate clause and head noun. In (16), the subordinate clause kianea karo piki 'there is a red car' modifies ailaroa 'people'.

(16)

 

Semantically, the people are possessors of the car, but the subordinate clause does not seem to have an unmentioned possessor in it. An alternative would be to say that the head noun is the location of the subordinate clause, and that location is being metaphorically extended to encode possession.

Fortunately, we do not need to solve this problem, since the semantic relationship between the head noun and the modifying clause is not limited to relative-clause-like interpretations. Wiru NMCCs can modify the head noun in a variety of ways that Matsumoto et al. (2017:5) call "extended NMCCs." These include examples like (17) and (18), in which the content of the head noun ko 'story, talk' is described by the subordinate clause. In instances like these, the head noun itself is neither an argument nor adjunct of the subordinate clause.

(17)

 

(18)

 

A similar example is given in (19). Here, the head noun oi 'time' is characterized by a subordinate clause that describes a future event.

(19)

 

Examples (17)–(19) illustrate what Matsumoto (1997) calls "noun-host" GNMCCs. Instead of the subordinate clause providing a semantic frame for the interpretation of the head noun, which is typically so with relative clauses, here it is the head noun that "evokes a nominal frame which acts as a [semantic] host for the content of the modifying clause" (1997:70).

Note, however, that even when the semantic frame comes from the subordinate clause, it is not always possible to provide a relative-clause-like interpretation. For example, in (20) the subordinate clause ue naki 'they drink water' describes the purpose of the head noun mere 'container'.

(20)

 

We see from the examples above that NMCCs in Wiru can express a variety of semantic relationships between the subordinate clause and the head noun. Since these include both relative-clause-like relationships—relationships where the head noun expresses an argument or adjunct of the subordinate clause—and other relationships, this construction fits Matsumoto et al.'s definition of a GNMCC.

Before moving on to the formal properties of Wiru GNMCCs, we wish to make one more observation. It is possible for GNMCCs in Wiru to lack an overt nominal head. In such instances the construction usually refers to the event expressed by the subordinate clause, but treats it nominally—that is to say, referentially. GNMCCs without overt nominal heads may still contain other modifiers, as in (21), which contains the definite determiner eni but no head noun.

(21)

 

In this example, the subject is recalling an earlier, happier way of life. The subordinate clause ena kakona 'they lived like that (i.e., happily)' expresses that state, and the GNMCC simply refers to the state itself, which is the syntactic object of wedekakome 'he thought about'.

In the next three sections, we describe the formal properties of GNMCCs. We begin with the properties of the subordinate clause, then move on to properties of the head noun (section 4), and finally discuss holistic properties of the GNMCC construction as a whole (section 5).

3. SUBORDINATE CLAUSE PROPERTIES

Subordinate clauses in GNMCCs are structurally quite similar to matrix clauses. They can be transitive (22) or intransitive (23).

(22)

 

(23)

 

They may also contain a clause chain, although this is fairly rare. Example (22) contains a different-subject chain, and (24) contains a same-subject chain.

(24)

 

Serial verb constructions are somewhat more common in subordinate clauses (25).

(25)

 

The predicate inside the subordinate clause can also be nonverbal (26).

(26)

 

And subordinate clauses in GNMCCs can themselves contain GNMCCs (27).

(27)

 

As the examples above have shown, if the subordinate clause is verbal, its arguments may be elided, as they can be in main clauses. Examples above include subordinate clauses with elided subjects (22), elided objects (11), and with both core arguments present (15).

Because eliding subordinate arguments is optional, it is possible for a GNMCC with a relative-clause-like interpretation to contain an overt noun phrase which is coreferential with the head noun of the GNMCC. An example of this is (28), where the subordinate clause modifies mairakoma 'children', but also contains mairakoma itself.

(28)

 

The marking of tense and aspect in GNMCCs differs in some respects from that in matrix clauses. These features of GNMCC subordinate clauses are not shared by all subordinate clauses. For example, the marking of future tense, which is distinctive in GNMCCs, is not different in complement clauses introduced by the quotative particle wa. We restrict ourselves in this section to a discussion of the features of subordinate clauses that modify a head noun.

In GNMCCs, the simple present tense can occur with its usual meaning (29).

(29)

 

However, the habitual suffix -ya (30) is not allowed in GNMCCs. Instead, habitual aspect is also conveyed using the simple present (31).

(30)

 

(31)

 

GNMCCs also differ from matrix clauses in the expression of future tense. Matrix clauses have a dedicated future tense verb suffix -o, to which the person marker is then suffixed (32). GNMCCs, by contrast, express future time reference with the optative marker, -rVk (33).

(32)

 

(33)

 

The use of the optative in the future tense is a grammatical requirement; the simple future suffix is ungrammatical in a GNMCC (34).

(34)

 

Note that this use of the optative in GNMCCs differs from its use in matrix clauses. In matrix clauses the optative does not receive future tense readings, but rather counterfactual (35) or potential readings (36).

(35)

 

(36)

 

When the action of a GNMCC occurs before the action of the matrix clause, the GNMCC final verb may take simple past tense inflection if the matrix clause is in present (37) or future tense (38).

(37)

 

(38)

 

A GNMCC final verb may also take simple past tense inflection if it occurs synchronously with the action of a past tense matrix clause (39).

(39)

 

GNMCCs also often occur with the anterior perfect suffix -na. Kerr (n.d. a: 225) observes that this suffix marks a range of words and constructions that are "dependent on nouns." The anterior perfect indicates that the event it marks took place prior to some other time established in the discourse, and that it has relevance to that time. We will call this other time that the anterior references the "reference time." In (40), for example, the anterior-marked verb pukuna 'had gone' expresses an event that occurred before the event of aching, expressed by naka in the simple past.

(40)

 

Note that the anterior event must be construed as relevant to the reference time; simple anteriority, or temporal succession, is typically marked with clause chaining. But in (40) the event of walking is marked as relevant to the event of aching—in this instance because of a presumed causal connection.

In the following example, there are two anterior verbs. The first follows a verb with the medial future suffix, -de, which provides the sense of 'wanting'. Both anterior verbs once again signal events that occur before the final verb, which is in the simple past tense.

(41)

 

The semantics of present relevance can extend to the discourse context. Example (42), for example, carries an implication that the hearer should already know the proposition being conveyed.

(42)

 

A final example of matrix-clause anterior marking is given in (43), where it occurs on an optative verb, noarikuna 'I should have eaten'.

(43)

 

Here, the optative is used in its matrix-clause function to mark counterfactuality. The anterior aspect, then, indicates that the time when this potential event might have occurred preceded the event of the following clauses. In other words, eating could conceivably have taken place before the reference time, even though it did not. In addition, the anterior also indicates that the nonoccurrence of that event is relevant to the reference time.

When -na 'anterior' occurs in GNMCCs, the reference time is understood to be the time of the matrix clause event. Thus, when the matrix clause is inflected for past tense, a -na-marked GNMCC indicates that the subordinate event occurred even farther in the past than the matrix event. For example, in (44) the state that the subject is recalling precedes the event of recalling that state.

(44)

 

Similarly, in (45), the subject encounters his old drinking buddies after swearing off drink. Their drinking together preceded the current encounter and is marked as anterior.

(45)

 

Example (46) nicely illustrates the contrast between an event marked as simultaneous with the matrix-clause event and an event marked as anterior. The clause concerns a man being released from prison.

(46)

 

The first subordinate clause is aka noademe toka 'he was about to come out'. This event modifies the head noun pere 'time' and is synchronous with the matrix clause—almost by definition, since the point of the subordinate clause is to establish the temporal location of the matrix clause. Consequently, it is marked for past tense, since the matrix clause has past time reference. The second subordinate clause is korokoa merekona 'he had removed and given (things to them)'. This clause modifies yarene 'group', which refers to the clothes and other items the prisoner had surrendered when he was admitted to prison. His giving these possessions over to the police preceded their returning them, so the clause is marked anterior.

Readers will have noticed that all the anterior clauses are marked present tense. This is a consistent pattern: almost all anterior subordinate clauses are inflected in the present. We currently have no explanation for this and can only offer the single counterexample we have. In (47), a past anterior perfect subordinate clause modifies the head noun ta 'village'.

(47)

 

To summarize, tense marking in subordinate clauses can indicate relative tense between subordinate and matrix clauses. The contrasts are nicely illustrated in (48)–(50). In (48), the subordinate clause refers to an event that is completed at the time of the matrix clause.

(48)

 

In (49) the event of the first subordinate clause is simultaneous with the event of the matrix clause.

(49)

 

And, in (50), the event of the subordinate clause is subsequent to the event of the matrix clause.

(50)

 

4. PROPERTIES OF THE HEAD NOUN

As mentioned above, the head noun can be coreferential with any argument or adjunct in the subordinate clause, whether overtly present or not. We have already seen numerous examples of this above. In example (12), the head noun is coreferential with the subordinate-clause subject and in (25) with the object. Head nouns that refer to adjuncts are also common: we see a location in (13), a source in (15), and a temporal noun in (19).

Table 1.

WIRU NOMINALISING ENCLITICS

Wiru also possesses a set of enclitics that can occupy the head noun position of a GNMCC. That is to say, they can cliticize to an inflected verb and create a noun phrase. The inventory of clitics is given in table 1.

These forms have very broad meanings, referring to any kind of action, object, speech, or location. Examples in which they head GNMCCs are given below:

(51)

 

(52)

 

GNMCCs headed by these enclitics have some idiosyncratic properties. The clitic head is phonologically bound, for one. The evidence for this is that the enclitic begins with r (phonetically [ɾ]), which is actually a word-medial allophone of /t/. This sound does not occur word-initially except in loanwords.⁴

The subordinate clauses that they attach to also exhibit different behavior. Recall that the future tense is not allowed in subordinate clauses, and that future time reference is instead marked by the optative. While the nominalizing enclitics can attach to optative clauses, as in (52) and (53), they can also attach to future tense clauses (54).

(53)

 

(54)

 

Note that these enclitics do not only occur in GNMCCs but can also attach to other constituents such as question words (55) and demonstratives (56).

(55)

 

(56)

 

A more complete study of these forms will have to await future research.

The last property of head nouns that we wish to discuss is the fact that they can be omitted. This occurs exclusively with GNMCCs that are optative or anterior. This restriction seems to be motivated by the fact that optative and anterior inflections are already associated with GNMCCs—the optative is the required way to mark future tense, and the anterior perfect is a common subordinate aspectual category. This association between these inflections and subordinate clause function may serve as a signal to the hearer that the clause is to be interpreted nominally even in the absence of an overt nominal head.

Example (57) contains a headless GNMCC in the optative, and (58) contains an anterior headless GNMCC.

(57)

 

(58)

 

Note that neither of these subordinate clauses contains a noun that refers to its notional head; that is, there is no noun meaning 'time' in (57) or referring to what was given in (58). But subordinate clauses in headless GNMCCs can contain overt reference to their referents, as with ko 'story' in (59) and oi 'time' in (60).

(59)

 

(60)

 

The grammatical requirement that subordinate clauses in headless GNMCCs occur in the optative or the anterior is illustrated by the ungrammaticality of putting example (58) into the simple past or the present tense (61).

(61)

 

An interesting headless GNMCC is shown in (62). Here, an optative headless subordinate clause serves as the subject of the light verb toa 'do'. This construction seems to assert that the event of the subordinate clause was expected to, or supposed to, take place. This is a use of the GNMCC that requires more investigation.

(62)

 

5. HOLISTIC PROPERTIES

In natural speech, the subordinate, attributive clause in a GNMCC behaves in many ways like other noun phrase constituents. That is, in some ways it is incorrect to discuss the pairing of a subordinate clause and a head noun as a construction in its own right. Rather, these are simply two fillers that can occupy slots in the broader Wiru noun phrase construction. In this section, we first present evidence in favor of this view. We then present counterevidence in the form of a tonal analysis that shows that the Wiru GNMCC is indeed a single unique structure that connects a subordinate clause with a head noun.

The subordinate clause can occur in a right-dislocated "afterthought" position (63).

(63)

 

This utterance is still well formed, and the subordinate clause does not require a recapitulation of the head noun. However, such constructions are never produced in elicitation. GNMCC noun phrases can also contain other modifiers. In natural speech, these tend to be quantifiers, such as pea 'all' (64) or the pluralizing yarene 'group' (65).

(64)

 

(65)

 

In elicitation, GNMCC noun phrases can contain a definite demonstrative, an attributive nonfinite verb, and an adjective, all in addition to the subordinate clause and the head noun (66).

(66)

 

Demonstratives typically occupy the leftmost position in noun phrases, but subordinate clauses can occur to the left of demonstratives, if they are heavy enough. So (67) is a permissible shortened version of the noun phrase in (66).

(67)

 

As discussed above, the head noun of the GNMCC need not be overt. When it is absent, its absence does not necessarily entail the absence of these other modifiers; for example, the noun phrase in (68) is missing a head noun but does have the definite determiner eni.

(68)

 

Similarly, the subordinate clause in (69) lacks an overt head noun.

(69)

 

Enikeni generally serves as an adjective that means 'all sorts of N', and tuku 'search' is a verb adjunct that combines with toa 'do' to form a complex predicate. So this noun phrase contains a subordinate clause and an adjective, but no head noun.

Case markers in Wiru are enclitics to the noun phrase. Noun phrases with subordinate clauses are, as expected, no different from other noun phrases in this regard and can be marked with possessive =ne (70), agentive/instrumental =me (71), as well as oblique postpositions such as locative ke (72).

(70)

 

(71)

 

(72)

 

As expected, headless GNMCCs can still receive case marking, as illustrated with instrumental =me in (73). Here, in the absence of a head noun, the case enclitic attaches directly to the subordinate clause.

(73)

 

All of these features suggest that GNMCCs, understood as a construction composed of a subordinate noun-modifying clause and its head noun, should not really be recognized. Rather, the subordinate clause should instead be seen as simply one more possible constituent in the Wiru noun phrase. In this sense, the label "attributive clause," as used by Comrie (1998), would be more appropriate.

Figure 1.

PITCH TRACE OF (74)

But there is one important respect in which the GNMCC is a construction distinct from a normal noun phrase: it has its own tonal properties. A preliminary analysis of tones in Wiru nouns has shown there to be five lexical melodies: high (H), low (L), rising (LH), falling (HL), and rising–falling (LHL).⁵These are word-tones—that is, one tone is assigned to each noun, with the pattern spread across its syllables (Donohue 1997). However, in a GNMCC, the lexical tone of the head noun is overridden and replaced with a falling tone. We have not yet identified any other syntactic structures in Wiru that override lexical tone in this way, although analysis of Wiru tone is still in its early stages. For example, take the high-tone noun, tonō 'mountain', illustrated in figure 1, which is a spectrogram of (74). Tonō 'mountain' forms a minimal pair with tonô 'bone', which has LHL tone. Tonô 'bone' is illustrated in figure 2, corresponding to (75).

(74)

 

(75)

 

When tonō 'mountain' serves as the head of a GNMCC (76), its H tone is replaced with falling HL tone, as shown in figure 3.

(76)

 

The LHL tone of tonô 'bone' is likewise replaced with an HL tone when it functions as the head of GNMCC, as shown in figure 4. Thus, the tonal contrast between tonō 'mountain' and tonô 'bone' is neutralized in this environment.

Figure 2.

PITCH TRACE OF (75)

Figure 3.

PITCH TRACE OF (76)

Figure 4.

PITCH TRACE OF (77)

(77)

 

The same phenomenon is illustrated again for an LH-tone noun, tá 'rain', in figures 5 and 6, which are spectrograms of (78) and (79), respectively.

(78)

 

(79)

 

This tonal behavior is not a property of the subordinate clause; rather, it only emerges when the subordinate clause is followed by its head noun. This is illustrated by GNMCCs that contain no overt head noun but that do contain other modifiers, such as determiners and adjectives. The lexical tone of eni (80) is illustrated in figure 7.

Figure 5.

PITCH TRACE OF (78)

Figure 6.

PITCH TRACE OF (79)

Figure 7.

PITCH TRACE OF (80)

(80)

 

When this determiner is in a noun phrase with a subordinate clause but no head noun, as in (81), it does not receive HL tone in the way that a head noun does. This is illustrated in figure 8, where the LH tone of ení is clearly visible.

(81)

 

Similarly, the lexical tone of the adjective enikenî 'various' is LHL, as illustrated in figure 9, a pitch trace of enikenî pea [various all] 'all those things'.

Figure 10 shows that this tone is maintained in (82), where the object of kĩõrokoa 'straighten, prepare' is a noun phrase that is composed of a subordinate clause and enikeni, but no overt head noun.

Figure 8.

PITCH TRACE OF (81)

Figure 9.

PITCH TRACE OF ENIKENÎ PEA 'ALL THOSE THINGS'

(82)

 

Figure 10 also makes it clear that the fall in pitch that is associated with the GNMCC is not a boundary tone. This observation is also confirmed by examining headless GNMCCs, such as (57), the pitch trace of which is given in figure 11. Here, the tone of the final verb of the GNMCC, noareko 'he would come', is level.

The evidence suggests that these tonal properties are, in fact, a property of the GNMCC construction that emerges from the particular combination of a subordinate modifying clause and a head noun. It must thus be concluded that the Wiru GNMCC is a single unique construction in Wiru.

6. AGREEMENT PROPERTIES

In this section, we describe the external syntactic properties of the Wiru GNMCC, in particular its agreement properties. As mentioned in section 3, it is possible for the head noun of a GNMCC to

Figure 10.

PITCH TRACE OF MOST OF (82)

Figure 11.

PITCH TRACE OF THE BEGINNING OF (57)

recapitulate an overt nominal from the subordinate clause. Another example of this is given in (83), where the expressions eni yarene, in the subordinate clause, and yarene, in head noun position, both have the same referent.

(83)

 

GNMCCs do not all have third-person reference; it is also possible for them to have first- or second-person reference. In such instances, the GNMCC is not headed by a pronoun but by a semantically inert noun. For example, in (84), the subordinate clause is no skuul wene meka muku 'I don't like school', and this clause modifies the head noun aroa 'woman'. The resulting GNMCC serves as the subject of the predicate skuul ke puku 'am going to school'.

(84)

 

Interestingly, the subject noun phrase in this example requires first-person agreement on the matrix verb. That is, the noun phrase no skuul wene meka muku aroa, which might be rendered into English as 'the I-don't-like-school woman', controls first-person agreement on the matrix verb puku 'go'. This is in spite of the fact that the structural head of this noun phrase is aroa 'woman', which would normally be a third-person noun. It is ungrammatical for the verb in (84) to be marked with third-person agreement morphology.

(85)

 

Note also that this agreement patterning is not due to the overt first-person pronoun in the subordinate clause. The same agreement marking is seen when no '1sg' is omitted.

(86)

As mentioned, this is also possible for second-person referents. This can be seen with the subordinate clause ne skuul wene meka moko 'you don't like school' in (87). But since Wiru indicative agreement morphology does not distinguish second from third person in either the singular or the plural, the agreement mismatch that arises with first-person referents does not arise here.

(87)

 

This ambiguity becomes more pronounced when the pronoun is omitted from the subordinate clause (88). This example is compatible with both a second-person and a third-person reading.

(88)

 

It seems there are two possible analyses for this data. First, we could say that nouns in Wiru are lexically unspecified for person. Then, when a head noun like aroa 'woman' in (86) combines with a first-person subordinate clause, it receives the person value of that clause. This analysis predicts that aroa, being unspecified for person, could combine with first-person marking on the verb even in the absence of any modifiers, but that is ungrammatical (89).

(89)

 

Rather, to achieve something like the meaning intended in (89), speakers use a pronoun and modify it with a postpositional phrase (90).

(90)

 

The alternative analysis would be to say that nouns like aroa 'woman' are indeed inherently third person, but that agreement in Wiru is conditioned semantically rather than structurally. In this analysis, a subordinate clause does not have to agree with the person value of its head noun—indeed, we do not have to claim that subordinate clauses are specified for a person value at all—and verb agreement is instead triggered by the referent of the subject, rather than by one of its structural features. (With the proviso, given the frequency of zero anaphora in Wiru that the subject does not have to be overt.) This appears to comport with other domains of Wiru grammar where semantic considerations trump structural ones. Switch reference marking is one such domain, as is verb agreement in some single-clause sentences. Consider example (91), in which the subject is first-person singular, yet the verb is marked for first-person plural. The agreement here once again prioritizes semantics over syntax. Note that the subject is not 'Sally and I', for which Wiru would use an inclusory pronominal structure (Lichtenberk 2000), Sali tora [Sali 1du]. Rather, the subject is no, and the predicate is modified by the comitative adjunct Sali pala 'with Sali'.

(91)

 

Finally, we observe that in GNMCCs with first- or second-person referents there is a strong preference for the head noun to be one of the following four: aroa 'woman', ago 'man', ali 'man', and yarene 'group'.⁶ The choice depends on the gender and number of the referent. (The choice between ago and ali seems to depend on speaker preference.) GNMCCs with first-person referents and other human head nouns, such as ana 'husband' or adona 'married couple', are ungrammatical. But in second-person imperative contexts, other head nouns appear more acceptable (92).

(92)

 

7. COMPARISONS

Having described the structure of Wiru GNMCCs in some detail, we can now turn to some typological comparisons. The occurrence of resumptive pronouns inside relative clauses has, of course, been recognized for a long time (Keenan and Comrie 1977), and resumptive pronouns are also commonly found in GNMCCs. This is seen, for example, in Cantonese (Matthews and Yip 2017), Bezhta and Hinuq (Comrie et al. 2017), and Tundra Nenets (Nikolaeva 2017).

In general, resumptive pronouns in these languages seem more common with head nouns of low accessibility in Keenan and Comrie's accessibility hierarchy (1977). However, Comrie et al. (2017:132) argue that there is no evidence for this being a syntactically motivated constraint in GNMCCs, as it is with English relative clauses. Indeed, they contend that the GNMCC approach predicts the presence of modifying clauses both with and without resumptive pronouns in languages with low referential density. A GNMCC without a resumptive pronoun corresponds to a main clause with zero anaphora, whilst a GNMCC with a resumptive pronoun corresponds to a main clause without zero anaphora.

The recapitulation of a head noun inside a noun-modifying clause is much rarer but not unprecedented. Let us consider Kombai, another TNG language described by de Vries (1993). The following is an example of an extended semantic relationship between noun and modifying clause:

(93) Kombai

De Vries describes relative clauses in the language as generally having both a "grammatical head" and a "topical head" (1993:77). This construction has been analyzed as a "double-headed" relative clause (Cinque 2011), and Kombai is the only language recorded in the World Atlas of Language Structures Online as having such relative clauses (Dryer 2013).

In Kombai relative clauses, the grammatical head is a noun such as rumu 'person' or ro 'thing' and is external to the modifying clause. The topical head is a more semantically specific, coreferential noun that sits within the modifying clause. In (94), yare 'old man' and rumu 'person' are coreferential. Yare is the topical head, as it is the more semantically specific term. Rumu, as the more general term, is external to the modifying clause and is the grammatical head.

(94) Kombai

 

In de Vries's analysis, the combination of subordinate clause plus grammatical head "functions as a restrictive apposition to the topical antecedent" (1993:77). On this account, a more literal translation of (94) is 'the old man, the person who joins the work', with the internal structure shown in (95a). But, it seems likely to us that this Kombai construction is rather a GNMCC that, like Wiru, allows coreferential nouns inside and outside the subordinate clause. On this analysis, the "topical head" is inside the modifying clause as in (95b).

(95)

a. Yare [gamo kheraja bogi-n-o] rumu …

b. [Yare gamo kheraja bogi-n-o] rumu …

If Kombai relative clauses are indeed GNMCCs, it is nevertheless clear that they differ from Wiru GNMCCs in some respects. First, the set of nouns that can function as a head noun (de Vries's "grammatical head") appears to be a closed set in Kombai (1993:78). Not so in Wiru, where any noun can head a GNMCC and occur with either a copy of that noun in the subordinate clause, or a resumptive pronoun. Second, we have not found any instances in Wiru of a coreferential noun (as opposed to a pronoun) in the subordinate clause that takes a different form from the head noun. This feature is not unique to Kombai; it is also allowed, for example, in Japanese. The following examples from Kuno (1981:237) were, like those of Kombai, identified by Cinque (2011) as examples of "double-headed" relative clauses. Kuno writes that they are "awkward, but not ungrammatical." Note, in particular, (96)b, in which the noun within the subordinate clause differs from the head noun.

(96) Japanese

Returning to Kombai, de Vries also identifies constructions that display the kind of noun duplication that we have observed in Wiru (section 3).

(97) Kombai

 

Another parallel between Wiru GNMCCs and Kombai relative clauses is that the latter can occur without a grammatical head (98). This construction strongly resembles the headless GNMCCs we have described in Wiru (section 4).

(98) Kombai

 

A similar pattern is found in Kewa (Franklin 1971). Here, a noun in a subordinate clause can be coreferential with the head noun, as with áá 'man' in (99), and the head noun can also be omitted.

(99) Kewa

 

Kewa also appears to allow extended semantic relationships to exist between a subordinate clause and its head (100).

(100) Kewa

 

Wiru also exhibits similarities with other highlands TNG languages. In both Korowai (van Enk and de Vries 1997) and Oksapmin (Loughnane 2009), it is possible for extended semantic relationships to exist between head noun and subordinate clause, which suggests that these languages may have GNMCCs.

(101) Korowai

 

(102) Oksapmin

 

Nggem (Etherington 2002) also allows extended modifier relations, but only with the head nouns wene 'message' and adem 'central point' (103).

(103)

 

All three of these languages allow headless relative clauses, as does Mian (Fedden 2011):

(104) Oksapmin

 

(105) Korowai

 

(106) Nggem

 

(107) Mian

 

Of the languages cited above, Wiru, Kombai, and Kewa appear to be the only ones that allow nouns in the subordinate clause that are coreferential with the head noun. While some grammarians make no explicit statement about the allowability of this construction, we do know they are disallowed in Mian (Fedden 2011:501), Oksapmin (Loughnane 2009:197), and Kewapi (Yarapea 2013).

We see, then, that the Wiru GNMCC shares several features with other highlands NMCCs. NMCCs in Kombai appear most similar to the Wiru GNMCC; headlessness is permitted, as well as coreferential head and subordinate clause nouns. There is some evidence to support a GNMCC analysis in some other highlands languages. Headless NMCCs appear quite common in the New Guinea Highlands, both in languages with apparent GNMCCs and in languages without them. We have not yet identified any other language that exhibits the semantic agreement that we described for Wiru in section 6. In addition, we have not found any other mention of the interaction of GNMCCs with tones.

8. CONCLUSIONS

Wiru has a GNMCC: a single construction is used for relative-clause-like interpretations as well as extended semantic relationships between the head noun and its modifying clause. We have demonstrated a number of distinctive features of the Wiru GNMCC. The future tense and habitual suffixes are disallowed in subordinate clauses, with the optative taking over the role of the future and the present taking over the habitual. Wiru GNMCCs may contain a recapitulation of the head noun, or an overt head noun may be entirely lacking. The GNMCC has its own tonal properties that appear to override the lexical tone of its head, when the head is a noun. In addition to third-person reference, Wiru GNMCCs can also have first- or second-person reference. In these instances, the head of the GNMCC is a noun that would usually have a third-person referent. Where such a GNMCC is the subject of a clause, the person marking on the verb must agree with the referent, not the third-person feature of the head noun.

The data used in this paper are limited to just two native Wiru speakers. Further data and speakers will improve our understanding of Wiru GNMCCs. Our typological comparisons demonstrate that GNMCCs may be more widespread among highlands TNG languages than has been realized, and also that the permissibility of headless NMCCs may be an areal feature.