An argument analysis of cognate objects in Dschang (Yemba)

Focusing on the Foto dialect of Dschang (Yemba), an understudied Grassfields Bantu language spoken in Cameroon, this paper offers a cross-linguistic perspective on Cognate Objects (CO). An argument analysis of Dschang COs is supported by both cross-linguistic comparison, e.g. forms of corresponding whquestions, the compatibility with strong determiners, quantifiers and possessors, and the ability to be pronominalized and relativized, and Dschang-internal evidence including word order variations and tonal marking in object position.

b. à lè ! khʉ́-ʉ̀ lè-khʉ̀ 3SG DST.PST run-OM5 5-run 'S/he ran a run.' The rest of the paper is organized as follows. Section 2 presents an overview of Dschang. In particular, we will introduce patterns of object concord and verbal tone in Dschang. Section 3 gives an overview of verb ~ noun alternation in Dschang. Argument properties of Dschang COs are laid out in section 4, where we look at both cross-linguistic comparisons and Dschang-internal evidence. Section 5 concludes the paper.
2. Background of Dschang. This section will provide some basic background information about Dschang. The language has basic SVO word order, as shown in (3).
(3) nìŋ lè lá-à m̀-bàp man DST.PST cook-OM9 9-meat 'The man cooked the meat.' 2.1. TONE. Like all Grassfields languages, Dschang is a tone language. In pre-pausal position, there are four contrastive tones: High, Low Level, Low Falling, Downstep High. The four surface contrasts derive from two phonological tones, High and Low (Hyman & Tadadjeu 1976, Hyman 1985. For simplicity, in this paper, we mark both types of low tone in the same way, as the contrast between low level and low falling tones does not play a role in the analysis of cognate objects.  (Harro & Haynes 1991) 2.2. NOUN CLASSES. Dschang has a subset of the noun classes found in Bantu. Since there is some amount of homophony of noun class prefixes, the class of many nouns in Dschang can only be determined by the form of concord elements, in addition to the class prefix. Nouns from other classes pair with one of class 2, 6, or 8 to form a number of genders. The noun classes are presented in Table 2, along with examples of a member of that class.

Class Prefix
Concord  (Tadadjeu 1975, Hyman 1980. The morphological structure of some of these tenses appears to derive from earlier auxiliary constructions, but they are synchronically treated as tenses. Examples of some of the tenses are given in (4  CONCORD. In addition to noun class prefixes and deictic concord elements (found in possessives and demonstratives), there is a set of object concord markers that appear between the verb and object. The form of the concord element depends on the noun class of the object, and can be either segmental and tonal, or only tonal. These two possibilities are demonstrated in (5): (5) a. à lè tɔŋ-ɔ́ ! séŋ 3SG DST.PST call-OM1 1.bird 'He called the bird (some time ago).' b. à lè tɔŋ-̀ m̀-! bhʉ́ 3SG DST.PST call-OM9 9-dog 'He called the dog (some time ago).' When the object of 'call' is the class 1 noun séŋ 'bird', a high-toned vowel with the same quality as the vowel in verb root appears between the verb and object. When the object of the verb is a class 9 noun like m̀!bhʉ́ 'dog', a low tone, and no vowel, occur instead. In this case, the verb is pronounced as a single syllable with a falling tone. Object concord in the Foto dialect has not yet been systematically studied, but for the Bafou dialect, Harro and Haynes (1991) provide the following inventory of object concord elements:

Segment Tone
Noun class a H 1 a L 7 e L 2, 3, 5, 6, 8 Table 3. Inventory of object concord elements (Harro & Haynes 1991) Harro and Haynes discuss some processes of assimilation that determine the surface form of the object concord vowels. For our speaker, the segmental portion of the object concord marker, other than that for class 1, is nearly always absent. For class 1 objects, the vowel typically appears as additional length on root-final vowel, or as a copy of the root vowel, for consonant-final roots.
These object concord elements disappear (or the distinction between them, either tonal, segmentally, or both) under negation. A low-tone object concord in an affirmative sentence (6a) fails to appear when that sentence is negated (6b): (6) a. à lè tɔŋ-̀ m̀-! bhʉ́ 3SG DST.PST call-OM9 9-dog 'He called the dog (some time ago).' b. à lè tè tɔ́ŋ ḿ-! bhʉ́ ʉ́ 3SG DST.PST NEG1 call 9-dog NEG2 'He did not call the dog (some time ago).' The relation of both object concord and negation to the status of cognate objects in Dschang will be discussed in greater detail below.

Overview of verb~noun alternation.
There is a productive verb ~ noun alternation in Dschang. Nouns derived from verbs consist of the verb root, plus a noun class prefix. Noun class prefixes on cognate nouns are drawn from the same set used to mark other nouns. There are a number of different types of nominal derivation in Dschang, with each type taking a different noun class prefix (see Harro & Haynes 1991: 13-15 for more detail). 3 Some examples of cognate nouns are given in Table 4. 4 Infinitive Noun Gloss Cognate nouns may also appear in subject position: (9) a. lè-shʉ̀' è fò lè ! pɔŋ-ɔ́ ! gá 5-arrival ASSC chief DST.PST be.good 1.OBJ 'The chief's arrival made me happy.' b. lè-wǐ ń ! nɛ́ fòtò 5-smile:COP inside photo 'There is a smile in the photo (lit. a smile is in the photo).' 3.2. RESTRICTIONS ON COGNATE OBJECT CONSTRUCTION. One well-known crosslinguistic restriction on the distribution of cognate objects concerns the type of verb that may occur in the cognate object construction. Kuno and Takami (2004) propose the unergative restriction on the cognate object construction. As demonstrated for English in (10), only unergative verbs may appear in the cognate object construction, and no unaccusative verbs may.

Argument properties of Dschang COs.
In this section, we lay out the argument properties of Dschang COs. On the one hand, we compare COs in Dschang and their counterparts in several other languages. In particular, we look at their compatibility with determiners, whether they can be pronominalized (in the object position), whether they can undergo relativization and topicalization, and how they can be questioned. On the other hand, we also present language-internal evidence for the argument status of Dschang COs. More specifically, we argue that COs and regular noun objects show parallel distributions in Dschang regarding tonal object concord and word order variations.

CROSS-LINGUISTIC COMPARISONS.
To address the question whether COs are arguments or adjuncts in a given language, previous studies have adopted various syntactic diagnostics (Massam 1990, Moltmann 1990, Pereltsvaig 1999, 2002, Akkuş & Öztürk 2017. In this section, using these diagnostics, we argue that Dschang COs systematically show properties of arguments compared to COs in languages including Russian, Hebrew, Sason Arabic and English. To begin with an obvious fact, as we have already seen in (7) and (8), Dschang CO constructions do not require obligatory modification, unlike their English counterparts: (12) English (Kuno & Takami 2004: 121) a. * Mary laughed a laugh at the meeting.
b. * Bob grinned a grin. c. * Bill sighed a sigh.
Next, we will illustrate the argument properties of Dschang COs using five cross-linguistically adopted diagnostics. The first one concerns whether they are compatible with determiners. Pereltsvaig (1999Pereltsvaig ( , 2002 argues that Russian and Hebrew have both argumental and adverbial COs, and one difference between these two types of COs is related to their compatibility with determiners. For instance, in Russian, some COs are marked with accusative case whereas others are marked with instrumental case, and it is proposed that accusative-marked COs behave like arguments whereas instrumental-marked ones are adjuncts. One piece of evidence comes from the fact that only the former (13a) but not the latter ones (13b) are compatible with demonstratives and quantifiers: (13) Russian (Pereltsvaig 1999: 13b, 15b) a.  (15) suggest that none of these constraints apply to COs in Dschang: they can co-occur with demonstratives, possessors and quantifiers.
(16) Russian (Pereltsvaig 1999: 18) a. Notice that in (16b), the pronoun ego remains in the same (object) position as the CO tanec in (16a), a pattern that is claimed to be impossible in English, where a subject-object asymmetry is found: COs can be antecedents of subject pronouns but not object pronouns (Matsumoto 1996).
(18) English (Matsumoto 1996: 26a, 32a) a. Mary smiled a mysterious smile and it was attractive. b. * Mary smiled a beautiful smile and Jane smiled it, too. COs in Dschang behave similarly to accusative-marked COs in Russian: they can be pronominalized in the object position. (20a) illustrates a case where a regular noun object like ŋgap 'chicken' is interpreted as the antecedent of an object pronoun yi in the second sentence, and this is equally available for COs as well (20b-c).
(20) a. Mary lè lá-à ŋ̀-! gápi, John láà yii sɛ. Mary DST.PST cook-OM9 9-chicken John cook 3SG too 'Mary cooked the chicken, (and) John cooked it too.' 6 It is worth mentioning that the ungrammaticality of (19b) is not related to the issue whether COs in Sason Arabic can occur in the left periphery. As mentioned in ( The third property is related to relativization. One piece of evidence for an argument analysis of English COs is that they can be relativized with a gap in a non-predicate position within the relative clause (Massam 1990): (21) English (Massam 1990: 16) a. Mona smiled a sarcastic smile, which John photographed ___. b. Elsie prayed a prayer, which my father wrote ___.
Similarly, (22) (21), other properties of English COs have led to the conclusion that they are adverbial. One such property is that they cannot be topicalized: (23) English (Moltmann 1990: 9) a. * A shrill scream, John screamed. b. * A painful death, John died. This is, however, not an issue for Dschang COs: like a regular noun object niŋ 'man' (24), they can also undergo topicalization and occur in a left peripheral position (25).
The last property concerns how a CO can be questioned. One argument for the analysis that Sason Arabic COs are adjuncts is that, unlike regular noun objects, they are questioned with ıştaba 'how', but not şıne 'what'. To answer a 'what' question like (27a), both regular (noun) objects like 'Mary' or 'me' can be used and are interpreted as goal arguments, and more importantly, the CO l�-wì is also a possible answer. Another example of 'what' question-answer pairs involving only CO answers is shown in (28), where answers alternative to the unmodified CO lɨ-khʉ̀ 'the/a run' consist of the same CO but with a (possessive) modifier.
5-laugh 5-laugh 1-chief 'A run / a run dedicated to the chief (lit. a run of the chief).' Hence it is clear that the above cross-linguistic comparison demonstrates that, compared to their counterparts in English, Russian and Sason Arabic, Dschang COs consistently show properties of argument: they (i) can occur independently from modifiers, (ii) can co-occur with various types of determiners including demonstratives, quantifiers and possessors, (iii) can be the antecedents of object pronouns, (iv) can undergo topicalization and relativization, and (v) can be questioned with 'what'.

DSCHANG-INTERNAL EVIDENCE.
In this section, we turn to language-internal evidence for the argument status of Dschang COs. More specifically, we show that there exists a strong parallel between regular noun objects and COs with respect to verbal tone and word order variations. In the first part of the section, we focus on evidence from tone and object marking. Tone on intransitive verbs in cognate object constructions is the same as verb tone in transitive sentences, but different from those with post-verbal adjuncts. More specifically, tonal object concord (see §2.4) in cognate object constructions is identical to that found with canonical direct objects in Dschang.
As shown Table 3, the object concord marker for class 5 nouns and class 6 nouns is identical. In our speaker's dialect, both are marked by a low tone intervening between the verb and the direct object. In (29a), we see that in the cognate object construction, a low tone occurs between the verb ! wí and the class 5 cognate object lè-wì. This is identical to how a class 6 object mè-séŋ is marked in (29b), when it is the object of the homophonous transitive verb 'laugh at'. The tonal object marking in the cognate object construction contrasts with the marking for a 1 st person singular pronominal object (class 1), (29c). In contrast, when followed by an adjunct, wì shows a different pattern. In both the affirmative ) and negative ), it has a level downstep high tone, and never triggers donwstep (in contrast with (29c)). 8 In (32), the verb wì 'laugh' is followed by a non-argument mè tʉ̀ e 'loudly (lit. with strength). The tone at the right edge of the verb does not change -it is a consistent downstep high tone. This is homophonous with the pattern for class 1 objects, but distinct from the pattern for non-class 1 objects, including cognate objects, as shown in (30) The second part of this section presents evidence from the perspective of word order variations. Although, as mentioned in section 2, the basic word order in Dschang is SVO, we do see some OV variations. One such context involves negation. To begin with, (33a) illustrates a typical indicative sentence in Dschang. In (33b), we see one strategy for negation. An invariant morpheme tè occurs preverbally and an additional negative morpheme is realized at the right edge of the clause, as a vowel and high tone after the post-verbal object. 9 (33) a. John lè lá-à ŋ-gáp V-O John DST.PST cook-OM9 9-chicken 'John cooked chicken.' b. John lè tè láá ŋ-! gáp ə́ NEG1-V-O-NEG2 John DST.PST NEG1 cook 9-chicken NEG2 'John didn't cook chicken.' c. John lè té ŋ-! gáp lá NEG1-O-V John DST.PST NEG1 9-chicken cook 'John didn't cook chicken.' Notice that (33b) has the canonical VO order. In (33c), however, the object occurs in a preverbal position when the right edge negation is absent, only the invariant negation tè is used. Interestingly, COs in Dschang show a parallel VO~OV variation in negative indicative sentences: when bipartite negation is used (34b), COs occurs post-verbally and they become preverbal when the right edge negation is absent (34c).
(34) a. John lè ! khʉ́-ʉ̀ lè-khʉ̀ V-CO John DST.PST run-OM5 5-run 'John ran a run.' b. John lè tè ! khʉ́ʉ́ lè-khʉ̀ ʉ́ NEG1-V-CO-NEG2 John DST.PST NEG1 run 5-run NEG2 'John didn't run a run.' c. John lè té lé-khʉ̀ khʉ̀ NEG1-CO-V John DST.PST NEG1 5-run run 'John didn't run a run.' It is worth pointing out that such parallel between COs and regular noun object regarding word order is not always seen across other languages. For instance, one of the arguments for an adverbial analysis of Sason Arabic COs is their distributional differences compared to regular noun objects (Akkuş & Öztürk 2017). The default position for non-specific regular noun objects is post-verbal (35a). However, this is not possible for COs (35b). This is one of a number of properties that should be further explored in the analysis of the Dschang cognate object construction. However, the evidence discussed throughout the paper shows that cognate objects have the properties of arguments, rather than adjuncts.