In this squib, I present new data from adjective orders in Ojibwe (Algonquian) that are problematic for Cinque’s (2010) phrasal movement analysis of adjective ordering.1Cinque (2010) claims that adjectives are merged in a universal hierarchy and that phrasal movement (specifically, movement of phrases that contain the NP) displaces adjectives. These movements give rise to the crosslinguistically attested adjective orders. By comparison, a base-generation account of adjective order variation would have to stipulate each attested and unattested order. I demonstrate that Cinque’s analysis cannot account for the crosslinguistic adjective-ordering facts. In particular, Ojibwe exhibits a previously undescribed and unattested adjective order that cannot be derived by Cinque’s analysis. While I do not propose a solution to this problem, I will suggest that a base-generation account can no longer be ruled out.
This squib is organized as follows. In section 1, I outline Cinque’s theory of adjective orders and his arguments against a base-generation analysis. In section 2, I present the relevant data from adjective ordering in Ojibwe and discuss why the data are problematic for Cinque’s analysis. In section 3, I address two alternative analyses of Ojibwe adjectives that could preserve Cinque’s approach; I subsequently reject both. In section 4, I conclude the squib.
1 Phrasal Movement vs. Base-Generation of Adjectives
Following Sproat and Shih (1990), Cinque (2010) proposes that adnominal adjectives have two sources. The first source, direct modification, is lower, ordered, and characterized by APs that are merged in the specifiers of dedicated functional heads (cf. Cinque 1994). The second, indirect modification, is higher, is unordered, and forms predicates in (reduced) relative clauses (in either an IP/TP or a CP). Cinque claims that only phrasal movement (i.e., movement of a constituent that contains the NP) can produce the attested orders of adjectives (and relative clauses) crosslinguistically.
Support for a movement account over a base-generation account of direct modification APs comes from the attested orders of adjectives crosslinguistically. Consider the typology of orders in (1) for the adjective classes of size, color, and nationality (Cinque 2010:38, (33)). Only the orders in (1a,c,d) are attested; the order in (1b) is not.
English, Chinese, . . .
Asize > Acolor > Anationality > N
*Anationality > Acolor > Asize > N
Welsh, Irish, . . .
N > Asize > Acolor > Anationality
Indonesian, Yoruba, . . .
N > Anationality > Acolor > Asize
According to Cinque, the orders in (1) are unexplained under the base-generation account because there is no clear motivation for the base-generation of only some of the logically possible orders. It has been proposed that more abstract adjectives merge farther away from the head noun, which could account for the possible orders in (1a,d) and the impossible order in (1b) (see, e.g., Sproat and Shih 1990). However, under this approach it is unclear how a language would have the order seen in (1c), since the most abstract adjective (Asize) is closest to the noun. Instead, Cinque argues that APs merge in a universal order, as schematized in (2), and that each head supports only one specifier (following Kayne 1994).
(2) [f1p APsize F1 [f2p APcolor F2 [f3p APnationality F3 [np N]]]]
Cinque suggests that either movement of the NP alone or piedpiping of the NP plus a larger constituent can account for the three possible orders, while blocking the fourth, unattested one (cf. Cinque 2005). Cinque’s assumptions about movement are given in (3) (adapted from Abels and Neeleman 2012).
The underlying hierarchy in the extended projection of the noun is Agrw > W > AgrX > X > AgrY > Y > N, where > indicates c-command and where W, X, and Y host APs in their specifiers.
All movements move a subtree containing N.
All movements target a c-commanding position.
Cinque provides two movement patterns to derive the possible adjective orders, which are schematized in (4). I go through the derivations in turn below.
To derive the order size > color > nationality > N in (1a), the NP does not move. The derivations of the other two attested orders both begin with movement of the NP to Spec,AgrYP, as indicated by the dashed arrow. If the NP moves in a specifier-to-specifier fashion to the AgrP projection that dominates each phrase that hosts a direct AP, the result is the order N > size > color > nationality in (1c). The third possible order, N > nationality > color > size in (1d), is derived in a ‘‘roll-up’’ fashion: first, the NP moves to Spec, AgrYP, and then AgrYP pied-pipes to Spec,AgrXP, followed by AgrXP pied-piping to Spec,AgrWP.
Cinque assumes that only NPs can move. Individual APs cannot move by themselves because they do not bear the relevant [+wh] feature (see (3b)). Thus, under Cinque’s analysis, the order in (1b) is underivable because the NP has not moved, yet the order of adjectives is different from the universal order in (2).
I have outlined Cinque’s phrasal movement analysis, and how it can account for both the crosslinguistically attested adjective orders and the adjective order previously thought to be unattested.2 In the next section, I provide previously undescribed data from Ojibwe that are problematic for Cinque’s proposal.
2 Ordering of Attributive Modifiers in Ojibwe
‘a red dress’
Adjectives in Ojibwe exhibit rather free ordering. A color adjective, such as misko- ‘red’, can be freely ordered with respect to a nationality adjective like Anishinaabe- ‘Ojibwe’, as in (6). The same is true for the orders of color and size (e.g., gichi- ‘big’) adjectives, as illustrated in (7).
‘a red Ojibwe bag’
‘a red Ojibwe bag’
‘a big red house’
‘a big red house’
However, (8) illustrates a restriction on adjective orders: size adjectives must precede nationality ones.
‘a big Ojibwe bag’
Intended: ‘a big Ojibwe bag’
A more complex ordering pattern emerges with three adjectives; see (9). The nationality modifier Anishinaabe- ‘Ojibwe’ can precede the size modifier gichi- ‘big’, but only if another modifier, such as misko- ‘red’, intervenes between them, as in (9a). Note that this is a previously undescribed adjective order. Recall that the nationality > color > size order has been thought to be unattested (see (1b)), but (9a) shows that this order is available in Ojibwe. Since Anishinaabe- ‘Ojibwe’ cannot precede gichi- ‘big’ in (8b), the ungrammaticality of (9b) and (9c) is also expected. The remaining possible orders of these three adjectives in (9d–f ) are all grammatical.
‘a big red Ojibwe table’
The adjective-ordering pattern seen in (6)–(9) can be summarized as follows: size and color adjectives are freely ordered with respect to each other; color and nationality adjectives also are freely ordered; but nationality adjectives cannot directly precede size adjectives.4 The data in (6)–(9) show that the order of adjectives in Ojibwe is more flexible than predicted by an analysis based on single specifier positions for each attributive modifier, such as the analysis in Cinque 2010 (see (2)). For example, misko- ‘red’ can have any order with respect to other adjectives, which is not expected under Cinque’s account. Deviation from Cinque’s universal hierarchy should only be possible if the NP (or some constituent that contains the NP) moves. This clearly is not the case in Ojibwe since the position of the noun remains constant. While many of the adjective orders found in (6)–(9) seem to be problematic for a strict universal hierarchy, for space reasons I focus only on the Ojibwe adjective order in (9a).
As (9a) indicates, the order nationality > color > size > N is possible. This is exactly the order that should not be possible under Cinque’s phrasal movement analysis (see (1b)). As mentioned in section 1, the order seen in (9a) is problematic because an adjective would have to move to the exclusion of the noun.5 Thus, the grammaticality of (9a) suggests that an account whereby adjective orders are derived exclusively through phrasal movement, as Cinque proposes, cannot fully account for the crosslinguistic variation in adjective ordering.
3 Adjectives in Ojibwe Are in Direct Modification
In this section, I present two possible options for analyzing the order in (9a) that would obviate the apparent problem it poses for Cinque’s analysis. I conclude that both options are untenable. First, one could argue that adjectives form a type of compound with the noun in Ojibwe (as Bloomfield (1958) first suggested), since they are bound to the head noun. If adjectives in Ojibwe form an [A N] compound, the possible adjective orders would be outside the scope of Cinque’s theory, which is only intended to account for phrasal modifiers. Second, one could argue that adjectives can be merged as (reduced) relative clauses. Under Cinque’s theory, (reduced) relative clauses merge higher than direct modifier APs and are also not rigidly ordered. I will argue that neither option is possible, as adjectives in Ojibwe do not form compounds with the noun they modify, and they do not conform to the diagnostics for relative clauses in Ojibwe or the interpretation of adjectives that are contained in relative clauses.
The first suggestion, that Ojibwe adjectives form a compound with the noun that they modify, is problematic for two reasons. The first piece of evidence against a compound analysis comes from the scope of adjectives in coordinated noun phrases. In (10), gichi- ‘big’ attaches only to the noun in the first conjunct. In this case, the adjective may also scope over the noun in the second conjunct, giving rise to the interpretation in which both the man and the woman are big. The scope facts extend to the color and nationality classes, as seen in (11) and (12).
gichi-inini gaye ikwe
big-man.na and woman.na
‘a big man and (a) woman’ or ‘a big man and (a) big woman’
misko-mishiimin gaye odaabaanens
red-apple.na and car.na.dim
‘a red apple and (a) car’ or ‘a red apple and (a) red car’
Anishinaabe-mashkimod gaye adoopowin
Ojibwe-bag.ni and table.ni
‘an Ojibwe bag and (a) table’ or ‘an Ojibwe bag and (an)
This clearly shows that the scope of the adjective is not restricted to the first conjunct. If adjectives formed compounds with their head noun, we would not expect them to scope into the second conjunct (see Schäfer 2009).
Second, consider the idiomatic noun phrase gichi-mazina’igan ‘Bible’ (lit. ‘big book’), which contains the adjective gichi- ‘big’; see (13a). In (8), we saw that Ojibwe adjectives obey the crosslinguistically attested size > nationality ordering (Truswell 2009, Watanabe 2012). However, (13b) demonstrates that the ordering constraints are violated when the combination of a noun plus an adjective creates an idiomatic expression: the nationality modifier must precede the size modifier that is part of the idiom. If another adjective intervenes between gichi- ‘big’ and mazina’igan ‘book’, as in (13c), the idiomatic reading is lost.
‘an Ojibwe Bible’
= ‘a big Ojibwe book’
≠ ‘an Ojibwe Bible
If the idiomatic noun phrase gichi-mazina’igan ‘Bible’ in (13) is a type of [A N] compound, we would predict that the strict ordering size > nationality would not need to be obeyed if a nationality adjective modified ‘Bible’ since gichi- forms a word with mazina’igan ‘book’. That is, the addition of the nationality adjective to gichi-mazina’igan is expected to be grammatical, and this is borne out in (13b). I claim that the combination of an adjective and a noun as in (13a) is a modificational compound, whereas the other examples of a noun plus an adjective are true instances of direct (attributive) modification. Therefore, one cannot argue that Cinque’s account does not apply to adjectives in Ojibwe because they form compounds with the noun that they modify (for further discussion, see Goddard 1990, Valentine 2001, Mathieu 2013, Barrie and Mathieu 2015).
The second possible alternative is to analyze adjectives as (reduced) relative clauses. Under this account, an adjective is merged as a predicate in a CP or IP/TP. Indirect modification structures are merged in a functional projection above direct modification, and modifiers in this structure have free ordering with respect to each other (cf. Sproat and Shih 1990). Cinque (2010) admits this possibility for other sets of data within his theory. Thus, if adjectives in Ojibwe were in an indirect modification structure, then the order in (9a) simply would not be relevant to Cinque’s claim that phrasal movement alone derives adjective orderings.
This potential analysis is a bit harder to refute. However, certain facts about Ojibwe relative clauses and the interpretation of adjectives indicate that adjectives in Ojibwe should not be treated as indirect modifiers. First, consider the relative clause in (14), where waa-aabjitoowaad ‘which they were going to use’ follows the head noun mshkikwaaboo ‘liquid medicine’.
Mii dash gii-zhitoo-waad iw mshkik-waaboo
and then pst-make-3pl that medicine-liquid
‘They made the liquid medicine which they were going to use.’
(Odawa; Valentine 2001:582)
In Ojibwe, interrogative and relative clauses show the presence of ablaut (called ‘‘initial change’’ in the Algonquianist literature) on the left edge of the verbal complex. The future prefix, which appears as wii- in declarative and matrix clauses, has changed to waa- in the relative clause in (14).6 Following work by Lochbihler and Mathieu (2013) on Ojibwe, I take this ablaut process to constitute a form of wh-agreement on I/T that is inherited from C.7
These data suggest that adjectives are not predicates in relative clauses, as there is no local I/T to trigger ablaut on adjectives (Lochbihler and Mathieu 2013). This rules out the possibility that adjectives in Ojibwe are in CP-sized relative clauses.8
Second, if adjectives in Ojibwe were predicates in reduced relative clauses, then Cinque’s (2010) account would predict that they must have an intersective reading. Cinque crucially relies on such interpretations to distinguish between direct modifiers and indirect modifiers in reduced relative clauses in English and Italian. Consider gete- ‘old, former’ in (16).
‘an old worker’ or ‘a former worker’
= ‘A worker is old.’
≠ ‘She or he is no longer a worker.’ (‘A worker is former.’)
The adjective gete- can have the nonintersective interpretation ‘former’ when it appears in prenominal position (see (16a)) but not when it is the predicate of the clause (see (16b)). Since indirect modification adjectives would have the same interpretation as predicate adjectives under Cinque’s account, the data in (16) are not readily explained under a relative clause analysis of Ojibwe adjectives.
Adjectives in Ojibwe do not undergo the ablaut process that other relative clauses undergo, and Ojibwe has nonintersective adjectives like gete- ‘former’. Thus, I conclude that it is untenable to claim that Ojibwe adjectives are in relative clauses.
Furthermore, Cinque (2010) notes that direct modification adjectives often form a small closed class and that they frequently appear as nominal prefixes. These two characteristics are consistent with Ojibwe adjectives: there are around twenty attributive adjectives, and as noted above, they are bound (as prefixes) to the noun. Despite their bound nature, adjectives must head phrases since they can be modified by gichi-. Above, we saw that gichi- means ‘big’, but it can also be an intensifier ‘very’. Both uses are seen in (17).
‘a big very red table’
Overall, the data point to the conclusion that these adjectives are in a direct modification structure; that is, they are bare APs. Possible alternative solutions that appeal to either a compounding or an indirect modification structure are not tenable for Ojibwe adjectives. The conclusion that adjectives are bare APs in Ojibwe confirms that the example in (9a) is problematic for Cinque’s (2010) phrasal derivation of adjective orders.
In this squib, I presented novel data from adjective ordering in the Algonquian language Ojibwe. I showed that the data are problematic for an account that relies solely on phrasal movement to derive variability of adjective orderings crosslinguistically. In order to derive the correct order, one must violate one of Cinque’s (2010) necessary assumptions— namely, that only phrases containing the NP can move. However, violating this assumption would weaken Cinque’s argument for phrasal movement over base-generation, since one could argue that the crosslinguistically attested adjective orderings are accidental under both accounts. If both NP-movement and AP-movement were possible, then all orders would be derivable, contrary to fact. Thus, both a basegeneration account and a movement account would make the same predictions: all logically possible orders could be derived if both NPs and APs could move to create the various surface adjective orders. It is not entirely clear how to rule out the unattested ones. On this basis, a base-generated account remains a viable option. I will not attempt to pursue the larger consequences of these data for Cinque’s proposal; however, the data in this squib are problematic for his phrasal movement analysis.
I am extremely grateful to my Ojibwe consultants (E.O. and S.J.) for teaching me about their language. Special thanks to Meredith Johnson and Yafei Li for invaluable comments and guidance throughout this project. Thank you to Monica Macaulay, Will Oxford, Mateja Schuck, Becky Shields, Rand Valentine, and the audiences at Chicago Linguistic Society 49 (April 2013) and the 46th Algonquian Conference (October 2014) for helpful feedback. This squib also greatly benefited from the comments of the anonymous LI reviewers.
1 Ojibwe is an Algonquian language spoken in Michigan, Minnesota, Montana, North Dakota, and Wisconsin (United States) as well as central and eastern Canada. There are many dialects of Ojibwe; the speakers consulted here speak a dialect that is on the border of the Oji-Cree and Northern Ojibwe regions in Ontario, Canada.
2Cinque (2010:41) notes that the phrasal movement account also provides a means to derive the fact that the adjectives that are postnominal in Romance languages are prenominal in Germanic languages. I do not address this claim here.
3 In the Algonquianist literature, these adjectives have been called ‘‘prenouns’’ since they appear directly before nouns (see also Oxford 2007 for Innuaimun). Other Ojibwe words that correspond to English adjectives are encoded as verbal predicates. All examples are elicited as objects of ‘make’, ‘buy’, or ‘see’ in pragmatically neutral contexts. My methodology follows the standard techniques of translation and acceptability judgment tasks (see Matthewson 2004 for more details).
The following abbreviations are used in the examples: 3 = 3rd person; conj = conjunct; dim = diminutive; fut = future; ic = initial change; na = animate noun; ni = inanimate noun; pl = plural; pst = past.
4 For an explanation of the ordering facts, see Rosen 2013. Under that account, adjectives in Ojibwe divide into three independently motivated classes, and each adjective class has its own selectional restrictions. In Rosen 2013, I assumed for the purposes of exhibition that adjectives are heads; however, the analysis can be straightforwardly adjusted to fit a phrasal account of adjectives.
5 It is worth pointing out that a nationality adjective can precede both color and quality adjectives in Ojibwe, as in (i).
‘a new black English house’
6 In Ojibwe, initial change (ablaut) has seven alternations: i > e; a > e; o > (w)e; ii > aa; aa > yaa; e > ye; oo > waa (Valentine 2001:155).
7 Interrogative and embedded verbs also bear a special set of agreement markers, known as conjunct order (conj) in the Algonquianist literature. Previous analyses of conjunct morphology in Algonquian languages have proposed that it appears when the verb has a relationship with C: the verb moves to C to check a conjunct feature [CJ] (Brittain 2001 for Western Naskapi); the verb does not move to C, which results in antiagreement with a wh-word (Richards 2004 for Wampanoag); or C possesses discourse features rather thanLochbihler and Mathieu 2013 for Ojibwe), among others.
On the other hand, adjectives in Ojibwe are morphologically marked by the vowel /i/, and the appearance of other vowels at the end of adjectives is often the result of independent phonological processes in the language (Valentine 2001). For example, Ojibwe has a rule whereby i and w coalesce to o. The adjective misko- ‘red’ has the underlying form /miskw-/ (e.g., miskw-aa ‘it (inanimate) is red’; Valentine 2001); thus, the final -w and the adjective producing -i combine to form -o. Following Slavin (2012), I treat this -i as the realization of a ‘‘little a’’ head of aP (an adjective) in Distributed Morphology (Marantz 1997).
8 One might wonder if adjectives in Ojibwe are always in relative clauses and thus already have the changed form. Then the forms in (15) would be ungrammatical because they cannot undergo a second process of ablaut. However, this is not tenable either since the ablaut process never yields the vowel [a] (e.g., Anishinaabe- ‘Ojibwe’).