Head Movement and Noun Incorporation

Abstract

Head movement (HM), once a mainstay of generative syntax, has undergone a tumultuous series of overhauls over the years (Baker 2009, Boeckx and Stjepanović 2001, Chomsky 2001, Fanselow 2003, Mahajan 2003, Matushansky 2006, Roberts 2010). The problems with HM have been discussed in the literature just cited, and we do not comment further on them here. Rather, we wish to address the issue of HM and noun incorporation (NI) in light of Roberts’s (2010) recent reworking of HM. In a nutshell, we show that the mechanism for HM adopted by Roberts (who uses NI to support his agreement proposal), while well-suited for capturing other properties Roberts discusses such as clitics in Romance and Slavic languages, fails to capture various properties of NI in several languages—notably Fox and Ojibwe (Algonquian) and various Northern Iroquoian languages (Mohawk, Onondaga, and Oneida), though we discuss other languages below. We do not argue that Roberts’s approach for HM is on the whole untenable; we merely contend that it cannot be the right analysis of NI.1This squib is organized as follows. In section 1, we give some background to HM and NI. In section 2, we outline the mechanics of HMas presented by Roberts, along with Roberts’s preliminary analysis of NI in Mohawk. In section 3, we present the relevant properties of NI in the languages mentioned above. In section 4, we discuss the problems facing Roberts’s proposal for HM and NI. In section 5, we present our conclusions and the implications of the current discussion for NI.One holdover from the Government-Binding era is the distinction between HM and phrasal movement, which persists to this day (Aboh 2004, Baker 2009, Roberts 2010). However, with the demise of any formal distinction between X0 and XP and the introduction of bare phrase structure (Chomsky 1995), the concept of HM has been called into question (Fanselow 2003, Koopman and Szabolcsi 2000, Mahajan 2003). Chomsky (2000, 2001) suggests relegating HM to PF, but leaves open the possibility that incorporation phenomena may still be part of the overt syntax (Chomsky 2001:37). NI certainly cannot be a purely PF phenomenon given its semantic effects, such as frozen scope (Baker 1996, Van Geenhoven 1998), and syntactic properties, such as changes in argument structure (Sadock 1986). In response to the growing skepticism regarding the existence of HM, Baker (2009) reaffirms that HM indeed is still needed for NI, but he does not address how NI can be implemented in a bare phrase structure framework.Roberts (2010) recasts HM as a reflex of Agree between a probe and a defective goal, bringing it in line with mainstream minimalist syntax. He takes a defective goal to be one whose features are a proper subset of the features of the probe. Following Roberts, we illustrate the operation with Romance clitics. Roberts assumes that an active, transitive v* has an interpretable V-feature, [iV] (to categorize the lexical root), and [uΦ]. Following Cardinaletti and Starke (1999) and Déchaine and Wiltschko (2002), he further assumes that a clitic is a bare ΦP devoid of Case features—in other words, a bundle of interpretable Φ-features, [iΦ]. When the V-v* complex enters into an Agree relation with the clitic, the [uΦ] feature set on v* is valued by the clitic (Roberts 2010: 60, (30)).Roberts argues that there is no formal difference between Agree that exhausts all the features of the goal, on the one hand, and Move/ internal Merge of the goal to the probe, on the other. In other words, the outcome of (1b) Agree(v*,Φ) is indistinguishable from Merge(v*,Φ) precisely because the set of features on Φ is a subset of the features on v*. Contrast the two structures in (2). In both cases, the same set of features appears on v* after either Agree or Move.Finally, HM fails with a full DP object because it contains features that are not part of v* (such as D and Case).Roberts reviews Baker’s (1996) properties of polysynthetic languages and proposes that NI can be captured under the same rubric as cliticization, thus following the tradition started by Baker (1988) in treating NI as HM. Roberts considers the following example of NI in Mohawk (Baker 1996:12, (9)):2The relevant probe in Mohawk (and, we assume, in all languages with NI) is a v* endowed with a D-feature, which Roberts argues is present because of rich object agreement (just as T has a D-feature for languages with rich subject agreement). NI, then, proceeds as follows. The nominal root incorporates into a nominalizer head (N-n), just as V raises to v, endowing the nominal expression with an N-feature.3 Roberts further makes the reasonable assumption that a D-feature entails an N-feature. Thus, the set of features on the goal, {[N]}, is a subset of the features on the probe, {[D], [V], [uΦ]}, given that [D] entails [N]. Roberts suggests, leaving various questions unanswered, that once the v* probe enters into an Agree relation with the goal (the incorporated nominal), the goal necessarily incorporates since its features are a subset of the probe’s. This is because, as with clitic movement above, once v* enters into the Agree relation with the N-n complex, the result is indistinguishable from actual movement, so incorporation ensues.We have sketched Roberts’s analysis of HM and how it applies to NI. In the following section, we describe some further properties of NI and then continue to explain how they are problematic for Roberts’s proposal.We briefly describe the facts of NI relevant to the current discussion, concentrating mainly on Ojibwe (Algonquian) and Northern Iroquoian; however, we do introduce data from other languages where they are pertinent to the discussion (for a fuller discussion of NI, see Gerdts 1998, Massam 2009). Consider first two examples of NI in Onondaga (Northern Iroquoian).(4a) shows a bare root incorporated into the verbal complex, and (4b) shows the incorporated element (adogw ‘spoon’, which derives from a verbal base meaning ‘scatter’) with a nominalizer. While in many languages the incorporated nominal (IN) can consist only of a bare root, it is quite common for nominalizers to appear as part of the IN complex. Ojibwe also exhibits NI with overt nominalizers, as shown in (5).4Furthermore, the IN in Ojibwe can be modified, as shown in (6).Note that this is not simply a lexicalized compound. While it’s conceivable that ‘big breakfast’ might be common enough to be stored as a whole, other modifiers that can incorporate include numerals, which strongly suggests a nonlexical source for these constructions. See Sadock 1980:316 for similar facts in Greenlandic.In Fox (Algonquian), this type of incorporation can also involve nouns that are possessed, as illustrated in (7).5In Fox, it is also possible for demonstratives to incorporate into the verb, as (8) shows. On the reasonable assumption that demonstratives are Ds, then clearly they are not defective.6 They nevertheless can incorporate.Finally, we illustrate NI into light verbs. This kind of NI has been studied extensively by Johns (2007) for Inuktitut. To illustrate, in Ojibwe (9) the noun naboob ‘soup’ merges with the morpheme -ke, a verbal form that means ‘make’, ‘build’, ‘hunt’, ‘pick’, ‘do’, and so on. The verbal form has a very general meaning akin to an auxiliary and unlike a lexical verb (see Mathieu, to appear, for a full description of these denominal verbs).7Ojibwe nominals in such incorporating constructions can also contain diminutive and pejorative morphology, (10), and number (singular)/ gender (animate) morphology, (11), both showing again that the incorporated element is much larger than a root or N-n complex.To sum up, we have shown that NI constructions can incorporate material beyond a nominal root and a nominalizer. In particular, we showed that the IN can be modified by adjectives or can appear with additional inflectional material such as number or possessive markers.Given that Roberts (2010) provides the most viable revision of HM within a bare phrase structure perspective, it behooves us to examine the ramifications of this proposal for NI. As it turns out, Roberts’s account of NI faces several problems. Recall that for NI to proceed as HM, it is crucial that the IN not be distinct from the probe, v*, and that v* consist of active object agreement to license the incorporation of nominal material. Crucially, we find crosslinguistically that (a) NI involves incorporation of distinct material, and (b) NI often lacks object agreement.The NI constructions in Algonquian presented in (6)–(8) are problematic because they include material not contained in the feature makeup of the v* probe. Spencer (1995:481) also discusses data from Chukchi where INs are modified. Moreover, the IN cannot be said to be a clitic (a defective element) since unlike a clitic it is perfectly capable of appearing separately from the verb, as in (12) (Ojibwe) and (13) (Mohawk).The case in (14) shows the same result with a different kind of construction. Adverb incorporation (AdvI) is found in a variety of languages, including Greek (Alexiadou 1997, Rivero 1992). Consider the following minimal pair. In the non-AdvI form, the freestanding adverbial is marked with adverbial morphology. In the AdvI form, the lexical root appears with a linking vowel (Ralli 2003).Ralli suggests that the linking vowel may be a morphological element present in compounds. If so, this root+LINK combination is not a defective goal in Roberts’s terms, as it contains whatever features are present on the linker that are not present in the probe, v*.A clearer case of AdvI with additional functional material comes from Blackfoot (Algonquian). Consider the following example, where the incorporated adverb appears with a degree word:If we follow Roberts (2010), we could argue that the nouns in (5) are formed via the merger of a nominal root, N, with n and that it is this N-n complex that merges with the lexical verb. However, INs in such contexts can be modified as shown in (6), thus endowing the IN with an Adj categorial feature (akin to N and V), which is not present on the v* probe. The examples of AdvI above also contain features not present on the probe—namely, an Adj categorial feature (assuming adverbs are built out of adjectives), and, in the case of Blackfoot, a Degree feature.Next, we turn to the issue of lack of object agreement. For Roberts’s proposal it is crucial that object agreement is found on the v* probe in order to ensure that the goal’s (the IN’s) features are nondistinct from the probe’s. It is well-known that Northern Iroquoian languages have robust subject and object agreement. However, Northern Iroquoian languages typically lack any kind of agreement with inanimate objects (see Koenig and Michelson 2008 for the source of the argument developed here). Transitive forms exhibit both subject and object agreement that does not vary with aspect. Intransitive forms typically have subject agreement with their single argument (although there are many lexical exceptions) in nonperfective aspects and object agreement in the perfective (stative in Iroquoianist terms) (Lounsbury 1949). Consider the following example with NI:The fact that the notional subject of the sentence is referenced by patient agreement shows that this sentence is morphosyntactically intransitive. If there were a phonologically null agreement marker referencing the object, then we would see agent agreement on the subject.Consider finally the Onondaga example of possessor raising in (17) (see Michelson 1991 for more details). Here, verbal agreement references the notional subject and the possessor, but not the IN. Northern Iroquoian languages can reference a maximum of two arguments. Since both agreement markers are used up by other arguments, the IN cannot be referenced by any agreement marker. In other words, v has a D-feature (and hence an n-feature), but it does not agree with the IN; it agrees with Mary.To sum up, in this section we have provided numerous illustrations of incorporation of nondefective material, such as AdvI and NI with modifiers forming part of the incorporated element. We also showed that agreement in Northern Iroquoian is in complementary distribution with NI (as originally argued in Baker 1996). Both observations lead to the conclusion that Roberts’s mechanism for head movement is not tenable for NI.HM continues to play an active role in syntactic theorizing. Roberts (2010) offers an innovative and viable solution to many of the problems raised by HM in Universal Grammar. When this solution is applied rigorously to NI, however, many insurmountable problems arise. It is crucial for Roberts’s mechanism of HM that the goal’s formal features be a subset of the probe’s. Roberts argues that NI occurs when a D-feature appears on the v* probe, which is present as a result of rich object agreement in languages with NI. The D-feature entails an N-feature, found on the IN (the goal), thereby ensuring that the goal’s features are nondistinct from the probe’s. We showed that numerous languages with productive NI are capable of incorporating distinct material such as adverbs (Chukchi, Greek, Blackfoot) and adjectives that modify the IN (Ojibwe, Chukchi). Furthermore, we showed that in Northern Iroquoian, the requisite object agreement is crucially absent in NI constructions, or, if present, does not agree with the IN; rather, it agrees with another argument.Although NI has a long tradition of being analyzed as HM (Baker 1988), the discussions mentioned at the beginning of this squib have sparked serious reconsiderations of HM. Although Roberts (2010) has proposed a viable and coherent solution to many of the problems associated with HM, we have shown that this solution will not work for NI. We suggest, then, that other avenues must be pursued for analyzing NI, such as XP-movement (Mathieu 2008) or roll-up movement (Barrie 2010).8

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 37 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.