Deconstructing subcategorization: Conditions on insertion versus position Laura Kalin and Nicholas Rolle (Princeton University) — Draft July 2020 1 Introduction The notion of subcategorization has been utilized for decades to account for certain idiosyncratic behaviors of lexical items. We hone in on one particular use of the term subcategorization, as relates to properties of individual exponents (morphs); see e.g. Lieber 1980; Kiparsky 1982; Selkirk 1982; Inkelas 1990; Orgun 1996; Paster 2005, 2006, 2009; Yu 2007; Bye and Svenonius 2012; McPherson 2019; inter alia. As Paster (2009, 21) puts it, the basic idea is that “affixation is a process that matches an affix with missing elements”, which must be present in the affix's local environment, as specified in the affix's “subcategorization frame”. Proposals have extended to utilize subcategoriza- tion for exponents of roots in addition to affixes, and to include virtually all linguistic categories as possible elements in a subcategorization frame, most relevant here being morphological features (e.g., [+past], [+latinate], stem vs. root) and phonological features, both segmental (e.g., C, V, [+labial]) and prosodic (e.g., phonological phrases and phonological words). Across these proposals, subcategorization has been used to subsume (at least) two types of restrictions on exponents. The first type—which we refer to as conditions on insertion (here- after COINs)—constrain when/whether an exponent is allowed to be inserted. The second type— conditions on position (hereafter COPs)—regulate where an exponent is positioned, in particular, when it does not appear in its otherwise expected place.1 Our goal in this squib is to argue that— even restricted to the fine level of granularity of regulating individual exponents—subcategorization must be formally separated into two distinct mechanisms; COINs and COPs cannot be collapsed. We show that (i) COINs and COPs differ typologically with respect to the sorts of elements in their frames, (ii) an exponent may have both a COIN as well as a distinct COP, and (iii) COINs hold at a derivationally earlier point than COPs. Our findings have implications for the architecture of the grammar, at both the morphology-syntax and morphology-phonology interfaces. This squib is organized as follows. We first review the relevant ways subcategorization is employed in the literature (x2), and show that this raises the question of how the grammar knows when to treat a subcategorization frame as a COIN or a COP (x3). From there we argue for the bifurcation of subcategorization (x4) and discuss the implications of these findings (x5). 1We have adapted this whether/where framing from Paster (2009, 19-20). 1 2 Background In the literature, subcategorization has a dual function. On the one hand, subcategorization is used to constrain the insertion of an exponent (our condition on insertion/COIN). On the other hand, it is also used to regulate an exponent's (idiosyncratic) position (our condition on position/COP). A COIN places an environmental pre-condition on an exponent. Consider, for example, the 3rd person possessive prefix in Tzeltal (Mayan), which has two phonologically-conditioned suppletive al- lomorphs: s- before C-initial stems (e.g., s-mul, `his sin') but y- before V-initial stems (e.g., y-ahwal, `his ruler'), as stated in (1). Throughout, we adopt a simple and intuitive notation for subcatego- rization frames—[feature] exponent : condition—followed by a description in prose.2 (1) Segmental COINs in Tzeltal (Paster 2006, 59, citing Slocum 1948, 80) a. 3.poss y- : V(« “y must be before a vowel”) b. 3.poss s- : C(« “s must be before a consonant”) COINs are commonly used to account for suppletive allomorphy, as above, as well as for morphologi- cal compatibility and morphological gaps (see, e.g., Lieber 1980; Jensen 1990; Booij and Lieber 1993; Halle and Marantz 1993, 1994; Booij 1998; Bobaljik 2000; Paster 2006, 2009; Bye 2008; Hannahs 2013; Harley 2014; McPherson 2014, 2019; inter alia). In contrast, COPs determine the position of an exponent with respect to some anchor. For example, in Chamorro (Austronesian), the verbalizer appears before the first V in the stem, as an infix (e.g., trăum¡isti `become sad'). This too is expressed via subcategorization in the literature: (2) Segmental COP in Chamorro (Yu 2007, 89, citing Topping 1973, 185) verbalizer -um- : V(« “um must be before a vowel”) Importantly, the conditions on y- in Tzeltal, (1), and -um- in Chamorro, (2), are identical, both subcategorizing for a following vowel. However, as a COIN, the frame [ V ] requires that the exponent y- be inserted only before a vowel (in its default linear position/insertion site, see fn. 3); otherwise it is not inserted. As a COP, the frame [ V ] still requires the exponent -um- to be before a vowel, but the exponent may displace from its default insertion site to satisfy this condition; the insertion of the exponent is not constrained, only its surface position. 2We do not intend for these to be taken as the ultimate representations of subcategorization frames, which is an issue outside of the scope of this squib. 2 COPs have notably been employed to model unexpected constituency disruption—e.g., infixation, second positionhood, and `special clitics' (Zwicky, 1977)—but also can be used to model idiosyncratic prosodic domains and phonological rule blocking (see, e.g., Spring, 1992; Downing, 1998b,a; Chung, 2003; Yu, 2003, 2007; Zec, 2005; Bickel et al., 2007; Caballero, 2010; Hyde and Paramore, 2016; Zec and Filipovi´cĐur¡evi´c,2017; Bennett et al., 2018; Rolle and Hyman, 2019; Rolle and O'Hagan, 2019; Tyler, 2019; inter alia). Numerous other phenomena involving quirks of affix ordering may be able to be grouped here as well, such as types of `local dislocation' in Distributed Morphology (Embick and Noyer, 2001), `morphotactics' (Arregi and Nevins, 2012), `templates' (e.g., the Bantu carp template; Hyman (2003)), and bigrams (Ryan, 2010, 2019).3 3 The puzzle The two distinct uses of exponent-related subcategorization raises an obvious question: given a particular subcategorization frame for a particular exponent, how does the grammar know whether this frame can be satisfied by displacing the exponent (as in Chamorro) or not (as in Tzeltal)? In other words, where in the grammar is it encoded whether the subcategorization frame expresses a COIN or a COP? Most of the literature cited above employs subcategorization for either a COIN or a COP, and so there is little discussion of this ambiguity, nor an explicit solution to the puzzle. One answer is offered by Yu 2007, 229 (also referencing Carstairs-McCarthy 1998), who proposes that “languages may respond to the failure to satisfy a phonological subcategorization requirement in different ways”. One of these ways is infixation (the topic of Yu's 2007 work), where displacement is used as a strategy to satisfy the frame. However, “when morpheme interruption is prohibited” (ibid:229) and infixation therefore unavailable, the exponent must instead satisfy its frame in its default position at the beginning or end of the stem it combines with. If, in this position, the frame is not satisfied, then the exponent is blocked from appearing—thus, there will be a morphological gap, which may or may not be filled by another exponent with a complementary distribution. A simple factorial typology (within OT) can more precisely illustrate this point (as alluded to in Yu 2007, 229). Consider three toy constraints (which stand in for larger families of constraints): (3) a. Subcat = a subcategorization frame must be satisfied 3As should be clear from this discussion, we are not referring to the basic linearization of morphemes here, but rather displacement from an otherwise expected linear position. In this squib we will not be concerned with how basic linearization happens, and assume that it must be established independently from both COINs and COPs (such as being calculated in some algorithmic way based off of the morpho(syntactic) structure à la Kayne 1994). 3 b. Linear = constituents must be uninterrupted (i.e., no infixation/displacement) c. Parse = there must be an available parse (i.e., no ineffability/gap) Now consider an exponent α that subcategorizes for adjacency to an element γ (subcategorization frame [ γ ]), and imagine a linearized input /α-β-γ/, where β and γ are some type of morpho- logical or phonological element (e.g., segments, exponents, features) that form a constituent to the exclusion of α. By re-ranking the constraints in (3), we can derive different outputs, (4). (4) Simple factorial typology (/input/ Ñ zoutputz)[to be argued against] a. If Subcat,Parse " Linear then /α-β-γ/ Ñ zβ<α>γz displacement (frame = COP) b. If Subcat,Linear " Parse then /α-β-γ/ Ñ d (the null parse) gap (frame = COIN) c. If Linear,Parse " Subcat then /α-β-γ/ Ñ zαβγz no change (frame ignored) In such an approach, a single subcategorization frame would behave as a COIN or a COP depending on other properties of the grammar in question, namely, the constraint ranking. According to this approach, then, COPs and COINs are two sides of the same theoretical coin. (Note that in a system like (4c), there would be no synchronic reason to posit a subcategorization frame at all.) Despite the elegance of such a proposal, we explicitly reject it in the next section. 4 Proposal: Deconstructing subcategorization In this section, we argue subcategorization at the exponent level must be formally split into two dis- tinct mechanisms, COINs and COPs. In other words, these two types of idiosyncratic subcategorized structure cannot be encoded in the same type of condition. (5) summarizes our proposal: (5) Subcategorization bifurcation: An exponent may have a Condition on its Insertion and/or a Condition on its Position, which are formally independent of one another, oper- ate over an (overlapping but distinct) set of primitives, and are not interchangeable.