2.3 English Stress and Derivational Morphology
2.3.2 Possible approaches in a piece-based framework
2.3.2.1 A Morphosyntactic Hypothesis: Inner- vs. Outer-Attachment
In Distributed Morphology, we assume that the syntax produces category-defining heads and that (at least some) of these heads cause phonological spell-out (Marantz 2001, 2007, Embick 2014, see also Marvin 2002; Embick and Marantz 2008; Embick 2010a). Following Embick’s (2010a) proposal, we maintain that merging category-defining heads causes spell-out of their complements. This proposal creates a distinction between affixes in the inner domain, e.g., root-attached affixes, and affixes in the outer domain, e.g., non-root-attached affixes. This distinction is necessary because, given the theoretical framework of spell-out, only a root-attached category-defining affix would be accessible during spell-out of the root. This distinction between inner and outer domains has been shown to play an important role in allomorphy and allosemy.
2.3.2.1.1 Inner- and outer-attachment matter for Allomorphy and Allosemy
In the inner domain, the selection of allomorphs of heads is potentially idiosyncratic to the root, but in the outer domain, it is not (Embick 2010a). For example, root-attached nominal heads show a wide variety of forms, while outer attached nominal heads do not, as shown in (36).
The data in (36a) shows a variety of allomorphs for the nominal in the root-derived and simple cases, but only one allomorph in the gerund cases. This is predicted by Embick (2010a) because the root-derived and simple cases are analyzed as being root attached n while the gerund cases have an interveningvhead, as shown in (36b). The vocabulary items (36c) of the innernheads are conditioned on a variety of lists to which the various roots belong, while the outern head has no conditioning (or perhaps is conditioned by being next to av).
(36) Allomorphy of nominals in inner and outer domain (Embick 2010a, pp.44–46) a. Nominals and allomorphy
Derived/Simple Gerund
refus-al refus-ing
marri-age marry-ing
destruct-ion destroy-ing
break-; break-ing
b. Structure for Root-attachednvs. non-Root-attachedn
n n p ROOT Root-attachedn n n v v p ROOT non-Root-attachedn
c. Vocabulary Items for Inner vs. Outer Domain
Inner Domain Outer Domain
n↔-al/LIST1 n↔-ing
n↔-age/LIST2
n↔-tion/LIST3 ..
.
The same inner- and outer-domain distinction is used in some subpart of the meaning system, namely those that relate to polysemy resolution (see, for example, Marantz 2013). That is, if a root- attached head restricts the meaning of a root to a certain subset of possible meanings, meanings not included in that subset cannot be used by outer heads.
(37) Inner- and outer-domain restrictions on meaning choice ofpFIRE:
form structure possible meanings ofpFIRE
p
FIRE {"a chemical reaction of burning fuel", "excited, passionate", . . . }
fiery pFIRE+a {"a chemical reaction of burning fuel", "excited, passionate"}
fire pFIRE+n {"a chemical reaction of burning fuel"}
For example, in (37), let us assume that the rootpFIREhas several possible meanings including "a chemical reaction of burning fuel" and "excited, passionate". The head attached in the inner domain of the root may select for a subset of these meanings. The adjective head (-y), for example, selects for the subset which contains both those meanings while the null nominal head selects for the subset which only contains the meaning "a chemical reaction of burning fuel". Outer-domain heads cannot access meanings of the root that were excluded from the subset chosen by the inner head, so if an adjective head (-like) is attached to the pFIRE+nstructure, it only has access to the meaning "a chemical reaction of burning fuel" (as determined by the null nominal head) and not to the other possible meanings ofpFIRE. That is,fire-likecannot access the root meaning "excited, passionate", and thus cannot have a meaning based upon that.
2.3.2.1.2 Inner- and outer-attachment do not matter for phonology
Given that the inner- and outer-attachment of heads is important in allomorphy and allosemy, a first hypothesis about the different phonological nature of affixes could be that inner domain affixes show stem-level phonology, while outer domain affixes would be word-level. If this correspondence were true, it would explain stem level affixes attaching to "stems" (here called "roots") while word- level affixes attach to words, since word-level affixes would attach outside a [root + functional head] structure. This hypothesis, however, is not true. There are four pieces of evidence for the falseness of this hypothesis: Class 2 affixes attaching to roots, level-ordering paradoxes, multiple Class 1 affixes, and morphosyntactically identical affixes showing different behavior depending on what exponent is inserted.
Examples of Class 2 suffixes attaching to roots were shown in (33) and examples of Class 2 suffixes attaching outside Class 1 suffixes were shown in (34). In the framework under discussion, we expect exactly one Class 1 affix to attach to a root so having none (in the case of Class 2 on the root)21or a Class 1 affix attached elsewhere (as in the level-ordering paradoxes) is a problem.
A further piece of evidence is that words may have multiple Class 1 suffixes, as shown in (38):
21The examples of Class 2 attaching to roots could be construed as having a null head in between the root and the overt affix, for example,ruthlessaspRUTH-n;-aless. As such, this particular argument on its own is not terribly strong. The level ordering paradoxes and the instances of multiple Class 1 affixes are a stronger argument against inner/outer attachment for phonology.
(38) Multiple Class 1 suffixes:
atom-ic1-ity1, educat-ion1-al1, monstr-os1-ity1, rotat-ion1-al1-ity1
If only the root-attached head were able to have Class 1 phonology, there should not be the possibility of having multiple Class 1 affixes because only one syntactic head can be immediately adjacent to a root.
Finally, there are cases where the same morphosyntactic structure resulting in a difference in phonological behavior dependent on the affix inserted. For example, the data in (39) show that the same head (n) in the same environment (attaching to [ root + a ] ) with approximately the same semantics (forming an abstract noun), results in a different phonological output depending on the exponent chosen.
(39) Same morphosyntactic structure shows stress shift inatomícitybut not inatómicness n n -ity1 a a -ic1 p ATOM atom atomícity n n -ness2 a a -ic1 p ATOM atom atómicness
There is no morphosyntactic reason to believe that the structure ofatomicityandatomicnessis different22; Both words seem to be a noun derived from the same adjective head and root. However, the stress shift inatomicityshows that-ityis a Class 1 suffix while-nessis Class 2 (presuming the stress shift is a stem-level process, following Chomsky and Halle 1968; Halle and Vergnaud 1987; Halle 1998, etc.).
It is clear that every instance of a category-defining head does not cause a run of the phonological block, suggesting that the morphosyntactic spell-out process is not equivalent to a pass of the phonology (that is to say, syntactically phase-cyclic is not the same as phonologically cyclic, see Embick 2014). However, it is also not the case that the morphosyntactic structure (inner- vs. outer- attachment) can determine whether a pass of stem-level phonology is run.
22If these twonheads (-ityand-ness) are exactly identical, this raises questions for how blocking works in this theory. We might expect only one exponent to be always inserted resulting in the other being ungrammatical. One solution would be to assume that thesen-heads are slightly different in some way (perhaps some morpho-semantic features, see, e.g. Plag 2003, pp. 66-67), but that they are the same morphosyntactically.
If the information about whether to run the stem-level phonological block is not an automatic result of morphosyntactic spell-out or structure, then it must instead be encoded somewhere else. I will end up positing a diacritic solution in Section 2.3.2.3 below, however, we do not want want to give up on the idea of a difference in morphosyntactic structure lightly.
2.3.2.2 On the attachment and structure of Class 1 and Class 2
Ideally, all phonological structures would reflect syntactic structures and different phonological behaviors would be the result of underlying syntactic differences. As we saw above, for Class 1 and Class 2 affixes, this does not initially appear to be the case. The difference in phonological behavior does not immediately seem to reflect any syntactic difference. However, syntactic differences have been posited for these classes.
Lowenstamm (2010), for example, posits that the derivational affixes seen in English are not exponents of category defining heads but are roots of their own. Conceptually, his argument is that (at least some) derivational exponents appear to surface as more than one part of speech. For example, words suffixed with -icare usually adjectives, but not always (e.g., tunic). Treating derivational affixes as roots presents an interesting solution to this. Lowenstamm posits that these derivational affix-roots have uninterpretable features which force them to syntactically select for either other roots or for full phrases. In the situation where an affix-root select for other roots, the result is that all of the derivational morphology is underneath the category defining head, meaning it is in the same syntactic spell-out cycle. This behavior, then, derives Class 1 affixation, allowing phonological interaction between the exponents because they are in the same cycle. Affix-roots which select for full phrases are in a separate spell-out cycle than the phrase they select and therefore do not interact phonologically. That is, they are Class 2 affixes.
For example, in (40a), both the affix-rootsicandityhave the uninterpretable feature [up] which indicates that they select for root phrases only. The only category defining head (here,n) is merged above the three roots. This triggers spells out of the root and the Class 1 affixes in the same cycle. By contrast, in (40b),nesshas the uninterpretable feature [u xP] which indicates that it selects for a phrase. Because of this, it cannot attach directly to the root complex. Instead, it can attach to the
phrase that is a category defining head plus the root complex. The result is that the lower roots (atom
andic) are in one spell-out phase, butnessis in a different phase. (40) Structures as proposed by Lowenstamm (2010)
a. Class 1 Affixation nP pP pP p ATOM p IC[up] p ITY[up] n b. Class 2 Affixation nP pP aP pP p ATOM p IC[up] a p NESS[u xP] n
⇒
nP pP aP pP p ATOM p IC[up] a tp n n p NESSWhile this idea does provide an account of the interactions in the phonology with respect to the syntactic phases, it seems strange for other reasons. First, affixes do not behave like roots elsewhere in the system. We think of roots as open class items which carry some real-world semantic reference, whereas affixes only modify the existing structure and meaning. Furthermore, in Lowenstamm’s system there are never overt exponents of category defining heads, which seems strange. Rather, all overt exponents (at least within the derivational morphology domain) are considered roots and all category defining heads are phonologically null. There is no a priori reason why the system should work this way, although it could. However, without derivation affixes carrying the semantics of category defining heads, we lose the fact that there is some compositionality of meaning with multiple affixes. For example, atomicity has some relation to the adjective meaning of atomic
(although there are exceptions to this compositionality). Similarly, it would be difficult to explain the semantics of doublets such ascómparable“approximately equal” andcompárable“able to be compared” without reference to the component parts; in this case, the fact thatcompárablecontains the verb form ofcompare(this case is discussed more below).
Most problematic for Lowenstamm’s account, however, is that fact that this system incorrectly predicts the phonological behavior of Class 2 affixes. For example,. Lowenstamm proposes the structure in (41) formoneyless, because-lessis a Class 2 suffix:
(41) Lowenstamm’s (2010) structure formoneyless
aP pP nP pP p MONEY n p LESS[u xP] a
⇒
aP pP nP pP p MONEY n tp a a p LESSThis structure predicts two separate domains for phonological behavior, themoneydomain and the-lessdomain. While this is correct in that the addition of-lessdoes not change the stress pattern of money, it is problematic in that empirically there is no stress assigned to -less. The structure proposed by Lowenstamm looks very much like compounding rather than affixation (cf. Harley 2009, although the head movement is different in Harley’s analysis). With Lowenstamm’s structure, we expect secondary stress on-less, like we would in a compound. This, however, is impossible.
(42) No secondary stress on-less
a. móneyless["m2.ni.l1s], *["m2.ni.lEs]
b. pénnylessvs.pénny lèss
i. “he is pénnyless”
["pE.ni.l1s], *["pE.ni.lEs]
ii. “just a pénny lèss than him”
["pE.ni.lEs], *["pE.ni.l1s]
As shown in (42), it is impossible to put stress on the-lessofmoneylessas diagnosed by vowel reduction. This is made apparent in minimal pairs, such as in (42b), in whichless as a separate word is given secondary stress and surfaces as[E]instead of reduced[1].23Note that this fact is still true with two Class 2 suffixes, e.g.,móneyl[1]ssn[1]sswhich cannot take secondary stress on-less
(*móneyl[E]ssness).
23Here I use[1]to represent a high lax unstressed vowel for speakers with a distinction between high and mid lax vowels (i.e.Rosa’s[ôoz@z]vs.roses[ôoz1z]).
Thus, while Lowenstamm’s proposal is a valiant attempt to make the syntax match the phonol- ogy, it drastically modifies the morphosyntax in a way which damages previous theoretical findings about the structures and relationships therein, and additionally fails to capture all of the facts about the phonological output.24
There have been other proposals to account for the Class 1/Class 2 behavior as well, mostly relying on some sort of prespecification of Class 2 as a certain domain in the phonology. Some accounts propose that Class 2 suffixes contain their own cycle or stem-level domain, resulting in their separation from the phonology of their stems (Halle and Mohanan 1985; Mohanan 1986; Baker 2005; Buckler and Bermúdez-Otero 2012; Bermúdez-Otero to appear). These proposals have the same problem as Lowenstamm’s account: They predict cyclic or stem-level stress on the Class 2 suffixes, which does not occur.
Another possibility is that Class 1 and Class 2 do not differ in their syntactic features but instead have some sort of difference in attachment to the syntactic structure. For example, building upon idea of autosegmental planes from Autosegmental Phonology (Goldsmith 1976, 1990; McCarthy 1986), Halle and Vergnaud (1987) suggest that Class 1 and Class 2 morphemes are have their own planes. Halle and Vergnaud do not intend this to be differences in syntactic attachment, but we could posit a reification of these “different planes” into some meaningful difference in the way morphemes are merged into the syntactic structure. For example, in (43), Class 1 attachment is shown with straight lines and Class 2 attachment is shown with a coiled line. These line types are meant to distinguish two ways of attaching to the morphosyntactic structure.
(43) Different attachment types between-ityand-ness n n -ity a a -ic p ATOM atom n n -ness a a -ic p ATOM atom
This is certainly a possibility, but it must be considered that this attachment difference seems to
24Lowenstamm’s proposal also predicts domains for allomorphy that are too large because he posits that several affixes (which mainstream DM would call category defining affixes) may within the same cyclic spell-out domain and thus able to be allomorphically sensitive to each other.
only affect the phonological outcome. There does not seem to be any syntactic difference between these two types of attachment. This solution is essentially the same as the vocabulary item diacritics suggested below, but it seems strange to posit that the diacritic status is encoded in the syntax because it has no effect on the syntax. Rather, the effect is seen in the phonology, thus I posit the vocabulary item diacritics below, because that is where the phonological forms are introduced.
2.3.2.3 A Morphophonological Hypothesis: Vocabulary Item Diacritics
As already proposed for Slavic yers in Section 2.2, the analysis which will be pursued here follows Halle and Vergnaud (1987) by proposing that there are diacritics on vocabulary items which deter- mine whether a pass of the stem-level phonology is to be run (see also Marvin 2002 for a similar approach).
For Halle and Vergnaud, the composition of all morphemes takes place before all phonology (as it does in the syntax all-the-way-down model of DM). Affixes are marked diacritically for whether or not they trigger cyclic phonology (equivalent here to what I have been calling stem-level). This rejects the architectural levels or strata of Lexical Phonology (and also Stratal OT).
The difference between Class 1 and Class 2 affixes is proposed to be a difference in the instruc- tions to the phonology, notated as+ in (44) below. Vocabulary Items with the+ diacritic trigger a pass of cyclic or stem-level phonology, while those without the diacritic do not.25
(44) English Sample "Class 1"/"cyclic"/"stem-level" vs. "Class 2"/"noncyclic"/"word-level" Vo- cabulary Items
a. n↔/-ity/+ b. n↔/-ness/
2.3.2.3.1 Addressing Selkirk’s (1982) objections to diacritic-based analysis
Selkirk (1982, p.112–119) presents a conceptually similar diacritic-based analysis as an alternative to her context-free grammar analysis, but rejects it for several reasons. The basic scheme of her
25It could very well be the case that the diacritic is on the non-cyclic affixes rather than the cyclic ones. See discussion in Section 2.5.2.
diacritic analysis given in (45):
(45) Selkirk’s (1982) diacritic-based analysis (Selkirk 1982, p. 113)
Category Subcategorization Frame
Class 1 [Af;α, +L] [_β[+L]] or [β[+L]_]
Class 2 [Af;α,−L] [_β[±L]] or [β[±L]_]
whereαandβare syntactic category features (noun, verb, etc.)
The idea is that affixes have a diacritic ([±L] for lexical) which is used in the subcategorization of other affixes to control what affixes can be attached. Class 1 affixes are [+L] and they can attach only to other [+L] affixes, while Class 2 affixes are [−L] but are subcategorized to attach after either [+L] or [−L] affixes, allowing them to attach after Class 1 or Class 2 affixes. Selkirk also allows for affixes that are members of both Class 1 and Class 2 (such as, in her analysis,-able) by having a undefined [uL] feature
One important difference between Selkirk’s analysis and the analysis proposed here is that Selkirk’s subcategorization corresponds to two distinct operations in Distributed Morphology, the combinatoric system (syntactic selection) and allomorphy (46). The combinatoric system is a func- tion of the syntax and determines whether or not a headycan be merged with a headx. Allomorphy, on the other hand, is a function of competing vocabulary items during Vocabulary Insertion and determines what phonological exponent will be inserted for a given morpheme in a given context.
(46) Schematic of Combinatorics and Allomorphy in DM (Embick 2010a)
Combinatorics Allomorphy D y x * y z x↔/ϕ1/ / Context1 x↔/ϕ2/ / Context2 x↔/ϕ3/
Selkirk makes several important objections to her proposed diacritic-based analysis. Each will be addressed in the paragraphs that follow. Importantly The first two objections can be eliminated