Please add your comments on Wolf’s paper on Maltese stress here.
3 thoughts on “Wolf on Maltese”
Lena
This article made we think (again) about the lexically indexed constraints vs. the PREC constraints, and the reason is the final ranking of PREC constraints which Wolf derives for the V-final stems. He first states that in “na” suffixation they require suffixation before stress assignment, unlike C-final stems, in which these processes are reversed, and derives a ranking of PREC constraints (PREC (Ident(length), build PR-WRD) >> PREC (build PR-WRD, Insert-Obj)), which explain how it is possible. He then goes on and brings up cases, where in V-final stems stress assignment happens before suffixation, and derives two additional PREC constraints that are ranked above the PREC (Ident(length), build PR-WRD), thus cancelling its influence. Therefore, at the end we have several constraints which impose reversed orders of operations ranked above one another, and this analysis made me wonder whether indexed constraints would be a more elegant solution in this case. On the other hand though, perhaps, just going through the one subset of constraints without multiplying them for each suffix is more simple and efficient.
I have a blindingly simple question about chapter 4: what’s the difference between a DEE and an NDEB? I couldn’t see how “happens in a derived environment” and “doesn’t happen in an underived environment” were distinguishable, although it seemed to be a phonological/morphological divide.
With the DEEs fresh in my mind, it surprised me that Wolf posited that all verbal suffixes have a floating mora (sort of a coincidence, no?) instead of saying that the final vowel lengthening upon suffixation is a derived environment effect. Is this ineligible for DEE status for some reason, or does it work better this way? Maybe he wouldn’t be able to collapse final syllable stress attraction and final vowel lengthening using OI for DEEs?
I will reiterate my comment I made at Matt Wolf’s presentation of this work in the Grant Group. I think that taking care of the non-cyclicity of weak-third-radical stems like /mela(;)/ by means of the Prec-constraint Prec(Ident(long), build-PrWd) misses a “substantive” generalization, in the sense that a Prec-constraint is an arbitrary statement, while I think a more interesting and less arbitrary account could be given for the behavior of vowel-final/weak-3rd-radical stems.
Let us assume that vowel-final verb stems have underlying long vowels (e.g. /mela:/), which we could make follow from a morphotactic constraint demanding three morae in each stem (or something of that sort), so that we don’t have to assume /CVCV/ verb stems because of Richness of the Base. (Wolf assumes that weak-3rd-radical stems have underlying short vowels but concedes that the opposite claim can also be made, and has been made in the past.)
If vowel length in vowel-final stems is underlying and is preserved throughout the derivations of forms like /mela:-/-na/ [(i)mlI:-na], then it cannot be harmonically improving to ever stress the syllable (me) – because final and penultimate long vowels always take precedence to be stressed. (me) could only be stressed if the final long vowel were shortened /mela:/ -> [ME.la].
However, what would the motivation for preserving the long vowel be? It’s clearly not preserved in the 3sg perfect /mela:/ -> [mela]. It seems most plausible that the opportunity for the stem-final vowel to be long is somehow provided by the presence of the consonant-initial suffix -na (suppose that there is a good account of why that should be so), which is absent in the 3sg perfect.
Thus, we want to preserve the long vowel, and we can because there is a suffix that allows us to, and preserving this long vowel prevents us from ever stressing (me). This more “phonologically motivated” analysis makes intuitive sense, but unfortunately, it cannot be expressed in OT-CC.
Ideally, I would like the complex of *V:]PWd >> Ident(long) to trump the Prec-constraint demanding late spellout of the object marker -na, but this doesn’t work in OT-CC, as opposed to HS, because only the latter has the requirement of optimality at each step of the derivation. (sorry, no details – too long)
However, this doesn’t seem to pose that much of a threat, at least conceptually: we could get the effect of Prec(build-PrWd, insert-Obj) with something along the lines of morpheme linearization through prosody (cf. recent work by F. Torres-Tamarit). The object marker must hierarchically further from the root than the subject marker: [[mela: -/]-na], therefore we will normally want to linearize the subject marker with the root first by incorporating them into a prosodic word.
However, if there is the need to preserve the underlying long vowel of /mela:/, and this can only be done by premature linearization of -na, then we could presumably force this by ranking the complex of *V:]PrWd >> Ident(long) above the constraint demanding adherence to morphological bracketing (let’s call this constraint Bracketing here):
*V:]PrWd >> Id(long) >> Bracketing
The consonant-final stems will obey bracketing because they have no issues with final long vowels or preserving them, but vowel-final stems will try to obviate their final long vowel issues, and because of this, the step where a prosodic word is created is the moment of decision. Candidates that violate Bracketing by linearizing the object marker first and thus allowing the stem-final long vowel to exist will harmonically bound any candidate that shortens the final vowel because it builds a prosodic word over just the stem and the empty subject marker morph (|me.la|).
|me.LA:.na| +0 > |me.la-0| +na (I show unlinearized morphs preceded by a plus sign, and 0 is a zero/null sign)
Of course, this approach would suggest a typology where affix order can be meddled with on the basis of phonological requirements, and it’s not clear to me that this is attested. I’m aware of this problem, but I think it would be interesting to try to build an analysis of these facts in HS and see if the arbitrary Prec-constraints can indeed be replaced by something more motivated, which would probably also result in a less crude way of handling affix-spellout(/linearization) mismatch.
This article made we think (again) about the lexically indexed constraints vs. the PREC constraints, and the reason is the final ranking of PREC constraints which Wolf derives for the V-final stems. He first states that in “na” suffixation they require suffixation before stress assignment, unlike C-final stems, in which these processes are reversed, and derives a ranking of PREC constraints (PREC (Ident(length), build PR-WRD) >> PREC (build PR-WRD, Insert-Obj)), which explain how it is possible. He then goes on and brings up cases, where in V-final stems stress assignment happens before suffixation, and derives two additional PREC constraints that are ranked above the PREC (Ident(length), build PR-WRD), thus cancelling its influence. Therefore, at the end we have several constraints which impose reversed orders of operations ranked above one another, and this analysis made me wonder whether indexed constraints would be a more elegant solution in this case. On the other hand though, perhaps, just going through the one subset of constraints without multiplying them for each suffix is more simple and efficient.
I have a blindingly simple question about chapter 4: what’s the difference between a DEE and an NDEB? I couldn’t see how “happens in a derived environment” and “doesn’t happen in an underived environment” were distinguishable, although it seemed to be a phonological/morphological divide.
With the DEEs fresh in my mind, it surprised me that Wolf posited that all verbal suffixes have a floating mora (sort of a coincidence, no?) instead of saying that the final vowel lengthening upon suffixation is a derived environment effect. Is this ineligible for DEE status for some reason, or does it work better this way? Maybe he wouldn’t be able to collapse final syllable stress attraction and final vowel lengthening using OI for DEEs?
I will reiterate my comment I made at Matt Wolf’s presentation of this work in the Grant Group. I think that taking care of the non-cyclicity of weak-third-radical stems like /mela(;)/ by means of the Prec-constraint Prec(Ident(long), build-PrWd) misses a “substantive” generalization, in the sense that a Prec-constraint is an arbitrary statement, while I think a more interesting and less arbitrary account could be given for the behavior of vowel-final/weak-3rd-radical stems.
Let us assume that vowel-final verb stems have underlying long vowels (e.g. /mela:/), which we could make follow from a morphotactic constraint demanding three morae in each stem (or something of that sort), so that we don’t have to assume /CVCV/ verb stems because of Richness of the Base. (Wolf assumes that weak-3rd-radical stems have underlying short vowels but concedes that the opposite claim can also be made, and has been made in the past.)
If vowel length in vowel-final stems is underlying and is preserved throughout the derivations of forms like /mela:-/-na/ [(i)mlI:-na], then it cannot be harmonically improving to ever stress the syllable (me) – because final and penultimate long vowels always take precedence to be stressed. (me) could only be stressed if the final long vowel were shortened /mela:/ -> [ME.la].
However, what would the motivation for preserving the long vowel be? It’s clearly not preserved in the 3sg perfect /mela:/ -> [mela]. It seems most plausible that the opportunity for the stem-final vowel to be long is somehow provided by the presence of the consonant-initial suffix -na (suppose that there is a good account of why that should be so), which is absent in the 3sg perfect.
Thus, we want to preserve the long vowel, and we can because there is a suffix that allows us to, and preserving this long vowel prevents us from ever stressing (me). This more “phonologically motivated” analysis makes intuitive sense, but unfortunately, it cannot be expressed in OT-CC.
Ideally, I would like the complex of *V:]PWd >> Ident(long) to trump the Prec-constraint demanding late spellout of the object marker -na, but this doesn’t work in OT-CC, as opposed to HS, because only the latter has the requirement of optimality at each step of the derivation. (sorry, no details – too long)
However, this doesn’t seem to pose that much of a threat, at least conceptually: we could get the effect of Prec(build-PrWd, insert-Obj) with something along the lines of morpheme linearization through prosody (cf. recent work by F. Torres-Tamarit). The object marker must hierarchically further from the root than the subject marker: [[mela: -/]-na], therefore we will normally want to linearize the subject marker with the root first by incorporating them into a prosodic word.
However, if there is the need to preserve the underlying long vowel of /mela:/, and this can only be done by premature linearization of -na, then we could presumably force this by ranking the complex of *V:]PrWd >> Ident(long) above the constraint demanding adherence to morphological bracketing (let’s call this constraint Bracketing here):
*V:]PrWd >> Id(long) >> Bracketing
The consonant-final stems will obey bracketing because they have no issues with final long vowels or preserving them, but vowel-final stems will try to obviate their final long vowel issues, and because of this, the step where a prosodic word is created is the moment of decision. Candidates that violate Bracketing by linearizing the object marker first and thus allowing the stem-final long vowel to exist will harmonically bound any candidate that shortens the final vowel because it builds a prosodic word over just the stem and the empty subject marker morph (|me.la|).
|me.LA:.na| +0 > |me.la-0| +na (I show unlinearized morphs preceded by a plus sign, and 0 is a zero/null sign)
Of course, this approach would suggest a typology where affix order can be meddled with on the basis of phonological requirements, and it’s not clear to me that this is attested. I’m aware of this problem, but I think it would be interesting to try to build an analysis of these facts in HS and see if the arbitrary Prec-constraints can indeed be replaced by something more motivated, which would probably also result in a less crude way of handling affix-spellout(/linearization) mismatch.