Courtesy of Tamina
Stephenson. Used with
permission. Phonology I – 24.961
Problem Set 2 – due class 2 of week # 4: Italian length
Given: Three Phonotactic Constraints in Italian, which I have labeled (I)-(III)
(I) If a syllable is stressed then it is heavy.
(II) No final long V
(III) Long C occurs only in the context V_V or V_[+son, +cont]
I will give an analysis roughly following Sommerstein’s system, where rules can be
motivated by phonotactic constraints. I will make one key departure from Sommerstein’s
program, which is that I will not assume that all phonotactic constraints must be surface-
true, or unviolable. (The possibility of adapting Sommerstein’s proposal this way has
been discussed in class.) In particular, I will allow rules to be positively motivated by
Phonotactic Constraint (I) even though there do exist surface forms that violate (I),
notably the citation forms of cittá, caffé, and parló, where the final, stressed vowel is
short. I think this move is justified because even though (I) is not surface-true in all
cases, there seem to be a number of otherwise separate processes (vowel lengthening,
gemination, and resyllabification) that “conspire” to make it surface-true in most cases.
If we were to follow Sommerstein and not allow (I) to be a real phonotactic constraint,
then we would have no account for the connection between these processes.
The Analysis: I will posit five rules, some of which will be positively and/or
negatively motivated by phonotactic constraints. Four of the rules are based on those
given in the handout; the fifth is a rule of de-gemination which accounts for the fact that
when a determiner1 ending in C:V (e.g., bello) is truncated (in this case, to bel) before a
consonant-initial noun, the long consonant is also shortened. The rules are formulated
below, with crucial ordering restrictions noted. If a rule is designated “positively
motivated by Phonotactic Constraint X,” this means that it only applies if the input
violates X (and if the rule will make it satisfy X). If a rule is designated “Blocked by
Phonotactic Constraint Y,” this means that it will NOT apply if the output will violate Y
(even if the input violates a constraint that positively motivates the rule). In other words,
in a sense I am taking the “negative motivation” conditions to be implicitly ordered
before the “positive motivation” condition, even though I haven’t written that way. (I
could have designated such an ordering for each rule, but I never found it necessary to
order a blocking condition after a positive condition.)
Rule 1: Vowel Lengthening
- Positively motivated by Phonotactic Constraint (I)
(ordered before Rules 2-3)
- EXCEPT: Blocked by Phonotactic Constraint (II)
(1) V V:
Note that designating Rule (1) as being positively motivated by Constraint (I) allows the
rule to be formulated more simply than rule (c) on the handout. Specifically, we no
longer need a context term (saying that the rule only applies in a stressed syllable)
because this comes from the phonotactic constraint. Also, since we can designate this
rule as being blocked by Constraint (II), we don’t need to posit any additional “fix-it”
rules to change a final long vowel back to a short vowel, which we might otherwise need
to do. (Alternatively, we could have added more to the context term so that the rule
wouldn’t apply in a final syllable, but this also would have made the rule more complex.)
Rule 2: Gemination
- Positively motivated by Phonotactic Constraint (I)
(ordered after Rule 1 and before Rule 3)
- EXCEPT: Blocked by Phonotactic Constraint (III)
(2) V]σ σ[Ci VCi]σ σ[Ci
Again, this rule is somewhat simpler than its counterpart (g) from the handout. In this
case, we don’t have to specify that the rule only applies at a word boundary, because in
all other cases Rule (1) would have applied first, so that the form would already satisfy
Constraint (I), in which case Rule (2) would no longer apply. We also don’t need to
specify that it only applies to a stressed syllable, since that’s the only context where
Constraint (I) could potentially be violated.
Rule 3: Resyllabification
- Positively motivated by Phonotactic Constraint (I)
(ordered after Rules 1-2)
(3) V]σ σ[Ci … VCi]σ σ[…
This rule can be written in much more general terms than could rule (f) on the handout
(i.e., “Ci…” rather than “s stop” ), since this rule will only apply if Rules 1 and 2 have not
1
I would expect bello to be classified as a modifier, not a determiner, unless there are some important
2
done so. More importantly, by invoking a phonotactic constraint, we can capture the fact
that Rules (1)-(3) are all working towards making stressed syllables heavy, although they
do this in different ways. Then by ordering the rules we can capture the fact that making
a vowel long is the preferred way to effect this heaviness, while re-syllabification is a last
resort. The three rules together guarantee that a stressed syllable will be made heavy one
way or another unless it is open and not only word-final but utterance-final, which is the
correct result.2 It is particularly important to notice here that if we followed Sommerstein
strictly and did not consider (I) a real phonotactic constraint since it isn’t surface-true, we
would lose much of the simplicity of Rules (1)-(3) and would not be able to account for
their “conspiratorial” nature.
Rule 4: Vowel Truncation (in determiners)
(Not phonotactically motivated)
- Ordered before Rules 1-3
(4) V ∅ / _]determiner X]phrase, where X ≠ ∅
This rule is written the same way as in rule (d) on the handout, since it cannot be
simplified by invoking any of the phonotactic constraints (I)-(III). I think it makes sense
in this case not to try to invoke (or invent) a relevant phonotactic constraint, since the
phenomenon of vowel truncation only applies to a limited class of lexical items and
constructions. In particular, I believe that this only occurs with masculine
nouns/determiners, so that we get bel giardino but bella bambina (a well-attested phrase
from movies). It would have to be a mighty specific phonotactic constraint that would
cause a final “o” to be truncated from the masculine form but not a final “a” from the
feminine. Of course, we might want to account for the bel~bello alternation outside of
the phonology per se, perhaps in the syntax or even the lexicon (depending on one’s
conceptions of these), but for the purposes of this problem this issue does not matter
provided that vowel truncation occurs before all the other phonological rules here.
syntactic facts which I’m not aware of; but I’ll call it a determiner for the purposes of this problem.
2
It’s not clear in the data if gemination and resyllabification can only occur within syntactic phrases or if
they can occur anywhere in a single utterance; however, the way the rules were formulated on the handout
would predict that these processes can occur over phrase boundaries, so I will assume that that is correct.
3
Rule 5: De-gemination3
- Positively motivated by Phonotactic Constraint (III)
(ordered after Rule 4)
(5) C: C
If the rule for vowel-truncation above is formulated correctly, then when bello, for
example, is truncated, it is predicted that the underlying form is /bell/ rather than /bel/.
Since the surface form contains a short /l/ (at least before a consonant-initial noun – see
note 3), we need a rule like (5) to shorten the consonant. We can do this by invoking
Phonotactic Constraint (III), which allows us to formulate the rule very simply without a
context term. In a purely rule-based system, we would have to specify that Rule (5) only
applies if a long consonant is followed by a [-son] segment. This would actually be a
fairly simple rule to write, but it would fail to capture the connection between the process
of gemination given in rule (2) and this process of de-gemination. One of these rules is
blocked by a phonotactic constraint and the other is positively motivated by the same
constraint, meaning that the two processes do the opposite thing but have the common
goal of satisfying a general constraint on legal gemination.
In the table below, it is shown how this analysis accounts for the key data given in
part (2) of the handout. The surface form is in bold type.
UR Rule (4) Rule (1) Rule (2) Rule (3) Rule (5)
(4)»(1), (5) (1)»(2)»(3) (1)»(2)»(3) (1)»(2)»(3) (4)»(5)
/a.'mi.co/ a.'mi:.co
n/a n/a n/a n/a
/re.'ga.lo/ re.'ga:.lo
/fra.'tel.lo/
n/a n/a n/a n/a n/a
/'for.no/
/εε.#.'bel.lo/ n/a n/a n/a n/a n/a
/'bel.lo.# 'bel.l# 'bel.l# 'bel.#
n/a n/a
._ar.'di.no/ _ar.'di.no _ar.'di:.no _ar.'di:.no
/nes.'su.no.# nes.'sun.#
n/a n/a n/a n/a
.gat.to/ .gat.to
/_it.'ta/ n/a Blocked n/a (since no n/a (since no n/a
/par.'lo/ C following) C following)
3
The way this rule is formulated predicts that when bel comes before a vowel-initial noun, the “l” will be
long (the second half of the geminate forming the onset to the next syllable). The other option would have
been to somehow build the shortening of the consonant into the truncation rule. This would predict that
when bel came before a vowel-initial word, the l would be short. (If there is a process that allows a final
consonant from one word to become the onset of the next, and if the rest of my analysis is right, then this
would also predict that the /ε/ in bel would be lengthened.) Although I don’t know the relevant facts about
Italian, I suspect that the first prediction may be correct. As Sonny Webb pointed out to me, there is some
evidence for this from the orthography in forms such as bell’anno.
4
UR Rule (4) Rule (1) Rule (2) Rule (3) Rule (5)
(4)»(1), (5) (1)»(2)»(3) (1)»(2)»(3) (1)»(2)»(3) (4)»(5)
/_it.'ta.# [blocked on 'ta] _it.'tap.#
.pu.'li.ta/ [ 'li 'li: ] .pu.'li:.ta
n/a _it.'ta.# n/a n/a
.pu.'li:.ta
/_it.'ta.# Blocked Blocked _it.'tas.#
n/a n/a
.'spor.ka/ .'por.ka
Discussion: I have already noted how each rule is affected specifically by invoking
phonotactic constraints. In general, this allowed the rules to be written much more
simply, often without a context term since the relevant constraint completely determined
the context for rule application. In the case of rules (1)-(3), for example, this arguably
simplifies the overall grammar since a single phonotactic constraint takes the place of
three different context terms in three separate rules. Moreover, invoking constraints
allowed us to naturally capture the “conspiracies” apparent in the Italian alternations, e.g.,
the different strategies for making a stressed syllable heavy that show up in rules (1)-(3).
This feature of an analysis involving constraints was also particularly striking in rules (2)
and (5), which actually do more or less the opposite thing from each other (one creates a
geminate and the other gets rid of it), but seem to both be conditioned by the same
phonotactic constraint. Allowing a constraint to either positively or negatively motivate a
rule gives us a natural way to express the relationship between such “opposite” rules as
(2) and (5). In some cases, such as Rule (1), allowing a constraint to block a rule let us
avoid putting in extra “fix-up” rules or complicated context terms.
The moral of the story, then, as I see it, is that we can “buy” ourselves a lot by
invoking phonotactic constraints. I have done this conservatively, without claiming that
these constraints must be universal or getting rid of rules altogether, which seemed to
work well for this problem. I haven’t worked out a solution in OT, so it remains to be
seen whether a more radical departure from a rule-based system would do any better. I
actually suspect that OT wouldn’t do any better than a Sommerstein-type account for this
particular problem, since my solution got so much out of these particular constraints
without claiming universality or having to introduce a whole new class of carefully-
ranked faithfulness constraints. But there are, of course, many other considerations
which could potentially point to a completely constraint-based analysis.
