Professional Documents
Culture Documents
ABSTRACT:
In this paper, I propose a model for the interface between metrical and
phonological structure, drawing on the recent theory of poetic meter outlined in
Fabb & Halle (2008). Most generative metrical approaches may well be
described as templatic, i.e. understanding meter as an abstract pattern that
constrains linguistic structure. Instead, Fabb & Halle develop a generative
theory, which takes meter to be a structure-building computation. This
computation takes as input a representation of phonology proper, and produces
a meter-specific bracketed grid. Hence, I first emphasize that this proposal
takes meter to be a specialized computational system, rather than an outgrowth
of phonology, contrasting with core assumptions in generative metrics. I
provide evidence thereof, namely an instance of a mismatch between
phonological and metrical structure. Secondly, I apply the theory to two Italian
meters; spelling out the rules building the metrical grids, I also show that the
computation controls only some stresses (stress maxima), and therefore needs
to define them beforehand. Then, I raise the issue of adjacent stresses and take
a close look at the prosodic as well as metrical structure of some lines. The
comparison of the two levels suggests how the meter may choose a single
maximum between two clashing stresses, thus giving the correct scansion. I
conclude that in such cases, this meter has access to a representation of
prosodic structure. Finally, my argument gives rise to some speculations on the
consequences of this model for a literary−linguistic theory and for a theory of
meter in particular.
_____________________
*
This paper has greatly benefited from comments by my supervisor, Marina Nespor. I’m
grateful to Nigel Fabb, Morris Halle, Donna Jo Napoli and Carlos Piera for the giving me the
privilege to discuss the idea I develop here with them. I also thank the audience of the “2nd
ATINER conference on Languages, Linguistics and Literature” for helpful comments, and I’m
particularly grateful to two anonymous reviewers. Any shortcoming in the analysis remains of
course of my responsibility.
2
Most insightful efforts to characterize the cognitive abilities required by
such form have been made in the field of Generative Metrics. Since its outset,
this literary-linguistic tradition (see for instance Halle & Keyser 1971, Nespor
& Vogel 1986, Hayes 1989, Golston & Riad 1994, Hanson & Kiparsky 1996,
and Fabb & Halle 2008 a. o.) has been characterized by the attempts at
modelling the interaction between the abstract pattern of meter and the
phonological component. The various models vary in the way they interpret the
parametrical variation in the different languages, as well as in that they
developed from a rule-based conception of phonology through an evaluation of
the role played by the prosodic hierarchy in a constraint−based approach to
metrical representation, thus mirroring important tendencies in phonology in
the past decades.
In the present paper I take instead the particular approach developed in Fabb
& Halle (2008). They describe meter as a specialized computational system,
which takes as input a representation of phonology proper and gives as output a
bracketed grid.1 Thus, 1) the metrical template is not given and subsequently
matched onto the language string, but it is rather directly generated for every
single line, 2) the bracketed metrical grid in poetry is neither a pre-existing
object encoding the rhythmic representation of the language string, nor a
stylization of linguistic rhythm, but it is an abstract object generated by the
special counting procedure. The Fabb & Halle approach presents at least two
advantages with respects to other theories. First, other theories are not able to
explain some important facts concerning verselines. Fabb & Halle accounts for
a universal feature of verbal art, namely the role played in lines therein. Lines
are non-linguistic objects defined by some kind of control over their form, in
most case over their length. The Fabb & Halle approach, by proposing a way
of counting the elements in the line, provides a formal justification for lines
being of a certain length. Furthermore, other theories have not been able to
account for those forms known as loose meters; as I will show below, Fabb &
Halle provides instead a form-specific scansion for loose meters.
To begin with, I provide some reasons for separating phonological (which
includes the linguistic metrical grid) and the poetic-metrical representation by
drawing up their boundaries. Namely, I distinguish between two
representations of syllables. I then show how a standard meter is accounted for
along these lines, and fully spell out the rules for this computation. I assume
that another advantage of Fabb & Halle (2008) consists in the possibility of
scanning loose meter, that is, lines which have a loose control of syllable
grouping. After showing how the specialized computation works, I suggest that
the theory can be better adapted to the Italian case by further specifying the
kind of phonological representation which is visible to the meter. Namely, the
data I provide show that, in the frequent case of stress clashes, meter must have
access to a representation of prosodic phonological information in order to
1
Bracketed Grid Theory as in Fabb & Halle (2008) directly comes from the proposal
developed at various stages by William Idsardi. Cf. Idsardi (2009) for un updated version of
bracketed grid theory (Simplified Bracketed Grid). Its application to the scansion of poetic
metrical forms differs, however, in that the bracketing domain is the line, instead of the word,
or the phrase.
3
discriminate between metrically relevant stresses (stress maxima, according to
this theory) and non metrical ones. I conclude by showing that this can be
applied to similar problems in Italian strict meters. I then proceed to sum up the
results of this analysis, suggesting that the advantages of Fabb & Halle’s
machinery are fully exploited when one assumes that meter has access to a
representation of prosodic structure. I propose to name this development of the
theory “prosody−extended Fabb & Halle 2008” (henceforth ProExF&H), and I
finally hypothesize that it could account for a number of other varieties in
different metrical traditions as well.
As a start, let us look at two lines from Dante’s Divina Commedia that
instantiate the most widely used meter of the Italian tradition, i.e.
endecasillabo (henceforth, lexical stress is always marked in the examples):
↓
b. fe-dír tor-ne-a-mén-ti e cór-rer giós-tra;
1 2 3 4 5 6 7 8 9 10
The line in (1a) has 13 phonological syllables, but, as stated above, and
parsed in (2a), only 10 metrical positions. Thus, meter “counts” the diphthong
/ea/ (two syllables in the phonological representation) as one single metrical
position (n. 7). Other metrical rules are applied to derive the final metrical
representation. Instead, in (2b), the same meter requires exactly the same
diphthong to count as two metrical positions (n. 4 and n. 5). In sum, they both
instantiate a perfect endecasillabo and notwithstanding they have a different
number of phonological syllables. Thus, the examples display a mismatch
between the phonological and the metrical structure of the line, which I explain
2
Dante, Purg. XVIII: 73.
3
Dante, Inf., XXII: 6.
4
See for instance the account of such meter in Nespor & Vogel (1986).
4
by assuming that the form of a poetic meter is distinct from (though related
to) the linguistic form of its instantiations.
As anticipated above, Fabb & Halle (2008)5 is a theory of how the metrical
form of a line is directly generated by a set of rules and conditions. I now
exemplify the rules for generating the particular meter endecasillabo, scanning
the same line by Dante in (3a-l). The rules follow the line step by step (from i.
to vii.):
iii. GL0: starting just at the right (R) edge, insert iterative R parentheses,
form binary groups, heads R.
iv. Default projection rule: the head of each group projects up to the next
GLs7.
v. GL1: starting just at the left (L) edge, insert iterative L parentheses, form
binary groups, heads L. The last (right-most) group must be incomplete –
unary.
5
Cf. Fabb & Halle (2008): 101-108. I apply here the rules given by Carlos Piera for the
Spanish endecasílabo, so as they are provided in the chapter on Southern Romance meters.
6
Non−projected syllables are marked with Δ for visual convenience.
7
This rule is operative for every meter of the world.
5
vi. GL2: starting just at the L edge, insert iterative L parentheses, form
binary groups, heads L. The last (right-most) group must be incomplete –
unary.
vii. GL3: starting just at the L edge, insert iterative L parentheses, form
binary groups, heads L.
The last asterisk projected up to GL4 is the head of the verse, in that it defines
the particular pattern of this meter. It should be noted that no particular
phonological properties need be associated with a syllable, for it being the head
of the verse.
A strict meter is characterized by having a strict control of some stresses in
the line, which the theory names stress maxima.8 Each meter may have its own
definition of such stresses; for endecasillabo, I suggest to define them as in (4):
The stress maximum definition takes into account only lexical stress and word
boundaries: in other words, it follows from a simplified representation of
stress. Finally, a further condition on this particular meter is stated in (5)
below:
8
This concept is a revision of that used in Halle & Keyser (1971).
6
Another important reason for choosing this approach is that it allows to scan
loose metrical forms, which other approaches fail to describe. E.g., a
constraint-based approach cannot account for such lines, since the placement of
stresses therein is not restricted to given positions.
Loose meters may be defined as a kind of metrical form which has a loose
control of syllable grouping, in that it allows asterisks to be left ungrouped
within GL0; on the contrary, the example in (3) is an instance of a strict meter,
since none of the asterisks within the line is left ungrouped.
Loose meters can be commonly (though not only) found in modernist
contexts, and Italian poetry is no exception in this regard, although the Italian
metrical tradition has not fully acknowledged their presence, yet9. In (6a) I
quote an example of a 12-syllable line by Eugenio Montale10, and parse it in
(6b):
I suggest12 that all lines which are 11 and 12 syllables long instantiate a loose
variant of endecasillabo. The computation building the grid for this meter
differs from the previous one in that, first, stress maxima are defined13 as in
(7):
(7) the syllable bearing word stress except when it is immediately preceded
or followed in the same line by a syllable carrying greater stress (p. 71);
and secondly in that the special rule for loose meters in (8) applies on GL0:
Thus, in a loose meter, it is the placement of stress maxima that influences the
construction of GL0. Due to this rule, the meter of the line is somewhat closer
to its rhythm. Still, the theory maintains a substantial distinction between
rhythm and meter. In (9a-e) it is shown how the example is scanned:
9
Cf. Antonello (1992).
10
All examples from now on (except if otherwise specified) are quoted from Montale (1982.)
11
Cf. Montale (1982): p. 191 (Notizie dall’Amiata, II, 1).
12
I base the scansion on what I claimed in Versace (2008).
13
Fabb & Halle (2008): 71. Their definition for English is consistent with that applied here for
this Italian meter.
7
ii. G0: starting just at R edge, insert iterative L-parentheses, form binary
groups, heads R. Ungrouped syllables are permitted when parenthesis
insertion encounters two non immediately adjacent asterisks.
iii. GL1: starting just at R edge, insert iterative R parentheses, form binary
groups, heads R. Last (leftmost) group may be incomplete.
iv. GL2: starting just at L edge, insert iterative L parentheses, form binary
groups, heads R. Last (rightmost) group may be incomplete.
In (11) I show that the first of the examples above has the same number of
positions as (6):
Therefore, we may proceed to the scansion by applying the same rules as for
loose meters. However, when we are to mark the stress maximum in the line,
we encounter two adjacent word stresses (falling on position 4−5, marked with
14
Cf. Montale (1982): p. 70 (Fine dell’infanzia, 99).
15
Cf. Montale (1982): p. 179 (Barche sulla Marna, 24).
16
Cf. Montale (1982): p.789 (A galla, 33).
17
Cf. Montale (1982): p. 266 (Madrigali privati II, 2).
18
Cf. Montale (1982): p. 459 (Il trionfo della spazzatura, 23).
19
Cf. Montale (1982): p. 58 (Noi non sappiamo…, 24).
9
↑ below). Note that in this case, adjacency is the effect of synalepha applying
on the vowel sequence /−a # e−/, and deriving one single position.
From the mere perspective of the definition in (7), no difference can be made
between the two stresses. However, deciding which of the two stresses should
be counted as a stress maximum is crucial, as two distinct scansions may be
given:
In this example, the choice might seem not to be influencing the scansion, as in
both (13a) and (13b) the output of the computation is the desired 5 feet
structure. But, perhaps more strikingly, there are cases where stress maxima
assignment does make a difference for the computation. As in the two different
scansions in (14a−b):
Or in (15a−b):
(16) i. φ domain
The domain of φ consists of a clitic (C) which contains a lexical head
(X) and all Cs on its non−recursive side up to the C that contains
another head outside of the maximal projection of X.
ii. φ construction
Join into an n−ary branching all Cs included in a string delimited by
the definition of the domain of φ.
iii. φ relative prominence
In languages whose syntactic trees are right branching, the rightmost
node of φ is labelled s (strong). All sister nodes are labelled w
(weak).
The s node is named DTE (designated terminal element, i.e., the main stressed
syllable). Let us now turn back to the line in (10a). Its prosodic structure is
illustrated in (17). Conventionally, φ boundaries are marked with square
brackets and relative prominence within φs is notated with an x (thus, a double
x corresponds to the DTE of φ):
20
It should be noted that there has been some disagreement on the definition of the φ−domain.
Among the other proposals, cf. especially Ghini (1993).
11
(17)
(U) U
(I) I
(ϕ) ϕ ϕ ϕ
(CG) CG CG CG CG CG
(ω) ω ω ω ω ω ω ω ω ω
(F) F F F F F F F F F F
(σ) σ σ σ σ σ σ σ σ σ σ σ σ σ σ σ σ
[La fanciullézza]φ [éra mórta]φ [in un gíro a tóndo]φ
x x x x x
x x x
21
This proposal refers to Italian meters, though I do not exclude the possibility of applying a
similar definition to meters in other languages as well, e.g. english, or greek.
12
(18)
(U) U
(I) I
(ϕ) ϕ ϕ ϕ
(CG) CG CG CG CG CG
(ω) ω ω ω ω ω ω ω ω ω
(F) F F F F F F F F F F F
(σ) σ σ σ σ σ σ σ σ σ σ σ σ σ σ
[pur non mi rasségno]φ [a restár ómbra]φ [o trónco]φ
x x x x
x x x
Here the stress clash takes place within the second φ, but the result is the
same. Namely, the rightmost stress (that in ómbra) is more prominent than the
stress in restár. As it can be seen in (19), the preferred metrical scansion also
picks up the “right” stress, DTE of φ:
x x x
x x x x
(19) [pur non mi rasségno]φ [a restár ómbra]φ [o trónco]φ GL
* (* *(* *) (* *(* *) (* *) 0
However, it should be noted that meter does not refer to prosodic structure in
all its aspects. E.g. it is irrelevant whether the clashing stresses are located at
the juncture of φ, as in (17), or φ−internally, as in (18). Furthermore, syllables
are grouped regardless of the organization in prosodic categories. Thus, i) this
meter only cares for the output of prosodic rules (the final representation of
stress prominence), not for their input; ii) it has access to information on
prosodic structure only in the specific case of a stress clash, namely, when the
prominence of an element is not straightforwardly detectable, but somewhat
opaque. This is precisely when ProExF&H comes into play, that is, detecting
which of the clashing elements gains prominence at phrasal level.
Specifying which representation of phonological structure is the input of the
metrical computation can thus solve the problem discussed so far. It is fair to
acknowledge that a different position could be taken, arguing that when two
13
stresses clash, neither counts as a stress maximum22. I have argued that this is
true for this meter, but I have also shown how Fabb & Halle (2008) allows us
to define this meter as a loose variant of a strict, canonical endecasillabo. The
question is then if the model would solve similar problems in the strict form as
it does for the loose one, and it is in first place motivated by the frequent
incidence of stress clashes in Italian. In this regard, let us notice that Carlos
Piera, author of the chapter on Southern Romance in Fabb & Halle (2008)23,
already had to come to terms with the issue. When discussing the topic of
stress maxima distribution, he states the following conditions on the Spanish
version endecasílabo:
(20) All maxima must project on GL1. At most one of the three syllables
projecting to Gridline 2 may be unstressed.
The presence of a stress on the 4th and/or 6th position defines a traditional
hendecasyllabic verse design, and we notice that it is not only Petrarch that
allows stressed syllable adjacent to those. Indeed, Italian metrical verse admits
such configuration, from the origins up to present days. In (22a−b) two
hendecasyllabic lines are quoted, one by Dante and one by Montale
respectively:
1 2 3 4 5 6 7 8 9 10Δ
(22) a. nel fóco, perché spéran di veníre25
‘in the fire, because they hope to come’
1 2 3 4 5 6 7 8 9 10
b. Trováva lúce, óggi non più che al giórno (LaB, Personae separatae, 24)
‘(he) found the light, today not more than at the day’
22
It is fair to acknowledge that a different position could be taken, arguing that when two
stresses clash, neither counts as a stress maximum. In some the scansion discussed above (11
and 12b), this would yield to the same computation, but the model of interaction would be 1)
weaker, as it could not provide reasons for the move, and 2) less productive, because it would
not be able to provide any account of how lines are produced.
23
Cf. Fabb & Halle (2008): 111−114. Note that maxima, for this meter are defined as «the
syllable carrying lexical stress in a word if it is immediately preceded and followed by
unstressed syllables in the same line.» p. 111)
24
Cf. Fabb & Halle (2008): 116.
25
Cf. Dante, Inf., I: 119.
14
Respectively, position 5−6 and 4−5 are stressed. For the first line, the
computation proposed in Fabb & Halle (2008) yields the following grid:
x 5 x x
(23) [nel fóco,]φ [perché spéran]φ [di veníre]φ GL
)* *)* *) * *)* *)* *) 0 ←
(* * (* * (* 1 →
(* * (* 2 →
(* *( 3 →
* 4
This sheds interesting light on the condition (17b) above, as the asterisk
corresponding to the 5th position is not projected further than GL0, although
corresponding to a maximum. Phonological phrasing (φ−boundaries and
prominence are provided in the example) indicates that 1) the non−projected
maximum corresponds to a weaker node of φ, and 2) all DTEs of the 3 φ are
projected up to GL2. This confirms the general assumption made by
ProExF&H.
If we look further at the second line, we recognize a similar phenomenon:
5
(24) Trováva lúce, óggi non più che al giórno GL
)* *)* *)* *)* *) * *) 0 ←
(* * (* * (* 1 →
(* * (* 2 →
(* *( 3 →
* 4
One of the most fruitful ideas generative metrics brought into metrical (thus
literary) studies is that forms correspond to different levels of representation,
which require separate explanations.26 I have shown here which such levels we
need to call into play if we want to describe some peculiarities of Italian
metrical verse: metrical structure (including iterative and non−iterative rules,
projection etc.), and a “copy” of phonological structure (including syllables,
stress, and bits of prosodic structure). According to the theory followed here,
phonology does not determine (neither does syntax nor semantics) what is
known as “meter”: a line of metrical verse is instead directly generated by a
separate cognitive system, picking up representations from linguistic structure.
Fabb (2009) has proposed that “poetic language” radically differs from
ordinary language, in that it is not ordered in a complex hierarchy of
constituents. It is rather a concatenation of elements, which may include
«words, phrases, and other pieces of language». Accordingly, the examples
discussed show that bits of syntactic structure (phrases) are mapped onto
prosodic hierarchy, and assigned a phonological representation, which includes
relative prominence. This is the “jigsaw” a poet produces when writing a line
of metrical verse. And this is why, in the critical case of clashes, we see meter
choosing amid the inputs those determined by prosodic hierarchy.
SELECT BIBLIOGRAPHY
26
This perspective has been explored by Thoms & Versace (forthcoming) with regard to its
application to the analysis of literatury forms in general.
16
Fabb, N. & M. Halle. (2008). Meter in Poetry: A new theory. Cambridge, UK:
CUP.
Ghini, M. (1993). φ-formation in Italian: A new proposal. In Toronto working
papers in linguistics 12, 41-78. Department of Linguistics, University of
Toronto, Toronto, Ont.
Golston, C. & T. Riad. (1994). Prosodic Metrics. Uc Berkeley and Stockholm
University.
Halle, Morris, e Samuel J. Keyser. (1971). English Stress. Its Form, Its
Growth, and Its Role in Verse. New York: Harper and Row.
Hanson, K. & P. Kiparsky. (1996). A Parametric Theory of Poetic Meter.
Language 72: 287–335.
Hayes, B. (1989). The Prosodic Hierarchy in Meter. In P. Kiparsky & G.
Youmans, Phonetics and Phonology. Rhythm and Meter. New York:
Academic Press. 201-260.
Idsardi, William J. 2009). Calculating Metrical Structure. In E. Raimy & C.
Cairns, Contemporary Views on Architecture and Representation in
Phonology. Cambridge MA: MIT Press. 191-211.
Montale, E. (1982). Le poesie. Milano: Mondadori.
Nespor, M. & I. Vogel. (1986), (20062). Prosodic Phonology. Dordrecht: Foris.
Selkirk, E. O. (1984). Phonology and syntax. the relation between sound and
structure. Cambridge, Ma.: MIT Press.
Thoms, Gary, & S. Versace (forthcoming) How Does the Mind do Literary
Work? Paper delivered at the conference “Bidirectional perspectives in
the Cognitive Science”, Marburg, Germany, 28/02−01/03/2009.
Truckenbrodt, H. (1990.) On the Relation between Syntactic Phrases and
Phonological Phrases. Linguistic Inquiry, Vol. 30, No. 2 (Spring, 1999):
219-255.
Versace, S. (2008). ‘Montale’s Italian Sprung Rhythm?’ Paper presented at the
conference “Frontiers in Comparative Metrics”, November 21−23,
Tartu−Tallin, Estonia.