1 Overview

Chapter 1 outlines the theory in terms of what is likely most familiar to readers, i.e. English metres. The metrical grid builds from a basic projection from syllables (Gridline 0). Parameter settings determine how the projected positions (represented as asterisks) should be grouped (binary/ternary), from which edge the iteration of groups should run (R/L), which type of parenthesis is inserted (‘(‘ or ‘)’) and at which edge of each group the head occurs (R/L). Heads constitute asterisks at the next level (Gridline 1), which is subject to the same set of parameters. Each line of verse is parsed into a multi-layered, specific grid, with variably high columns, the highest of which identifies the position called ‘the head of the verse’. Asterisks that are not grouped by parenthesis insertion, remain ungrouped. Here is the analysis of a line from Edmund Spenser's Faerie Queene (p. 58):

1. (1)

   

The arrow at each gridline indicates the parsing direction, and the bracketing indicates the size and headedness of the groups.

The model makes a categorical distinction between strict and loose metre, based on the derivational order of iterative and non-iterative bracketing rules. Chapter 2 covers the English strict metres: iambic and trochaic, anapestic and dactylic, pairwise essentially each other's opposites. Iambic lines are generated by iterating final-headed binary grouping from the right, trochaic lines by iterating initial-headed binary grouping from the left. Anapestic and dactylic are generated in parallel but with ternary grouping. English metre controls ‘maxima’, a notion of prominence used in several verse systems, defined as in (2) (p. 47).

1. (2)

   

Once the grid is built, the maxima which occur are required to be heads of the groups given by iteration at Gridline 0. The higher structure of the grid, specifically the unique ‘head of the verse’, is needed for the model to be able to refer to a location in the middle of the line (e.g. for caesura or deletion). In the example above, the caesura occurs after the sixth syllable, now statable as ‘right after the head of the verse’. Information like this guides the various aspects of grid construction. There is no predetermined template for metres; rather, the approach is that one does what one can with the tools available to get the right result, hence the change of direction, headedness and group size between Gridlines 1 and 2 above. This entails a rather different methodological stance in comparison with other generative theories, which work from a fixed, external schema and then deduce the metrical properties in the matching with the text.

Chapter 3 treats English loose metres, where looseness contrasts with strictness by a more varied syllable count across lines, a more relaxed definition of maximum and the admission of ungrouped asterisks as well as incomplete groups in mid-line. In a loose metre, maxima are appointed as heads in the grid before iterative grouping takes place. The authors show that while the joint effect of non-iterative marking of more inclusive maxima will largely determine loose metre, the iterative rules do have a role to play, specifically in long stretches of non-prominent positions.

Chapter 4 is contributed by Carlos Piera, and deals with Spanish metre and some aspects of Italian and Galician-Portuguese metre. Like other Romance poetry, Southern Romance very frequently doesn't count a syllable that occurs in hiatus. In this model, this is called ‘non-projection’, common in Romance, less so in English, ubiquitous in Old English. Non-projection entails no adjustment of recitation, but is a purely metrical licence, applying when needed. What comes out as the truly characteristic feature of the Southern Romance metres is the fact that they use two notions of prominence: stress and (the narrower) maximum. Both of these may be jointly controlled within a line, e.g. in Petrarch, such that whenever there is a maximum in the third position in the endecasillabo, there will also be a stress in the sixth. When the third position contains something else, the sixth is also freer.

French metre is analysed in Chapter 5. Fabb & Halle explicate the importance of appealing to the underlying structure, specifically as regards the unstable vowel e caduc (schwa), whose pronunciation (or not) is immaterial to the metrical rules. French metres are divided into simple and compound (roughly, whether or not they are long enough to have caesura). Simple metre is illustrated by Verlaine, and compound metre by the alexandrin (iambic hexameter) of Racine and others. This chapter also contains a discussion of the metre of the Old French Chanson de Roland, which is analysed as a special case of a simple metre with two separate grids per verse line, rather than as a decasyllabic compound metre with caesura.

All the major metres of Classical Greek are treated in Chapter 6 (iambic, dactylic, anapestic) along with a subset of the so-called Aeolic lengths. Greek is the first quantitative system encountered in this book. It turns out that this has little consequence for grid construction, since the unit counted remains the syllable, rather than the mora. The proposed unifying feature of the Greek system is a principle which deletes the first of any sequence of two light syllables. This makes the bimoraic equivalence of H and LL (traditionally called ‘resolution’) go away, and instead emphasises what is shared between H and L, namely the fact that they are single syllables. This is clearly at odds with traditional analysis and also with phonological fact. It does, however, express a similarity between Greek and the Romance systems, where syllables are frequently discounted. The fairly novel approach to Greek metrics is matched by a fairly traditional definition of what an anapest is: LLH or HH. This is a simplification that creates more symmetry than is empirically warranted in the Greek metrical system. As West (Reference West1982: 94f, 191) points out, the anapest can be any of HH, LLH, HLL or LLLL.

Chapter 7 treats the many metres of Classical Arabic, where again the basis for counting is syllables. The main unifying property of Arabic metre in this model is that a syllable projecting to Gridline 2 must be heavy, and preceded by a light syllable if that syllable belongs to the same group at Gridline 0. This captures the traditional watid (‘peg’), typically an iamb. Just as in most earlier generative work (Halle Reference Halle1966, Maling Reference Maling1973, Prince Reference Prince, Kiparsky and Youmans1989), the basic classification of metres in circles is preserved. Circle 3 (rajaz, hazaj, ramal) is taken to be strict and basic, the four other circles being related to it. Circles 1, 2 and 5 contain only loose metres, while circle 4 contains strict metres with the peg reversed.

Chapter 8 deals with Sanskrit metres, whose major properties come out as rather unexpected within the Fabb & Halle model. The head of the verse is necessarily a light syllable, though heavy syllables tend to project elsewhere. The head should be preceded by a caesura, a fact that sometimes falls out automatically, and sometimes requires asterisk deletion. Deletion rules are quite potent in their effect on the grid structure, but are crucially restricted to the syllable in the head of the verse position. The authors look at Vedic metres and the Classical Sanskrit anuṣṭubh metre, before analysing some of the fixed syllabic metres of Classical Sanskrit. The chapter also contains the analysis of a strictly mora-counting metre, āryā, the only metre of this type included in the book. One might presume that Japanese metres (haiku, tanka) would also be treated like āryā, heavy syllables projecting two grid positions and light ones one.

Chapter 9 is on Latvian metre. There are the simple-looking dimeters occurring in dainas (folk songs), which come with the interesting requirement of a polysyllabic word at the end of each line. Further, loose anapestic dimeter and loose trochaic tetrameter are identified, the latter of which admits a variable number of groups per line, unlike corresponding loose metres in English, which abide by a strict number of feet per verse line.

Chapter 10, Meters of the world, serves the purpose of broadening the outlook. Earlier chapters have illustrated the partitionings of H vs. L for quantitative metres, and various versions of stressed vs. not stressed (maxima). This chapter introduces, for example, a metre based on the partitioning of superheavy vs. other (non-classical Arabic), metres employing tone and a metre distinguishing alliterating vs. non-alliterating syllables. Chinese and Vietnamese both exploit the tonal distinction for metrical purposes. Chinese lu shi distinguishes between level and oblique tones. The system is apparently quite complex, and I found the presentation rather too compact. The Vietnamese ca dao is a clearer case, where even-numbered syllables are controlled for tone, a fact that Fabb & Halle interpret as a signal of headedness. There are two tone classes for the purposes of metre, Even and Sharp, distributed over the four tonally controlled positions of the line in the following manner (unfortunately misprinted in the wrong order in the text): x Even x Sharp x Even x Even (where ‘x’ marks tonally uncontrolled positions). The fourth position (Sharp) is the designated head of the verse. The authors also look at metrical rhyme in ca dao. The logic of the model here requires a different grid, since statements regarding positions internal to the line can only be made via heads.

The last metre treated in this chapter is that of Old English Beowulf. In this analysis, Beowulf is a very loose metre, as it admits so many unstressed syllables in sequence that the model cannot build a proper metrical grid. The proposed theory-driven solution to this dilemma is that unstressed syllables simply do not count in Beowulf. There is a rich tradition of ignoring unstressed syllables in the analysis of Beowulf (e.g. Sievers Reference Sievers1893), but none as radical as this one. Alliterating syllables become heads by a non-iterative rule, and iteration then builds groups from what is left, i.e. stressed syllables. However, the model still overgenerates ungrouped syllables, a fact which calls for further pruning.

Chapter 11 is on the metrical poetry of the Old Testament. If the treatment of Beowulf is brief and rather hurried, the treatment of the Old Testament metre is more detailed. Here the authors engage in a discussion of previous proposals and philological interpretation not seen in any of the preceding chapters. The point of this is to establish the true line length, which is important to get the result for Psalm 137 (a picture poem). The chapter is basically unattached to the rest of the book, as the counting in this metre is not based on grid construction, but on rudimentary, ordinary counting. The definition of ‘metrical’ (designated line length) given in this book includes Old Testament poetry, but the theory as such doesn't.

2 Templates and rhythm

If the job of metrics is to count the syllables of metrical lines, group them and via that grouping make statements about some patterns of the line, then this theory is as good as any. The length of lines is central, specifically, the fact that metrical lines have length. That is what motivates the grid. The actual length of lines in a metre, however, is not an issue here. Like most other models, the actual length is treated as pretheoretic, and hence stipulated.Footnote ¹ Grids are assigned instead by trial and error. In principle, the entire repertoire of rules for grid construction is available at the metrification of each individual verse line and this means that there are commonly several viable analyses for a given line. Variation is kept in check by a principle of isometricality, whereby poems are assumed to be written in one and the same metre unless there is evidence to the contrary (as e.g. in Hopkins' sprung rhythm), and by a general condition on licensing of a metre within a tradition, i.e. cultural anchoring. These notions are also both pretheoretic.

A new and interesting perspective here is that the analysis is looking at lines rather than at metres, thereby shifting the focus somewhat from that of traditional and generative metrics, where the metre is not the result of analyses of individual lines, but is, as it were, a separate statement. There is then no place for templates in the Fabb & Halle model, the typical way of conceiving of metre in traditional as well as generative approaches. This is more than a technicality. In Fabb & Halle's view, the theory of metre is the parametric algorithm, the set of rules that assign a grid to a given line of metrical poetry. With the grid in place, some patterns can be organically expressed in relation to the head of the verse. Other patterns rather take the form of stipulations, as acknowledged by the authors, much as in traditional metrics. Even if they may be expressed in terms of the grid, they often remain extratheoretic.

Beside dispensing with templates, Fabb & Halle explicitly wish to avoid saying that the grid is inherently rhythmic. Rhythm arises as a side-effect of conditions on the grid, specifically via the notion of maximum and the necessity of maxima to project in the grid. In most definitions, however, a maximum is not just a syllable with a certain prominence property, but a syllable with a rhythmic configuration around it. A stressed syllable in a polysyllabic word (‘strong’ in previous models) means that we are dealing with minimally two syllables, one of which is more prominently stressed than the other. Furthermore, in the more elaborated definitions of maximum (there are at least four), the contextual expression is more direct. There is a certain sleight of hand here, inherited from earlier generative accounts, whereby the statement of conditions rhetorically address a single unit, but where the real content of the constraint concerns two or more syllables in a particular rhythmic configuration. Nevertheless, this is at least superficially closer to a relevant representation of rhythm in English metre than what is offered by the classical generative metrics model (e.g. Halle & Keyser Reference Halle and Keyser1971, Kiparsky Reference Kiparsky1979, Hanson Reference Hanson, Dresher and Friedberg2006), which assumes a metrical template, labelled with strong (s) and weak (w) positions, and which overstates rhythm within the definition of the template itself. The matching of linguistic structure with that metrical template is remarkably unconstrained, leaving the template underexploited.Footnote ² It seems to me a good idea not to hardwire rhythm in the universal characterisation of metre – and Fabb & Halle don't. There is no sense in which there is inherent rhythm, for example in verse made up of identical, quantitative elements, as in Japanese metre or in Greek spondaic metre.

3 Phonology in metre

It is everybody's intuition that there is a connection between phonology and metre, and the issue is raised in the very first paragraph of the book. How does this connection come out in Fabb & Halle's theory?

A good place to start looking is the basic categories. Fabb & Halle's theory has basic partitionings (properties that rules and conditions nail to positions in the grid) and units from which grid construction proceeds. The basic partitionings of elements that the authors suggest vary between straight phonological (e.g. heavy vs. light syllable, stressed vs. unstressed, tonal α vs. tonal β) and more indirectly phonological (e.g. alliterating vs. non-alliterating, maximum vs. non-maximum). Regarding grid construction, this study analyses metres that rely on syllables (most), moras (āryā) and primary stressed syllables (Beowulf), all phonological categories. With those three in place, we may anticipate that at least (all) stressed syllables and accented syllables should also be possible grid foundations in this model (no exhaustive list is given). The basic units, then, are essentially phonological.

When it comes to the phonology of a specific language and its metre, the connection is less clear. Phonological notions employed in metrical conditions are simply stated, or taken as self-evidently given from the language. There is no ‘theory of maxima’, for example, or group size at some grid level that is derived from the language. The fact that English metre is rhythmic is thus not organically connected with the phonology of the language. There are other instances where phonological connections are not made, even though the verse and the phonology look very similar. A case in point is French, where in the alexandrin metre there are two places in the line which must coincide with the end of a word: before the third verse foot counting from the end (caesura, marked ‘|’ in the example below (p. 141), from Jean Racine's Phèdre, IV.4.1170), and at the very end of the line.

1. (3)

   

These two half-line final syllables are also necessarily maxima, here stressed syllables. This end-strong pattern looks very much like the regular final stress of phrases in French. In Fabb & Halle's analysis (and not only theirs), however, it is stated as a purely metrical fact, suggesting that this is accidental. In principle, French alexandrins could just as well have required half-line medial maxima, I think counterintuitively.

A similar argument can be made for resolution in Greek and, in particular, Arabic. In Fabb & Halle's analysis, resolution plays no role in either of these metrical systems, largely due to the forefronting of the syllable count. In Greek, LL is reduced to L by a general routine taken to be the unifying feature of that metrical system. For Arabic, the solution is less elegant. With two metres, kāmil and wāfir, there are recurring positions which alternate between LL and H (six times in a normal line). The problem for the syllable count, and thereby grid construction, is the ‘extra’ L syllable in LL. To account for this, the authors propose that these metres admit ungrouped syllables, the hallmark of a loose metre. However, in the way in which the system is set up, an ungrouped syllable will now occur only in a LL sequence, and never in HL, LH or HH, making the analysis suspect.

The linguistic part of this theory is the computation, the actual counting and the basic units counted. Many linguistic models assume binarity as a basic property tied to the computation of linguistic structure. Fewer treat ternarity on a par with binarity. Otherwise, the gap between phonology and metre is maintained here, much as in traditional metrics. There is much at stake here, for any theory of metrics. Even if one takes an extreme view of the cultural influence on what metres occur in a given language, it is clearly the case that borrowed metres inevitably adapt to the borrowing language (much as in the case of word loans). The more of metre that we can derive from known linguistic properties, the better our chances of arriving at a properly constrained theory of metrics, which is grounded in language but which avoids duplicating it.

4 Ramifications of the theory

While the primary topic of this study is the presentation of a model and its workings in different systems, the authors have not quite managed to make clear the ramifications of the theory. For one thing, conditions are often stated in a semi-technical prose unsuited for extrapolation into generalisation and prediction. Here is a condition relating to Arabic metre (p. 198).

1. (4)

   

What other conditions are thereby admitted into the general, universal theory of metre? I have the feeling that there are rather too many for comfort.

An overgeneration problem is also in evidence at the level of the individual system. An example of this is the Spanish endecasílabo, where Gridline 2 asterisks correlate with stress. A typical grid is given below (p. 103), from Garcilaso de la Vega, Égloga I, line 43.

1. (5)

   

There are six different patterns of stress distribution across the ten metrical grid positions of the line, occasioned by variation of headedness at Gridlines 1 and 2, causing the highest column (the head of the verse) to end up in two different locations.Footnote ³ Subsequent deletions of asterisks at Gridlines 0 and 1 cause the head of the verse to shift locally, deriving the various patterns. Below, the patterns have been regrouped into two derivational families, with the mothers ((i) and (ii)) at the top. The head of the verse is underlined, and the varying headedness at Gridlines 1 and 2 is bolded.

1. (6)

   

This analysis captures the inductively established facts, but does not provide an understanding for why it is that 2–5–10 or 3–8–10, for example, are not permitted. These patterns can, however, be derived very easily from the two combinations of headedness at Gridlines 1 and 2 in (7).

1. (7)

   

As far as I can see, the proposed model does not appropriately constrain this metrical system, or provide a reason why certain patterns should occur rather than others, or why some patterns should be more frequent than others. Of course, my critique on this point relies on the assumption that the attested selection from the pool of patterns is indeed significant for the theory of metrics.

The set-up of the theory indicates what is relatively complex and what is not, what is basic and what is derived. This should yield expectations for the likelihood of a metre to occur in a given language, or within a metrical system. Fabb & Halle certainly refer to relative complexity, always preferring the simpler analysis, and they definitely relate structures to each other (see in particular the endecasílabo above, and the analysis of Arabic), but they don't extract analytical mileage from it, in terms of predictions of naturalness or frequency. Frequency of occurrence is otherwise a good source of empirical support. For the Arabic system, where metres are not as strongly correlated with genres or content as Greek meter, for example, all corpora single out ṭ awīl as the most popular metre, and three others (kāmil, wāfir and basīṭ ) as fairly frequent. The remaining eight to twelve metres are infrequent. This frequency pattern is related to structurally constrained rhythmic well-formedness in Golston & Riad (Reference Golston and Riad1997), and thereby serves as an empirical reflection of the model proposed there. In the Fabb & Halle model, no such correlation between what is basic and what is frequent can be made. Indeed, the circle 3 metres, which are basic in their analysis, are relatively infrequent.

Furthermore, performance may, but need not, provide cues to structure (e.g. Japanese metre (Cole & Miyashita Reference Cole, Miyashita, Dresher and Friedberg2006) and Sanskrit metre (Deo Reference Deo2007)). Performance is relevant in Fabb & Halle too, but only under the condition that the grid and the rhythm happen to correlate in a language, in this case Sanskrit.