It’s all about size

This squib (i) suggests a new analysis of English d/t-lenition, (ii) argues for a particular analysis of the internal structure of coronals and (iii) hints at a general theory of lenition.

1. The problem

Within the framework of Government Phonology (GP; Kaye, Lowenstamm & Vergnaud 1985, 1990), Harris & Kaye (1990) argued that feet (as in wáter) form governing domains, with the stressed nucleus governing the unstressed one. A foot-internal intervocalic consonant was claimed to be “in the way” and thus prone to lenition. This was offered as an explanation of intervocalic tapping (wá[ɾ]er) and glottalisation (wá[ʔ]er) as found in various varieties of English. Harris (1997) offered a variation of this and argued that foot-internal consonants are weak because they are far down on the licensing scale. That is, they are licensed by unstressed nuclei which are themselves licensed by stressed nuclei. In wáter, the stressed nucleus licenses the unstressed nucleus, the unstressed nucleus the intervocalic consonant, which is two steps removed from the ultimate head of the domain, the stressed vowel, and thus weak. Furthermore, stops are the most complex objects (highest number of elements), hence the first to be targeted by lenition. How d/t is reduced will depend on the variety in question, but both phenomena can be expressed as the loss of (different) elements. This is shown in (1).

(1)

a.

b.

c.

(1a) illustrates the relationship between a stressed and an unstressed nucleus, labelled “(1)”. In Harris & Kaye (1990) this is interpreted as a relationship of government, in Harris (1997) as one of licensing. The unstressed nucleus in turn licences its preceding onset, labelled “(2)”. O₂ finds itself in a detrimental position; either due to government (Harris & Kaye) or to a weak licenser (Harris 1997). Accordingly, it loses (different) elements, leading to (1a) or (1b).

While both analyses explain why foot-initial coronals (e.g. retáin) resist reduction (not trapped within a governing relationship/licensed directly by the domain head) and why only stops are affected (high complexity), they fail to address why only coronal stops are targeted. In the element calculus of the time (1990’s), coronals and, say, labials were equally complex: The elements common to d/t were R, ʔ; those common to b/p U, ʔ. Yet, labials do not lenite (pépper does not go to *pé[w]er, even though its intervocalic consonant intervenes in a governing domain). A disposition for lenition cannot be blamed on the number of elements.

Later on, at least in some flavours of the theory, the coronal element R was replaced by A (e.g. Cyran 1997, Goh 1997, Kaye 2000), i.e. the same element characterising non-high vowels. This allowed for a non-arbitrary analysis of English intrusive r, cf. Broadbent (1991, 1999). Intrusive r requires the nucleus preceding the insertion site to contain A, e.g. Shah-[r]-of Persia vs. Rabbi-*[r]-of Budapest. If r is nothing but an A in onset position, intrusive r simply becomes a special case of glide formation: An element spreads from a nucleus to an empty position, like see-[j]-ing (I in onset position from preceding nucleus) in some varieties of English. Further corroboration for this replacement came from the observation that R did not seem to occur in nuclear positions anyway, so the equation of R and A remedied that gap as well. However, equating the two did not lead to a better understanding of tapping. Labials and coronals were still equally complex. (Labeling coronals as “unmarked” is of course uninformative as it reveals nothing about the reason for that unmarked state.) With the special status of coronals unsolved, Harris & Kaye’s/Harris’s analyses failed to reach explanatory adequacy, remaining (partially) arbitrary.

2. What makes coronals special?

Why is it then that English singles out d/t for lenition? I submit that the solution comes from one particular recent development of the theory, viz. GP 2.0 (Pöchtrager 2006, Kaye & Pöchtrager 2013). GP 2.0 reinterprets several properties as structural which precursors of that theory (referred to here as GP 1.x) expressed by elements. Currently, there are three elements left in GP 2.0: I, U, and L. For the elements H (voicelessness/high tone), ʔ (stopness), and, crucially, A (coronal/non-high), GP 2.0 argues that they be replaced by (different kinds of) structure. That ʔ is replaced by structure goes back to an argument by Jensen (1994) to the extent that stopness arises out of a relation holding between two points. For example, in Pulaar, gemination of a fricative automatically turns it into a stop. Gemination means that a consonant is called upon to govern a second position (in GP 1.x parlance). Stopness is then not an element, but the result of a particular relationship (Jensen 1994, Pöchtrager 2006). This makes the change from fricative to stop non-arbitrary; it is not achieved by the addition of an element out of thin air, but by a certain structural relationship being established. Likewise, the difference between fortis/lenis consonants, as in English, should not be expressed by the element H: What sets apart the final consonants in a pair like bid/bit is not that t contains H while d does not, but rather that d is simply the structurally smaller version of t. Under such a reinterpretation, having a shorter vowel in bit than in bid (the famous “pre-fortis clipping”) is no longer surprising: Room that is not taken up by the consonant will be taken up by the vowel.

Arguments for a replacement of A by structure are similar in nature and come from the behaviour of both coronal consonants and non-high vowels in a variety of languages; examples from English, German, Finnish, Sason Arabic etc. are discussed in Pöchtrager (2010, 2012, 2015). We will only look at English here. In English, we normally only find VCC (ant, enter) or V:C (leave, beaver). Yet long vowels before clusters can occur if the cluster fulfills certain conditions on coronality (the presence of A). That is, longer structures (V:CC) are made possible by A. The patterns are complex and can be broken down by cluster type. Nasal + stop clusters have to be coronal throughout to allow long vowels: haunt vs. *haump, *haunk. Fricative + stop (particularly in Southern British English) clusters display a complex interplay with the vowel: While st (with A in both consonants) allows any long vowel before it (priest, boost, paste, host, last etc.), sk/sp (A in only one consonant) can only be preceded by long a (task, clasp, *teesk, *cloosp). In other words, if both consonants contain A, the vowel is free to be any long vowel, but if only one consonant does, the vowel must provide another A (which has to be by itself it seems). Whatever the particulars, A plays a key role in allowing longer structures.

Since A consistently interacts with structure, it must be structural itself. The difficult question is how exactly to represent that idea in the theory. One way (Pöchtrager 2006, 2010, Živanovič & Pöchtrager 2010) is to assume that A is replaced by a certain piece of structure where not every position within that structure is necessarily taken up. Positions that are not taken up could then be borrowed by a neighbouring object. If, say, in a cluster st there is extra room that is unused, a preceding vowel could borrow that room and be long. The formal implementation of that "certain piece of structure" was thought to be an adjunction, i.e. the splitting up of a head into two identical copies of itself. This is shown in (2), with an onset head without adjunction (labials, velars) in (2a) and adjunction structures in (2b–c).

(2)

a.

b.

c.

Note that the adjunction structure was assumed to come in two flavours, one where all positions are taken up ((2b); no borrowing possible) and one where there is an unused position (2c). Given these two flavours, the number of objects expressible with an adjunction structure (2b–c) is essentially twice of what can be expressed without (2a), which seems to be on the right track: In many languages, the number of coronals outnumbers other places of articulation. For example, in English we find [t d θ ð s z l n r ɾ], but only [p b f v m w].

As a consequence, objects that used to contain the (now defunct) element A are thus reinterpreted as structurally bigger than those without. (This applies to vowels and consonants alike, which we will come back to.) Thus, coronals are bigger than velars or labials, and this of course is crucial in understanding tapping: I agree with Harris & Kaye (1990) that stops are phonologically big objects (cf. what was said about ʔ above) and bigger than other manners, and if coronals are even bigger, then it follows that d/t are the biggest objects. Thus, it is unsurprising that English singles them out as a target for lenition. We have a non-arbitrary account that explains, rather than stipulates why d/t are affected: They are the biggest consonants in the system, hence the first (and in English: the only) ones to be targeted.

Let us come back to the size of stops. In the accounts of Harris & Kaye (1990) and Harris (1997), stops were more complex in terms of elements than corresponding fricatives; trivially, stops contained the stop element ʔ, fricatives or approximants did not. In GP 2.0, both ʔ and H are replaced by structure as well. Stops are not big in terms of melody (high number of elements), but in terms of structure. The following representations make this clearer:

(3)

a.

b.

c.

Non-coronal fricatives have a one-layered structure (3a), while non-coronal stops have two layers (3b). (The relationship between the onset head xO₁ and its sister x₂ in (3b), indicated by an arrow, is thought to be part of what makes up a stop, but of no relevance here. Further details can be found in Pöchtrager 2006 and Pöchtrager & Kaye 2013.) Since the coronal element is replaced by extra structure, coronal stops will be even bigger (3c), where the arrow in brackets between xO₁ and its sister x₂ represents the two options given in (2). The special status of coronal stops falls out.

This proposal does not claim that only coronals undergo lenition. In initial position, Danish (Harris 1999, Basbøll & Wagner 1985, Basbøll 2005) allows both fortis and lenis stops, while a stop in foot-internal position can only be lenis; irrespective of place. However, Danish fortis stops as a class are still the biggest objects of the system, suggesting that Danish is simply less “tolerant” about what it allows in its weak positions; the cut-off point is lower than in English.

3. Extension to vowels

This account of lenition in consonants also extends well to vowel reduction in unstressed position as e.g. in Portuguese or Catalan (Cristofaro-Silva 1992, Harris 1997, 2005, Mateus & Andrade 2000, Wetzels 1995, Wheeler 2005): In many cases non-high vowels become high(er), as seen in several varieties of Brazilian Portuguese (4).

The chart illustrates that in (different types of) unstressed position, fewer and at the same time more basic vowels occur. The reduction of unstressed o/e (earlier: containing an A element) to u/i (no A element) receives an explanation parallel to that of tapping. A is replaced by more structure, and reduction to u/i then implies the loss of structure in the weak part of the foot. This could easily have been explained as the loss of an A element as well and does not yet say anything about which account is preferrable. However, the complex reduction pattern in (4) makes clear that a more fine-grained model is called for, and here is where an account in terms of elements will be too crude: In pre-stressed position, open (but not closed) mid vowels are barred; an identical situation pertains in unstressed position in Italian. (Nevins 2012 argues that Northeastern Brazilian Portuguese allows only open mid vowels in unstressed position, but offers practically no evidence and ignores that those vowels are the result of a harmony process, cf. Cobb 2003, Segundo 1993. Given that, there is no contradiction to the claims made here.) Many versions of GP employ headedness, i.e. the relative weighting of elements, to express, amongst other things, the difference between open and closed mid vowels (with open mid vowels more A-like, closed mid vowels more I/U-like). But in such a formalism reduction of [ɛ/ɔ] to [e/o] involves the random replacement of one head by another. Both [e]~[i] and [ɛ]~[e] make the vowel higher, yet they involve different formal means (loss vs. head rearrangement). In a structural approach like GP 2.0, a unification can be achieved with the following modification to the adjunction approach outlined above: As proposed before, A is replaced by extra structure. But assume now that [ɛ] contains two layers of such extra structure, [e] one, and [i] none. In other words, the three vowels are points on a uniform scale, and both [e]~[i] and [ɛ]~[e] involve the loss of exactly one layer of structure. (5) exemplifies this idea.

(5)

In (5), vowels are represented as x-slots along with their projections (x′ etc.). The representational format is somewhat vague, but suffice it to say that heads are underlined (those are the x-slots that can project), that a head can project maximally twice (x″), and that vowels involve maximally two such projections on top of each other. For example, in the representation for [ɛ] we have a head projecting twice (up to x″), and that sister to that itself is a tree projecting up to x′. Such a format allows us to represent reduction in a uniform fashion, as a loss of a layer of structure. It also allows us to unite cases that contain melody, such as the front vowels in (5), with those without, such as the central/back vowels in (6).

(6)

The structures in (5)–(6) are more fine-grained than the ones in (2)–(3). Given that (5)–(6) allows the expression of reduction in a non-arbitrary fashion, it will be the structures in (2)–(3) that will have to be modified. Crucially, however, the explanation for tapping/glottalisation in English will not be lost; size is still the crucial factor. While the exact details of the formalism are still to be worked out, the generalisation emerging is clear: The larger the object, the more likely that it will undergo reduction.

The claim that (old) A (now: extra structure) underlies coronals has been criticised, cf. recently Backley (2011) or Nasukawa & Backley (2011), who suggest that either I or a combination of A and I characterises coronals. I believe (and have argued elsewhere) that many of their arguments are mistaken and/or susceptible to alternative analyses (Pöchtrager 2013). Note however that the above link between lenition of coronal stops and reduction of non-high vowels is only expressible if both form a natural class. This can be achieved by both old A and the new structures, whichever form they should take, but not if coronals are characterised by I.

4. How much does size matter?

The preceding discussion has shown that tapping and vowel reduction can be seen as instantiations of the same underlying idea, i.e. the loss of structure (which replaces old A). This proposal also raises the more general question whether all lenitions are about structural size. As pointed out before, in GP 2.0, A is replaced by structure, but so are the old elements ʔ (stop) and H (voicelessness). Certainly stopness is a lenition target (Spanish, Catalan, Danish) and so is voicelessness (Danish). It is then tempting to make the bold claim that all lenition/reduction is structural.

Such a claim seems to fly in the face of cases like Catalan, where [e]/[ɛ] reduce to schwa. This not only involves the loss of structure, but also of the element I (contained in [e]/[ɛ], but not in schwa). Notice, however, one crucial link: both structure and an element are lost. If it can be shown that melody is never lost without structure, then the bold claim of the preceding paragraph can be upheld. The type of reduction we find in Catalan does not necessarily contradict that claim. Reduction means the loss of a tree branch, and if I sits on that very branch, then it follows the the loss of melody is nothing but “collateral damage”. This is illustrated in (7).

(7)

In (5)–(6) the element I was sitting at the bottom of the structure. Assume now that a language allowed an element to sit at higher levels, too, as in (7). If reduction in that language means that all structure is lost save the lowest layer, then any element housed higher up will be lost as well. Both [ɛ] and [e] will end up as schwa. This allows for a simple explanation of the Catalan facts.

As illustrated, melody is lost as the result of structure being lost. This idea can be taken further: If elements can sit on different branches in different languages then they will differ in how easily they are amenable to reduction. For further elaboration of this idea cf. Pöchtrager (2016)

Summing up then, a structural approach to lenition and reduction then not only allows us to make sense of why coronal stops are singled out in English, but it can also help us to further our understanding of weakening phenomena in general.

References