Universal Boundary Theory: An Old Icelandic Case Study

Author: Sayantan Mandal, Université Concordia

1. Introduction

            This article discusses the nature and complexity of phonological representations, and more specifically the prosodic representations, with a focus on patterns of assimilation or preservation of final /r/ in underlying /lr/ and /nr/ sequences in Old Icelandic. I evaluate the role played by hierarchical syllable structures, and syllable node labels, in constituting barriers to assimilation, and highlight some issues inherent in Reiss’ (1994) syllable-structure driven approach to Old Icelandic. I then discuss a completely flat framework for prosodic representations, Universal Boundary Theory (UBT; Andersson, 2020), and proceed to show that despite of its severely impoverished representational complexity, UBT provides a more elegant account of Old Icelandic that circumvents the problems inherent to the traditional analyses, and along the way brings phonological representations closer to animal comparanda. I argue that this latter aspect, coupled with the more familiar arguments from economy, makes UBT a better fit for the biolinguistic program (Chomsky, 2007).

2. Syllable structure and barriers in Old Icelandic

            In Old Icelandic the underlying (UR) final consonant in the masculine nominative forms of strong nouns and adjectives is /r/. However, during the derivation of the surface representation (SR) from the UR in forms containing a preceding /l,n/ (for brevity, I will refer to /l/ and /n/ together with the capital /L/) the final /r/ can be either preserved (the first two tokens) or it can assimilate to the preceding /L/, surfacing as a geminate (the final four tokens). Consider the forms in the table below. All forms bear stress on the initial syllable.

In the table above the first pair of forms (Type A) contain a short vowel, and the surface forms illustrate the preservation of the underlying /r/. The second pair of forms (Type B), on the other hand, contain either a long vowel or a dipthong, and here we find assimilation of the final /r/ into the preceding /L/. Finally, in the last pair of words (Type C) the vowel preceding the /Lr/ sequence is also short, but since it is not in the initial syllable it is not stressed. Here too we find assimilation of the /r/ into the preceding /L/. The table below provides a descriptive summary, and highlights the fact that similar patterns obtain in medial positions as well.

Reiss (1994) argues that the Old Icelandic patterns illustrates the role that syllable-structures can play in creating barriers to assimilation. He argues that syllabification in Old Icelandic long vowels and dipthongs project a single branching nuclei, and following this, onsets are formed according to the standard onset maximization principle. Any remaining consonants can be assigned to the coda position, provided that sonority is decreasing from left to right. Additionally, a single unassigned consonants can be assigned to an appendix position that connects directly to syllabic root.

With this general schema in place, consider the syllabification of Type B forms below. The /f/ is the O(nset) position, the long vowel fills up both branches of the nucleus, and the /l/ is assigned to the C(oda) position. Reiss (1994: 331) argues that in Old Icelandic /l,n,r/ are identical in sonority, and as such the final /r/ cannot be assigned to a branching C, and must go under the A(ppendix) node.

above, assimilation delinks the X-slot from the /r/ and in turn is linked to the /l/ in the coda position. This yields the SR [fu:ll] from the underlying /fu:lr/, with assimilation spreading from the C node to the A node. Now consider the initial syllabification of a Type A form in (3) below. At first glance, there is little difference from the syllabification above, as seen in the figure on the left.

However, given that Proto-Germanic long-vowels remain long in Old Icelandic only in stressed syllables, Reiss (1994) argues that in the language stressed syllables project a branching nucleus. Accordingly, he proposes a resyllabification rule that allows for the nucleus incorporation of a coda L within the nucleus. This is illustrated in figure (4) in the sub-figure on the left below, while the figure on the right illustrates its effect on the Type A forms.

The re-syllabification process incorporates the /l/ in /selr/ within the nucleus, and it surfaces as [selr]. Next, consider the two figures in 5 below. The figure on the left illustrates the syllabification of a Type C form. The second syllable is unstressed and hence contains a non-branching nucleus, filled by the short vowel /a/. This leaves the L under the C node, and results in /r/ assimilating. The figure on the right shows that the process works similarly in medial syllables, with the only exception being that assimilation spreads from a C node to the O node.

Describing assimilation processes in terms of hierarchical syllable structures is not uncommon. For instance, Steriade (1988) reports that Chinese allows assimilation in the backness of a nucleus /ə/ to a glide in the coda, or to the onset but only when there are no codas. That is, α can assimilate to β only in (6a) and (6b) below, but not in (6c).

This option, Reiss acknowledges, is unavailable in Old Icelandic. First, note that in Old Icelandic assimilation targeting an underlying /r/ in /Lr/ sequences requires spreading features both within and across syllable boundaries. Second, given that features can spread from a C node to the O node of a following syllable it is also not possible to define this process in terms of c-command relations. Old Icelandic appears to require a negative definition, a domain only outside of which assimilation can occur – the nucleus. Alternatively, instead of stating the process in terms of the environment of the target, /r/, we could attempt to define the process in terms of the environment of the environment – assimilation of /r/ to a preceding /L/ is allowed when the /L/ is under a C node. In either case, note that a bare syllable-structure with c-command relations is not enough, and we obligatorily require labelled nodes to be a part of the repertoire of phonological UG.

3. Universal Boundary Theory

            In the history of phonological theory, analyses have often appealed to a wide variety of boundary symbols to mark various morphological and phonological domains. Chomsky and Halle (1968: 134), for instance, presents the underlying form /æd=vIs#or+y/ (advisory). An obvious problem associated with such postulations for a theory of mental grammars concerns the fact that, unlike segmental features, such boundary symbols are often hard to justify from a point of psychological and cognitive reality.  Since Goldsmith’s (1976) introduction of the autosegmental approach, a variety of hierarchically organized models, such metrical phonology (Liberman, 1975) and prosodic hierarchy (Selkirk, 1981), have sought to reduce the reliance on such diacritic symbols in phonological alternations by conceptualizing prosodic phenomena in terms of hierarchical syllabic structures. However, such structures also add to the representational complexity of phonological theory. Consider the analyses of Old Icelandic data presented in the preceding section. It makes no references to any boundary symbols, but it is stated in terms of syllable structures and labelled nodes. In fact, the crucial point of Reiss’(1994) argument is that patterns such as those seen in Old Icelandic require labelled nodes in addition to a hierarchical syllable template. Crucially, a bare notion of structural relations, such as c-command, fails to account for the observed patterns, necessitating explicit appeals to node labels. While not out of the realm of plausibility, such a grammar is highly enriched, containing both hierarchical structures and node labels, in addition to segmental features and mapping rules. In contrast, Universal Boundary Theory (UBT; Andersson, 2020) is a completely flat theory of prosodic representations that seeks to replace the various different boundary symbols proposed in phonological theory since the SPE (Chomsky & Halle, 1968), the hierarchical syllable templates, and all associated labels with a single, universal boundary symbol -- |, featurally encoded as [-Segment]. While superficially this might seem like a return to diacritics, UBT conceives of the universal boundary symbol exactly the same as any other segmental feature. The only quality that separates [-Segment] from other segmental features lies in the assumption that at the interface with phonetics, this features does not receive any phonetic reading, and hence has no perceptible phonetic realization. Besides this one difference, [-Segment] behaves exactly like any other feature – it can be inserted, or deleted, by extrinsically ordered rules, it can trigger a rule, and rules can make reference to it as an environment. Andersson (2020:771) provides an account of German glottal stop insertion using UBT. Briefly, in German Glottal stops are found word-initially, before a stressed vowel in hiatus, between prefixes and their roots, and with a subset of suffixes. While previous analyses of Glottal stop distribution in German appeals to the notion of hierarchical syllable templates, arguing that if the first syllable of a foot is vowel-initial, a glottal stop is inserted (Wiese 1996: 58-61), Andersson (2020) provides a completely flat account within UBT, which lacks any structural notion of syllables altogether. Besides the assumptions outlined above, Andersson (2020) also assumes that phonological derivation proceeds by phases, with each syntactic phase being spelled out by the phonology separately. It is assumed that the output of each phase is padded on both sides by an initial and final boundary, thus affording us two free boundaries with each phase. With just two rules, one each for boundary insertion and glottal stop insertion, ordered as below, Andersson (2020:771) successfully derives the observed glottal stop distribution[1].

What is crucial for our purposes, however, is Andersson’s (2020) argument that in the spirit of minimalism (Chomsky, 1995/2014) a theory of phonology that can account for such diverse phenomena without appeals to hierarchical syllabic representations is preferable, for reasons of economy, to one that requires hierarchy. The following discussions illustrate that UBT is, likewise, capable of accounting for the patterns of /Lr/ assimilation and preservation in Old Icelandic, requiring neither syllable trees nor node labels.

 In order to account for the distribution of the final /r/ in Old Icelandic /Lr/ sequences, I will propose two extrinsically ordered rules. The first rule, called boundary insertion (BI), inserts a boundary symbol after an underlying /L/ whenever there is a short stressed vowel (/Vˊ/) immediately preceding the /L/. The second rule allows for assimilation of /r/ (RA) in all /Lr/ sequences, so long as there are no intervening boundaries in between.

In the grammar in (8) above, a boundary is inserted after an underlying /L/ whenever there is a stressed short vowel immediately preceding it. This rule is ordered before the /r/ assimilation rule which assimilates any underlying /r/ to the immediately following L. Now consider the UR→SR derivations illustrated below.

The derivation table above leaves out the stress marks for sake of simplicity, but recall that in the forms above stress is always on the initial syllable. Thus, the forms in columns A, B and G are the only ones that are subject to the BI rule. In the forms in columns C, D and H the nucleus carries stress but the vowel is either long or a dipthong, and thus the structural description of the BI rule is not met and it fails to apply. Likewise, in the forms in columns E and F, the vowel preceding the /L/ is not in the initial syllable, and thus not stressed. Once again BI fails to apply in these forms. In the next step of the derivational process, RA enforces /r/ assimilation, but it fails to apply to the forms in columns A, B and G due to the intervening boundary between the r and L. These forms, therefore, surface with the underlying final /r/ preserved, while in every other form it undergoes complete assimilation with the preceding L. A quick comparison with tables 1 and 2 above will illustrate that we have now derived all the necessary Old Icelandic surface forms, but without any appeals to syllable structures or node labels. Indeed, in UBT there is no notion of a structural syllable, and the input to phonological representations are conceived of as completely flat beads-on-a-string, with interspersed boundary symbols sub-grouping the beads along the string. Further, unlike the hierarchical approach, the UBT rules require no negative domain definition, and resemble any other feature inserting/deleting rule familiar from rule driven frameworks.

4. Conclusion

            Given the discussions so far, it is reasonable to wonder what advantages UBT offers over more traditional hierarchical theories? There are, at least, two related reasons to assume that the framework discussed here is not without merits. First, as mentioned briefly before, is the reason of economy. A flat theory of phonology that can do without hierarchical templates, and labels, what other theories do with them is, by default, the more economic one. Assuming that phonological grammar consists only of features, and feature-bundles, significantly reduces the representational complexity of phonology. Likewise, assuming that the input to phonological computations are strings implies that our grammatical representations are leaner, containing only the bare minimum of structural relations – in this case, precedence. A second, but related, reason concerns the fact that UBT proposes a representational format that has significant parallels in the form of animal comparanda. Consider the representations of finch bird song structure and baboon social dominance relations below. Here n denotes notes, which are grouped into linear (bird-song) syllables, and these are in turn organized into motifs. Likewise, Baboon tribes obey a strict, transitive dominance hierarchy divided by matrilines; individuals from a single matriline take adjacent spots in the hierarchy, with mothers, daughters, and sisters from the matriline next to one another (Samuels, 2011: 14). In both cases, Samuels (2011) points out, the structures are flat, or linearly hierarchical with beads on a string grouped by interspersed boundaries.

These two structures are strikingly similar to the format that phonological representations take within UBT. Given the biolinguistic program such closeness to animal comparanda is only to be desired, especially when the framework proposed can still yield the empirical coverage of more complex models with significantly reduced representational complexity.

References:

·       Andersson, S. (2020). Creating boundaries and stops in German: An analysis in Universal Boundary Theory. Proceedings of the Linguistic Society of America, 5(1), 765-776.

·       Chomsky, N. (1995/2014). The minimalist program. MIT press.

·       Chomsky, N. (2007). Biolinguistic explorations: Design, development, evolution. International Journal of Philosophical Studies, 15(1), 1-21.

·       Chomsky, N., & Kenstowicz, M. (1999). Derivation by phase. An Annotated Syntax Reader, 482.

·       Goldsmith, J. A. (1976). Autosegmental phonology (Doctoral dissertation, Massachusetts Institute of Technology).

·       Liberman, M. (1975). The Intonational System of English. Boston, MA: MIT dissertation. 

·       Reiss, C. (1994). "The outcome of /Ir/ and /nr/ in Old Icelandic" The Linguistic Review, vol. 11, no. 3-4, 1994, pp. 329-350. https://doi.org/10.1515/tlir.1994.11.3-4.329

·       Samuels, B. D. (2011). Phonological architecture: A biolinguistic perspective (Vol. 2). Oxford University Press.

·       Selkirk, E. (1981). On prosodic structure and its relation to syntactic structure. InThorstein Fretheim (ed.), Nordic Prosody II. 111–140. Trondheim: TAPIR.

·       Wiese, R. (1996). The phonology of German. Oxford: Oxford University Press.

[1]Andersson’s (2020) discussion of German delves further into patterns of devoicing, which will not be delved into here for reasons of space.