Abstract: "Language processing and segmental OCP effects"
by Stefan Frisch
In Bruce Hayes, Robert Kirchner, and Donca Steriade, eds., Phonetically-Based Phonology, Cambridge University Press (2004).
Department of Communication Sciences and Disorders
University of South Florida
4202 E. Fowler Ave, PCD 1017
Tampa, FL 33620
The Arabic verbal roots are subject to a long-distance phonotactic constraint that is well known for its implications for autosegmental representation (McCarthy 1986, 1988, 1994). In this constraint, originally proposed as an instantiation of the Obligatory Contour Principle (Goldsmith 1979), repeated place of articulation features are not allowed within a root. Subsequent research has shown that the details of consonant occurrence in the Arabic roots are complex, with the strength of the phonotactic restriction gradiently dependent on the similarity of the consonants involved, the presence of intervening segments, and the contrasts available in the segmental inventory of Arabic (Pierrehumbert 1993; Frisch, Broe, & Pierrehumbert forthcoming).
The gradience of the phonotactic patterns in the Arabic lexicon provide strong evidence for a functional phonetic motivation for the constraint. The similarity avoidance constraint in Arabic is quantitatively dependent on similarity, distance between segments, segment frequency, and segmental position in the word. No formal model that prohibits feature cooccurrence like the autosegmental OCP can capture the richness of the patterning. A wide range of evidence from psycholinguistics suggests that processing a sequence of similar items is more difficult than processing a sequence of dissimilar items. Thus, we can account for the presence of similarity avoidance constraints in the phonotactics of Arabic as a consequence of functional pressure to make language processing as easy as possible. I claim that the richness of phonotactic patterns directly (quantitatively) reflects the functional explanation. In this way, statistical analysis of the lexicon provides a novel type of evidence for functionally-motivated constraints and rules out alternative formal explanations (see Hawkins 1994 for similar arguments at the syntactic level). Statistical patterns in the synchronic lexicon arise as the result of a diachronic influence of processing difficulty on language change. Over time, functional pressures on the language have shaped the lexicon by influencing borrowing, the creation of nonce forms, and the loss of lexical items.
Despite the diachronic origin of the similarity avoidance patterns, native speakers are aware of these patterns, so they must be considered a part of the synchronic linguistic knowledge of the speakers. The cooccurrence constraint on homorganic consonants influences metalinguistic judgments of acceptability for novel roots (Berent & Shimron 1997; Frisch & Zawaydeh 2001) and the accommodation of borrowed lexical items (Frisch, Broe, & Pierrehumbert forthcoming).
Moreover, a phonotactic constraint based on processing difficulty should be universal. In fact, similarity avoidance constraints for homorganic consonants like those in Arabic have been found in a wide range of languages, such as English, Javanese, and Ngbaka. Analogous constraints that apply to repeated laryngeal features rather than repeated place features are also attested across unrelated languages such as Sanskrit, Hausa, and Souletin Basque (MacEachern 1999). Further, in cases where lexical patterns have been analyzed statistically, the cooccurrence patterns are gradient and quantitatively depend on similarity (Berkley 1995, 2000; Buckley 1997; Frisch 1996; Pierrehumbert 1993).
In this chapter, a functional account of segmental OCP effects is given. Languages avoid sequences of repeated similar segments because they are difficult to serialize. An explanation for the difficulty of repetition can be found in models of language processing. Current language processing models use activation and competition in a neural network of linguistic units to account for similarity dependent error rates in perception and production. In an activation/competition model of the encoding of a serial sequence, the units (e.g. segments) to be serialized must be activated and deactivated in the proper sequence (Dell, Burger, & Svec 1997). For a segment that has already been encoded, the node in the network corresponding to that segment has fired, and that node must be inhibited so that it does not continue to fire. Simultaneously, for a segment that is soon to be encoded, the node corresponding to that segment must be excited so that it is ready to fire at the proper time. If a sequence involves a repeated segment, the periods of inhibition and excitation may overlap and disrupt encoding of the correct sequence.
The chapter is organized as follows. Section 2 provides an overview of current theories of language processing, focusing on models of segmental processing. Section 3 reviews the segmental OCP patterns in Arabic and other languages, and shows these patterns are gradient and similarity dependent. Section 4 presents the processing account of similarity avoidance constraints and reviews some outstanding problems with this account. Section 5 concludes the chapter with a brief summary.
Download a draft version of this chapter
Return to the Phonetically-Based Phonology home page