Contextual Neutralization of Vowel Length: Evidence from Dutch

Based on recent experimental studies of word-final devoicing in languages like German and Catalan which show production differences between the neutralized and nonneutralized consonants., i t has been claimed that phonological neutralization is' incomplete'. It seems, however, that this claim is quite premature given that most studies have considered only the neutralizing phenomenon of word-final devoicing. In this paper, we examine a qualitatively different neutralizing phenomenon the neutral tzalion of vowel length in open syllables. We compared the duration of Dutch long vowels which are derived by an open-syllable lenghtening rule to those that are underlyingly long (cf. [da:len] < /dal+en/ vs. [ba:len] < /bal+en/). Our study shows that there is no difference in the duration between these vowels and that, at least, in this instance, contextual neutralization does lead to identical surface realization of distinct phonological segments. The identical realization of distinct phonological segments in a specif ic environment is termed contextual neutralization. This is the situation in which underlying feature distinctions are lost in a given environment while being retained elsewhere. One of the most famil iar examples of neutralization is the loss of voicing dist inctions in syllable-final posit ion in languages l ike German, Dutch, and Polish. In German, for example, the feature [voice] is distinctive such that both [*voice] and [-voice] obstruents occur in underlying representations ; in word-final posit ion however, [+voice] consonants become [-voice]. Thus, we find alternations in the nominative and genitive forms of nouns: [ra:t] [ra:des] Rad 'bicycle' as against [ra:t] [ra:tes] Rat 'advice', where the noun Rad has two surface allomorphs (rat) and (rad) as a result of the neutralization of word-final voiced consonants. Contextual neutralizations of this sort occur frequently in natural languages. In Russian, /a/ and, /o/ are neutralized to [a] in unstressed posit ion; compare the nominative singular and plural forms [mril] [mali] 'small', but [stol] [stali] 'table'. Turkish geminates and single consonants are neutralized word-finally; compare the possessive and nominative forms for 92 Lahiri.u Schriefers/Kuijpers 'soil', [hakl] [hak], with the respective forms for'truth'' [hakkl] [hak]. On the one hand, neutralization proc€sses cause the loss of surface contrasts of phonemes, and on the other, lead to surface alternations of the same underlying morpheme. But although neutralization is a wellaccepted phonological fact, recent work in phonetics has given rise to a controversy as to whether phonological neutralization is a 'complete' phonetic neutralization. The contextual neutralization rule which has been referred to most in this controversy is the rule of final devoicing which is phonologically motivated in a number of languages. Several studies [O'Dell and Port, 1983; Port and O'Dell, 1985; Dinnsen and Charles-Luce, 19841 have suggested that final devoicing in German and Catalan may not be complete, so that voiceless obstruents which are derived by the final devoicing rule are phonetically different from underived (and therefore underlying) voiceless obstruents either in terms of closure durations, vowel length or aspiration. This evidence, however, has not been unchallenged. Fourakis and Iverson [1984] argue that the differences obtained in the studies of O'Dell and Port [1983] may be an artifact of the task and that in natural linguistic contexts voicing is completely neutralized in German. The results of Fourakis and Iverson [1984, p. la9] support the hypothesis that production differences in German final obstruents observable from data elicited by a reading task are 'hypercorrect manifestations of linguistic insecurity' due to the fact that spelling reflects the underlying distinction between the voiced and voiceless consonants (cf. Rat and Rad). Their experiments show that eliciting the relevant articulations without orthographic cues pointing to the underlying distinction can lead to a complete neutralization of the voicing of word-final consonants. In reply to Fourakis and Iverson, Port and O'Dell [1985] argue that hypercorrection should actually increase the tendency of stops to be completely neutralized since it is a widely known fact that German has final stop devoicing. However, they acknowledge that, given the 'very small differences observed' [Port and O'Dell, 1985, p. 470lin their production experiments based on reading isolated words, and given the absence of such a difference in 'casual unattended speech' in Fourakis and Iverson's [1984] study, the differences cannot play a significant role in normal speech. Based primarily on Port and O'Dell [985] and other similar studies, Dinnsen [1985] questions whether there is any actual occurrence of neutralization. Most empirical work on neutralization has centered around word-final devoicing. Even if one takes the existing results as an indication that word-final devoicing is 'incomplete', it remains an open question whether this holds for other cases of neutralization. It seems quite premature for Dinnsen [1985, p. 27 5l to claim that the 'standard view of neutralization. . . is unfortunately without empirical support' before other qualitatively different neutralizing phenomena have been examined. Therefore in this paper we examine such a phenomenon: the neutralization of vowel length in Dutch open syllables. This instance of neutralization is different from final devoicing since it involves a different class of sounds, vowels rather than consonants. However, it is similar to devoicing because the context of neutralization involves a boundarv in this case an ooen svllable. Vowel Leneth Neutralization in Dutch 93 Vowel Length in Dutch Dutch has both underlying short and long vowels. Their distributional properties have been studied in great detail by phonologists and wil l not be discussed here [Booij, 1981; Trommelen, 1983; van der Huls t , 19851. Suffice it to say that short vowels (not including schwa) do not occur syllablefinally and therefore not word-finally, whereas long vowels can [van der Hulst, 19851. Thus it is possible to get contrasts of the type