Predicting the Unpredictable: Interpreting Neutralized Segments in Dutch

Among the most fascinating data for phonology are those showing how speakers incorporate new words and foreign words into their language system, since these data provide cues to the actual principles underlying language. In this article, we address how speakers deal with neutralized obstruents in new words. We formulate four hypotheses and test them on the basis of Dutch word-final obstruents, which are neutral for [voice]. Our experiments show that speakers predict the characteristics of neutralized segments on the basis of phonologically similar morphemes stored in the mental lexicon. This effect of the similar morphemes can be modeled in several ways. We compare five models, among them stochastic optimality theory and analogical modeling of language; all perform approximately equally well, but they differ in their complexity, with analogical modeling of language providing the most economical explanation.^*

1. Introduction

The way language-users deal with foreign borrowings or words that they have never heard before may show us what knowledge language-users actually have about the distribution of sounds in their language, and how they organize and use this knowledge (see e.g. Hyman 1970, Hooper 1976:10, Derwing 1980, Gussenhoven & Jacobs 1998:38). For instance, adaptations of foreign borrowings into different languages often show (depending on sociocultural factors such as attitudes toward borrowing) that borrowings may be changed in the receiving language so that they conform to the phoneme inventory of that language, which suggests that languages have restricted phoneme inventories not because they happen to have words with no other phonemes, but because of highly ranked wellformedness constraints (Jacobs & Gussenhoven 2000).

We used experimental neologisms to come to grips with final devoicing in Dutch, the phenomenon that underlyingly voiced and voiceless obstruents are realized identically in syllable-final positions. Experimental neologisms can provide new insights into the knowledge speakers use when dealing with neutralized obstruents of which the underlying [voice]-specification is unavailable to them. We carried out an experiment in Dutch that allows us to answer questions such as the following. Do listeners know when hearing a segment in a neutralizing position in an unknown word that this segment may be realized differently in a non-neutralizing position, that is, that its underlying representation may be different from its surface representation? If, for instance, both /t/ and /d/ are realized as [t] at the end of words in a language, and listeners hear the new word [fat], do they know that the [t] is possibly [d] before affixes? And if listeners know this and are forced to determine the realization of the segment in non-neutralizing positions, is their choice for one of the possible realizations random? Is it based on the relative phonological strengths of the corresponding phonemes in the language, that is, on the extent to which these phonemes are resistant to assimilation processes and on the ease with which they can be phonetically realized? Or is their choice based on the characteristics of phonologically or phonetically similar words in the lexicon? The answers to these questions take us beyond the traditional wisdom that underlying [voice] specifications are idiosyncratic only, and call for models that can deal with the probabilistic aspects of lexical structure.

2. Neutralization of [voice] in dutch

In Dutch, the feature [voice] is distinctive for obstruents that precede a vowel-initial suffix (except -achtig), the past-tense morpheme, or vowels that belong to the same morpheme as the obstruents. For instance, the words verwijden ([vεrυεidәn]) and verwijten ([vεrυεitәn]) only differ in the [voice] specification of their coronal stop preceding the infinitive suffix -en ([-әn]), but differ in meaning (see 1a). The [+voice] specification of the [d] of verwijden seems to be an idiosyncratic property of this lemma. Similarly, the [−voice] specification of the [t] of verwijten seems to be lexically marked and unpredictable.

In word-final position, [voice] is nondistinctive. The realization of obstruents in this position as voiced or voiceless does not depend on their realization before vowel-initial suffixes, that is, on their underlying [voice] specification, but mainly on the type of following segment. They...