II.           Experimentation in Phonology 

        No claims in phonology are above doubt: the existence of the phoneme, syllable, or the feature [voice]; the reconstructed Proto-Indo-European form for Sanskrit budh-; that speakers know the posited rule-governed phonological link between the pair of words repose/ repository. All of the ses are potential subjects for experimental study. It is a matter of research strategy, the availability of reliable experimental methods, and the amount of personal commitment we have to one belief or another which determines which issues one chooses to address experimentally. So many experimental paradigms have been proposed for testing phonological hypotheses. I give representative examples from different domains in phonology.


                         i.           Experimental Assessment of the Distinctive Feature of Speech

One of the most fundamental tasks of phonology is to establish how different linguistic messages are conveyed by sound. Whether it is lexical differences or grammatical function, distinct messages must have distinct physical encodings, whether these are paradigmatic (different ciphers from a finite inventory of ciphers) and/or syntagmatic (different permutations of the ciphers). This is far from a trivial issue and certainly not one to be determined unequivocally by the unaided ear. Well-established methods exist for discovering the physical correlates of different linguistic messages in cases where they are uncertain or disputed.

Consistent differences may be sought in the physiological or acoustic domains but the relevance of any difference found must ultimately be validated in the perceptual domain (Lehiste 1970). For example, in a series of instrumental and experimental studies, Lisker and Abramson (1964, 1967, 1970) found that in initial position (before stress), the distinction between pairs of English words like paid vs. bade, tie vs. die, cool vs. ghoul, is carried largely by the relative timing of voice onset after the stop release, that is, what is called VOT (for Voice Onset Time): the phonemes /p t k/ showed a substantial delay in VOT (modal VOT = 50¡X70 msec) whereas /b d g/ had a short VOT (modal VOT = 0¡X20 msec). Phonetically, this contrast is said to be between voiceless aspirated stops and voiceless unaspirated stops. Perceptual studies demonstrated that VOT was the dominant cue for such lexical distinctions although several secondary cues also played a role (Lisker 1986). Although this contrast among stops is commonly attribute d to presence vs. absence of voice, voicing per se plays only a secondary role in this environment and in other positions in the word as well (Denes 1955; Raphael 1972). Lisker (1957) showed that in intervocalic position, in addition to voicing, the duration of stop closures helps to cue lexical distinctions such as rapid vs. rabid, where the voiceless stop is longer.

The stops that appear in prevocalic clusters after Is!, e.g., spade, sty, school, may only be voiceless unaspirated. Lotz et al. (1960) showed that to English speakers these are perceptually most similar to the stops in bade, die, and ghoul, i.e., /b d g/ (though they are not completely identical, Caisse 1981). Thus, although traditionally the prevocalic stops in paid and spade would be counted as allophones of the same phoneme / p / in English, there is greater physical and perceptual similarity between the stops in bade and spade.

                       ii.           Can Phonetically Different Sounds Be Psychologically the Same?

This still leaves open the question of whether native speakers regard the voiceless unaspirated stops in sC- clusters to be psychologically similar to the voiceless aspirated or the voiceless unaspirated in absolute initial position. This question was investigated by Jaeger (1980, 1986) who used the so-called concept formation method to address the question of the assignment of allophones to the /k/ phoneme. Without being given any more instructions than (approximately) ¡§assign the following words to two different categories depending on the pronunciation at their beginning,¡¨ linguistically naive subjects were first presented orally with uncontroversial examples with initial stops such as kiss, chasm, cattle, and quake designated ¡§category,¡¨ intermixed randomly with noncategorv examples, grip, gash, lime, ceiling, chest, and knife. Initially subjects were given feedback on each trial, i.e., they were told whether their category assignment was correct or not. If they reached some preset criterion of performance in this training, they were then presented with words containing the stop allophone whose phoneme membership was controversial, such as school and scold. This time there was no feedback. If they put these words in the same category as cool and cold it would imply that they regarded the [k] and [kh] as somehow psychologically equivalent. In fact, this is what they did. (See also Ohala 1986.)

                     iii.           Experiments on Morpheme Structure Constraints  

Esper (1925) explored the effect of analogy on the change in phonological shape of words and morphemes. His experiment was a task where he required his subjects to learn the names of 16 objects, each having one of four different shapes and one of four different colors. (He trained them on 14 object-name associations but tested them on 16 in order to see if they could generalize what they learned.) In three different experimental conditions, each with a different group of subjects, the relationship between the names and properties of the objects differed. 

The names presented to subjects in group I were of the sort naslig, sownlig, nasdeg, sowndeg, where nas- and sown coded color and -lig and -deg coded shape (though they were not told of their ¡§morphemic¡¨ constituents). Since these names consisted of two phonologically legal morphemes, this group could simplify their task by learning not 16 names but 8 morphemes (if they could discover them) plus the simple rule that the color morpheme preceded the shape morpheme in each name. Group 3, a control group, were presented names that had no morphemic structure; they had no recourse but to learn 16 idiosyncratic names. As expected, group 1 learned their names much faster and more accurately than group 3. Of interest was the performance of group 2 which, like group 1, were presented with bi-morphemic names and thus could, in principle, simplify their task by learning just eight morphemes. But, unlike group 1, the rnorphemes were not phonologically legal for English, e.g., nulgen, nuzgub, pelgen, pezgub (where now nu- and pe- were color morphemes and -lgen and -zgub were shape morphemes, the latter two, of course, violating English morpheme structure constraints). Could the subjects in group 2 extract the hidden morphemes and perform as well as those in group 1? Apparently not: their performance was similar to (and marginally worse than) that of group 3, which had 16 idiosyncratic names to learn. Furthermore, analysis of the errors of group 2, including how they generalized what they¡¦d learned to the two object-name associations excluded from the training session, revealed that they tried to make phonologically legal morphemes from the ill-formed ones. 

Esper¡¦s experiment achieved his goal of showing the force of analogy in language change, i.e., paradigm regularization, but it also demonstrates the psychological reality of morpheme structure constraints.


                     iv.           Experiments on Phonological Change  

One of the earliest accomplishments of phonology was the development of a method, the comparative method, which allowed one to reconstruct the history of languages, in particular the changes over time in the phonological forms of words (Rask 1818; Grimm 1322). To oversimplify, the comparative method consists in finding an optimal single unbranching path between pairs or groups of words judged to be cognates, where the ¡§path¡¨ consists of (a) intermediate forms between the two, one of which is the ¡§parent form¡¨ and (b) sound changes which operate unidirectionally and convert the parent form into the attested daughter forms. Historical phonology might seem at first to be an unlikely domain for experimentation since most of the events of interest occurred in the inaccessible past and thus cannot be manipulated by the experimenter. But if one is willing to make the unformitarian assumption, that is, that whatever caused sound changes in the past is still present and causing sound changes now, then although we cannot be there when Proto-Indo­European kw changed to Greek p, e.g., PIE ekwos ¡§horse¡¨ > Gk. hippos, we may be able to contrive circumstances where the same or similar changes occur in front of our eves or our microphones. In fact, the parallelism between dia­chronic and synchronic variation has often been remarked by researchers and sometimes has led to laboratory-based studies of sound change (Rousselot 1891; Haden 1938). One of the most fruitful areas of experimental phonology, then, involves studies on the phonetic influence on sound change or on pho­nological universals in general (see, e.g., Lindblom 1984; Wright 1986; Kawasaki 1986; Kawasaki-Fukumori 1992; Stevens 1989; Goldstein 1983; Ohala 1992, 1993).

One of the most common processes evident in sound change is assimila­tion and one of the common textbook examples of it is the case of medial heterorganic clusters assimilating in Italian: Late Latin octo> Italian otto ¡§eight¡¨. Such assimilations are overwhelmingly of the form -C1C1- > -C2C2-; rarely does C2 assimilate to the place of C1 (and many of these cases could be reanalyzed as involving a different process; see Murray 1982). Such a change is usually attributed to ease of articulation or conservation of energy (a heterorganic clus­ter requiring more energy than a homorganic one). But if so, why is it C1 that usually changes, not 02 Expenditure of articulatory energy is presumably cumulative through an utterance and thus would be greater by the time C2 was reached than C1. Thus we might expect C2 to assimilate to C1, just the reverse of what is found. Such doubts lead us to entertain an alternative ex­planation for this process.


                       v.           Experiments in Lexical Representation  

 Lahiri and Marslen-Wiison (1991, 1992) put underspecification theory into the empirical arena. They suggested that the lexical representations p05-ited by phonologists ¡§correspond, in some significant way, to the listener¡¦s mental representation of lexical forms. . .and that these representations have direct consequences for the way. . . the listener interprets the incoming acous­tic-phonetic information.¡¨ Lahiri (1991) argued specifically that ¡§the surface structures derived after postlexical spreading do not play a distinctive role in perception; rather, a more abstract underspecified representation determines the interpretation of a phonetic cue.¡¨

Ohala and Ohala (1993) attempted a replication of Lahiri and Marslen­Wilson¡¦s experiment,5 but restricted the subjects¡¦ responses to just one of two choices, e.g., when presented with an end-gated version of rube the choices specified on the answer sheet were room and rube. The results are seen ln figure 24.1 as triangles superimposed on the circles representing Lahiri and Marsien-Wilson. This curve appears to be quite similar to theirs, but there is a crucial difference: a statistical analysis is possible in the Ohala and Ohala case. In fact these results show that listeners made the correct identification of the stimuli up to the point where the consonant joined the vowel 82.8 percent of the time; this is highly significant (x2 = 92.03, 1 df, p < .001). Conversely, the same statistic shows that the subjects¡¦ choice of the incorrect CVN responses were much below chance level. 

( Adapted from  John J. Ohala. Experimental Phonology. The Handbook of Phonological Theory. John A Goldsmith edition.  1996 )