Glides
In my MA Thesis, I explored the behavior of the palatal glide in Turkish Vowel Harmony. Unlike vowels, the palatal glide in Turkish is transparent to harmony. By comparing the behavior of the glide and of the palatal lateral (which does participate in Vowel Harmony), I argued that Vowel Harmony cannot be a syllable-based process, but must instead occur at the featural level.
My MA thesis led me to look at the different behavior of glides cross-linguistically. In some languages, glides pattern with consonants in various phonological processes, while in others, they pattern with vowels. In my dissertation, I argued that there are two types of phonologically distinct glides: "phonemic" glides (which pattern with consonants) and "derived" or "allophonic" glides (which pattern with vowels). Using strict criteria, I argued for the existence of underlying glides (contra other researchers who have argued that all glides are derived from underlying vowels) based on data from 9 unrelated languages (Turkish, Pashto, Karuk, Sundanese, IT Berber, Pulaar, Cree, Tahltan, Yawelmani).
Turkish Prosody
My work on Turkish prosody has looked at both the lexical level of accent and on larger units such as genitive NPs (e.g., the dog's house) and compounds (e.g., dog house). In a paper published in JIPA, I measured F0, duration, and intensity in 20 noun and verb pairs that differed minimally in their location of lexical accent. Based on these measures, I argued that Turkish lexical accent behaves more like pitch-accent systems than stress-accent systems.
My work on Turkish genitive NPs, such as "my father's restaurant", examined the intonational contours of such phrases with different types of lexical accent. I systematically varied the location of accent (final vs. non-final) in both the first and the second NP in GNPs and compounds in order to test the interaction between lexical and intonational pitch-accents. I found that each NP constitutes a separate Accentual Phrase and carries its own lexical accent. The first NP of the GNP also exhibits a H- continuation rise on the final syllable if the lexical accent is non-final. I also found that speakers vary in whether they allow dephrasing of one of the two APs in the GNP. This study further confirmed that only the lexical accent of the first member of the compound is retained and the second member surfaces with no lexical accent.
Cross-Language Speaker Perception
In collaboration with Steve Winters and David Pisoni, I have also examined "indexical properties of speech". Indexical properties are those characteristics in the speech signal that provide talker-specific information (e.g. age, gender, dialect, size and shape of the vocal tract). Using a database of 20 German-English bilingual talkers saying 360 English and 360 German CVC words, we tested whether there is a sufficient amount of indexical information that transcends language ("language-independent" indexical properties). In one experiment, we trained 40 monolingual English listeners to identify 10 speakers (5 men, 5 women) over the course of four days. Twenty listeners were trained with English words, while the remaining 20 were trained with German words. After voice training, all 40 listeners were asked to identify the speakers in both languages (blocked by language). Results revealed several interesting facts. Both groups of listeners learned to identify the talkers at the same rate, indicating that the specific language of training (and the linguistic experience of the listeners) does not play a role in learning to identify talkers. An effect of language was found, however, when examining listeners' performance on the final day of testing. Listeners who were trained in English performed significantly worse when the language was switched to German. Listeners who were trained on German, however, did not show a decline in their performance in the untrained language (in this case English). In other words, listeners were better able to generalize to an untrained language when they have linguistic experience/competence in that language. Furthermore, though the performance of listeners who were trained on English declined when they were faced with the German stimuli, they maintained above chance performance, indicating that they were still able to perform the task.
Another study examined whether the "familiar talker advantage" (increased intelligibility for familiar talkers) transfers across languages. Listeners familiarized with bilinguals in English showed the expected benefit: They responded with fewer phoneme errors in a word recognition task. In contrast, listeners familiarized with bilinguals from German stimuli did not show this intelligibility boost. Listeners must therefore have knowledge of linguistically relevant articulations to exhibit an advantage in a linguistic processing task.
