This summer through CURF’s Jumpstart for Junior’s grant, I carried out a linguistic research project into the effects of conditioned mergers on underlying phonological representations by investigating the interaction of intervocalic flapping and pre-flap vowel duration.
The pronunciation of a word can be divided into two levels of representation: the surface phonetics, corresponding to how words are actually pronounced, and the underlying phonology, corresponding to how words are cognitively stored. These two levels of representation do not always have a one-to-one correspondence. Allophonic splits result in a single underlying sound having two distinct pronunciations. For example, the English /p/ sound has a puff of air after it in words like pot, but not after /s/ in words like spot. Conversely, multiple phonologically distinct sounds may fall together in certain environments through a conditioned merger such as intervocalic flapping, whereby /t/ and /d/ sounds are pronounced as a “flap” sound between vowels, results in coding and coating becoming homophones.
Conditioned mergers pose a problem from a language-acquisition perspective. If a child encounters a sound that is ambiguous between two underlying forms, how are they to assign an underlying phonological representation to that sound? For example, how do children decide on underlying representations for the words latter and ladder, pronounced indistinguishably by most North Americans? The goal of my research was to investigate to how the ambiguity introduced by conditioned mergers may trigger reanalysis of the phonological system by looking at the interaction of the conditioned merger of intervocalic flapping and pre-flap vowel duration (which is allophonically greater before underlying /d/ than underlying /t/). By hypothesis, there vowels should be longer before written d than t only when disambiguating evidence for the underlying status of the flap is made available through analogy with unambiguous words (e.g. rider and writer have ride and write with unambiguous /d/ and /t/ in contrast with ladder and latter which do not have such analogous forms), and vowels before remaining ambiguous flaps were expected to pattern with vowels before unambiguous /t/-flaps.
To investigate these questions, I worked with data from the Philadelphia Neighborhood Corpus, a database of several hundred sociolinguistic interviews conducted with native Philadelphians between 1972 and 2012. In order to analyze the data, I wrote a number of scripts in the speech analysis software Praat and the statistical computing software R. Statistical models showed tendencies that were consistent with my hypothesis but did not meet levels of statistical significance.
Although I am likely to shift the focus of my research for the upcoming year to other related phonological phenomena, I plan on using the results of this project as a significant part of my senior honors thesis in linguistics. Furthermore, this project has afforded me the opportunity to significantly sharpen my programming proficiency in Praat and R as well as to expand my familiarity with statistical modeling, all of which skills will prove invaluable to me as I apply to PhD programs and continue to pursue linguistic research.