Alyssa Allen
Title: Bilingual Speaker Phonetic Alignment in a Human-Machine Context
Abstract: This talk presents the results of an experiment which explores the degree to which Spanish (L1)-English (L2) bilingual speakers in Mexico phonetically converge to a human compared to a machine (synthetic) voice in terms of voice onset time and vowel duration. I’ll also explore if the same convergence patterns hold in both Spanish and English.
People seem to have accepted these agents as active conversational partners and have no issues making claims about the voice assistant’s intelligence, personality, gender etc (Schweitzer et al., 2019; Zellou et al., 2021; Bell 2003; Branigan et al. 2011). Users have even been shown to view virtual assistants as conversational partners, secretaries, friends, or even family members (Choi & Drumwright, 2021, citing Purington et al., 2017; Rhee & Choi, 2020; Wang, 2020; Zhao & Rau, 2020).
While voice assistants like Siri and Alexa were first only available in standard-American-English, non-English versions of the agents and agents with non-standard American English dialects are now available to users (Garcia, 2018). Regardless of an agent’s, linguistic setting, voice agents have historically had lower performance in understanding non-English or “non-standard” English dialects, which raises concerns over voice assistant and the potential ramifications for the end user (Klein, 2021; Bogers, 2019; Debajyoti, 2019; Moussalli, 2019; Rowe, 2021). Under Communication Accommodation Theory (CAT) , social factors such as sentiment toward a conversational partner can impact degree of convergence (Giles, 1973).
This project explores the Spanish-English bilingual phonetic accommodation in a human-machine context via a lexical shadowing task. Conducted online, Mexico residents were asked to first say words as they saw them on screen. They then completed a shadowing task in which they were asked to say the word after it was said by either a human or machine model talker. Results and key takeaways per language will be discussed in this presentation.
Lindon Dedvukaj
Title: The dialectal split of Gheg and Tosk Albanian: A case of contact-induced phonological change
Abstract: I propose an alternative hypothesis for the origin of the two main phonological isoglosses that separate the Gheg and Tosk Albanian dialects, namely vowel denasalization and rhotacism of [n] > [ɾ] in Tosk Albanian. Some linguists have also considered denasalization in Tosk and Slavic Macedonian as being historically related (Curtis 2012: 155-56). Trummer (1981) and Hamp (1981/82) consider the denasalization in Albanian and Slavic Macedonian as one historic isogloss. Hamp (2015: 3-4) also sees rhotacism and denasalization as part of the same process and the loss of nasality as an areal sound law that is shared with Bulgarian and Macedonian. I endorse Trummer’s and Hamp’s view, adding to it the claim that the denasalization can be accounted for through the phonological process of segmentalization (Hock 2021: 118) or nasal vowel unpacking (Paradis and Prunet 2000). Three factors support this overall account. First, there is a lack of rhotacism in Slavic loanwords in Albanian. Bilingual Albanian-South Slavic speakers (assuming Albanian is the L1) would not rhotacize or change the phonology of a Slavic word when in constant communication with Slavic speakers who themselves do not rhotacize the alveolar nasal. Second, Tosk denasalization coincides historically with the period in which denasalization occurred in the Southwestern Macedonian dialects of the Bulgarian empire, e.g. Old Church Slavonic dǫbъ > Southwestern Macedonian dəmp ‘oak’, but further these underwent the same segmentalization process as Tosk; a nasal vowel segmentalizing into an oral vowel and nasal consonant, e.g. Old Church Slavonic orǫdьje > Tosk Albanian orendi ‘equipment’. And lastly, there is toponymic evidence of bilingual communities which exhibit both regular Slavic sound change and Tosk rhotacism, e.g. Lychnidus > Oh(ë)rid.