Sociolinguistic work on variable ING (thinking~thinkin’) uses conversational speech as a window onto the locus of ING variation, and on the use of the variants -ing and -in’ in speech[1,2]. We use experimental priming to complement this literature by asking how the variants of ING are mentally represented (e.g. are -ing and -in’ represented as distinct pieces in the mental lexicon), and how they impact spoken word recognition.
261 participants completed two continuous auditory lexical decision experiments. The critical stimuli were disyllabic progressive verbs (frequency controlled). The critical conditions asked whether 'working' and 'workin’' prime 'thinking' and 'thinkin’' by comparing their response-times (RTs) to a baseline (e.g. jiggle–thinking). Participants heard 120 critical stimuli using a 400-600ms ISI, and 396 fillers (50% nonwords), and responded whether they heard a real word of spoken American English or a nonword. RTs measured from sound-file onset to response were analysed using LMERs. Fixed effects included PrimeCondition, TrialNumber, Prime/Target Frequency, and PrimeRT. Random intercepts for Participant and Target were included (random slopes did not improve the model).
The Exp1 results show significant priming for all critical conditions. The finding that 'working' primes 'thinking' (Priming Effect=44ms, 𝛽=-0.05, p<0.001) and 'workin’' primes 'thinkin’' (84ms, 𝛽=-0.09, p<0.001) suggests that -ing and -in’ are both mentally represented as pieces distinct from the whole words they occur in, since they elicit priming even when the prime and target stems are unrelated (i.e. work vs think). We further detect a processing asymmetry: -ing and -in’ primes yield equal facilitation for -ing targets (-ing=44ms vs -in’=36ms: 𝛽=0.008, p=0.55), but same-variant (-in’) primes facilitate -in’ targets better than cross-variant primes (-ing=51ms vs -in’=84ms: 𝛽=-0.03, p<0.001). We call this the -in’ boost.
An explanation is that the unexpectedness of noncanonical variants causes a processing speedup[3]. This boost should be short-lived[4]. Exp2 uses the same method as Exp1, but introduces an intervening word (e.g. blue) between half the primes and targets, with the goal of a) replicating the -in’ boost in the no-intervener condition, and b) testing whether the asymmetry is attenuated in the 1-intervener condition.
The no-intervener condition replicates the -in’ boost. Crucially, the 1-intervener condition eliminates the asymmetry and gives equal priming from both prime types to both target types (-ing targets: -ing=29ms, -in’=33ms; 𝛽=-0.004, p=0.06. -in’ targets: -ing=23ms, -in’=31ms; 𝛽=-0.008, p=0.18). This result supports the idea that shared surface form between sequential -in’ primes and targets is violating listeners’ expectations causing an -in’ boost (if this were exclusively a shared surface form effect we would expect a parallel ing-ing boost).
This constitutes evidence that listener expectations can influence the processing of variation. Upcoming studies manipulating variant proportions and model talker accent will probe this idea further.