Replication of Finn & Hudson Kam (2008) The curse of knowledge: First language knowledge impairs adult learners’ use of novel statistics for word segmentation, Exp. 3 Gina L. Iozzo, Katie E. Lamirato, and Joshua K. Hartshorne Department of Psychology, Boston College Contact joshua.hartshorne@bc.edu We replicated Exp. 3 of Finn & Hudson Kam (2008) The curse of knowledge: First language knowledge impairs adult learners’ use of novel statistics for word segmentation, Cognition, 108, 477-499. This experiment was part of a larger project to systematically replicate as many experiments involving statistical word learning in adults as possible. Keywords: statistical learning, unsupervised learning, language acquisition, replication In this paper, we report a replication of Experiment 3 from Finn & Hudson Kam (2008), henceforth FHK08. This exper- iment was part of a larger project to systematically replicate as many experiments involving statistical word learning in adults as possible. Method Subjects 50 subjects were recruited through Amazon Mechanical Turk and paid for their participation. We excluded 11 sub- jects who incorrectly answered more than one catch trial, leaving 39 (Ages 23-61, M=33). All subjects were native speakers of English and reported normal hearing. Known differences with original. • Our study had nearly double the number of subjects (20 vs. 39) and thus considerably more power and pre- cision. • Subjects in our study were likely older than the under- graduate subjects in the original. • Subjects in our study were compensated monetarily rather than with course credit. Materials Training. The stimuli cannot be easily recreated: The description in the original paper uses an obsolete phonetic alphabet for which we were unable to locate a key. The au- thors graciously provided us with the original stimuli. These stimuli are now available as part of the repository associ- ated with the present project, with permission of the authors (osf.io/bwa48/). Stimuli consisted of eight two syllable words (CCVCV), each beginning with a consonant cluster. The CC onsets vi- olate the word-initial phonotactic rules of English. FHK08 report that the stimuli were generated with the text-to-speech program SoftVoice. The synthesizer produced syllables with a monotonic F0 (fundamental frequency) of 83.62 Hz. All vowels were matched for length, and there were no co- articulation effects. During training, words were presented quasi-randomly with no pauses and no immediate repetitions. FHK08 report transitional probabilities of 1.0 for syllable transitions that are word internal and .143 at word boundaries. They further report that phoneme transitional probabilities (PTPs) within words were higher than those across word boundaries, with word-internal PTPs, ranging from .25 to 1.0 and PTPs across the word boundaries ranging from .035 to .143. Thus, there was no overlap in the within-word and across-word PTPs. Training lasted 17 min 59 sec. While FHK08 report that each word occurred 560 times, we suspect this is a type-o, since it would suggest 8.3 syllables/sec. The speech rate in the training files provided is much closer to 1 syllable/sec, which would be consistent with each word occuring 70 times for a total of 560 words in the training. However, we did not confirm this with an exact count of the words. Test. After exposure, participants were given a forced- choice test between a trained word and either a non-word (consisting of two syllables from the language but a transi- tional probability of 0) and a split-word (a trained word mi- nus the first consonant and with another trained word’s initial consonant at the end, resulting in a CVCVC structure). In essence, the researchers constructed the split-cluster words by shifting one phoneme to the right in the exposure stim- uli. This results in a foil that is a licit English word, pitted against a target which is not. The split-word foils measured the accuracy of parsing the consonant clusters in defiance of