Neural Tracking of Rhythmically Presented Morpheme Structure
Tovah Irwin1, Alec Marantz2,1; 1New York University Abu Dhabi, 2New York University
[INTRODUCTION] Understanding speech requires the brain to parse a continuous signal into a hierarchical structure of discrete units of meetings. Morphemes are the smallest meaningful lexical units of language. The word “millisecond” is composed of two morphemes - the stem “second”, and prefix, “milli-”. The extent to which morphemes are represented as structural units in the brain during language processing is debated. Hypothesis 1 is that words are first recognized as whole units. Hypothesis 2 is that the lexicon is organized around morphemes and that words are morphologically decomposed during language processing. Previous studies reveal that the brain shows frequency peaks in neural oscillations associated with the presentation rate at various levels of language structure, including syllables/words, and phrases (Ding et al. 2016). If morphological structure is automatically constructed in the brain along with words and phrases during speech recognition (Hypothesis 2), we expect also to see frequency peaks associated with morpheme presentation rates. [METHODS] We recorded the brain activity of 11 adults using Magnetoencephalography (MEG). Participants listened to lists of unrelated words, blocked by condition. Each condition contained 200 syllables, with two runs of each condition per participant. Syllables were presented at a rate of 3 Hz, with all audio generated syllable by syllable by a text to speech program to avoid including acoustic information on the locations of word or morpheme boundaries. Conditions included (a) single syllable single morpheme words such as “cat”, (b) two syllable single morpheme words such as “money”, (c) two syllable two morpheme words such as “artful”, and (d) four syllable two morpheme words such as “millisecond”. [RESULTS] The statistical analysis of significant peaks in the responses was performed by averaging power spectra from each run across subjects, and significance was tested based on the average of every 0.5 hz bins from 0.25-5 Hz following Ding et al. (2016) . The single syllable/word condition (a) evoked a strong 3 Hz peak associated with the word and syllable rate of presentation (p = 0.26), but no other lower frequency peaks indicative of a higher level structure. Both the two and one morpheme two syllable word conditions (b-c) revealed both a 3 Hz peak associated with syllable rate (p = 0.0361 and 0.0317) and a 1.5 Hz peak associated with word boundaries (p=0.1793 and 0.1780). The 3 morpheme 4 syllable condition (d) produced a 3 Hz syllable peak (p=0.0369), as well as a 1.5 Hz morpheme-rate peak (p=0.2893), and a 0.75 Hz word-rate peak (p=0.0011). Data collection and analysis is ongoing, and future analysis will also reveal the source localization of frequency peaks. [SUMMARY] The results support Hypothesis 2, which predicts the word, morpheme and syllable peaks for condition (d) “millisecond”, over Hypothesis 1, which does not predict the 1.5Hz morpheme peak in this condition. The experiment indicates that words are decomposed into a morphemic structure from the speech stream similarly to other layers of language structure.
Topic Areas: Morphology, Perception: Auditory