Decoding Language Discriminative Features In EEG Signals
This project presents an experimental study to probe the language discriminative cues in neural responses to speech stimuli presented from two languages. In this study using electroencephalography (EEG) recordings, one of the first ones in this direction, we explore the differences in human perception while listening to a familiar and an unfamiliar language. The main objective of this study is to analyze the changes in brain responses when a human subject is listening under familiar and foreign conditions in a linguistic sense. The stimuli-set for this study contains words from English (familiar language) and from Japanese (unfamiliar language). The language discriminative representations are probed by designing an off-line language classifier on the EEG signals. These experiments reveal that the time-frequency representations along with the phase information of EEG signal carry significant language discriminative information. We obtain an average accuracy of 64.4% among 12 subjects with a support vector machine based classifier, which is significantly above chance.
An EEG-based Study On the Evolution of Brain Representations during Language Learning
The early stages of language learning consists of learning new words from an unfamiliar language through phases of listening and articulating the words of the language over time. In this study, we design an experimental setup to understand the evolution of neural representations in a language learning task. The human subjects in the experiment are performing the task of listening and reproducing a set of words from a familiar language (English) and an unfamiliar language (Japanese). During the task, electroencephalogram (EEG) signals are recorded in addition to the spoken audio signals. With a detailed analysis on the recorded EEG signals and the audio signals, we propose to uncover the language learning process by highlighting the difference between these signals for the known and the unknown language. Upon exposure to multiple trials of the same word from the unfamiliar language, we find that the inter-trial distance between EEG signals (as well as the spoken audio signals) decreases, indicating a consistency in the neural representations. In addition, the brain regions (in terms of EEG channel locations) responsible for the learning task are identified.
Akshara Soman, C R Madhavan, Kinsuk Sarkar, and Sriram Ganapathy
An EEG Study On The Brain Representations in Language Learning
Submitted to Brain and Language, June 2018.
Office: LEAP Lab (C328) / aksharas AT iisc DOT ac DOT in / 080-2293-2362
Postal: C328 (LEAP Lab) / Department of Electrical Engineering / IISc Bangalore / India 560012