A generative incremental dependency parser enhanced with sequence encodings from a large languagemodel was used to calculate a syntactic surprisal measure in order to analyze the ambiguity present when listening to an audiobook. This metric was correlated with BOLD fMRI signal, confirming the hypoth- esis that derivations with low predicted probability require greater effort to understand. This surprisal metric was validated at various levels of parallel processing, providing evidence that increasing the level of parallelism creates a significantly better predictor for the data, up to a threshold which includes most viable derivations. The brain regions associated with increased activation for higher levels of parallelism across English and Chinese were identified as bilateral superior temporal gyrus activations.