In a set of behavioral, eye tracking, and ERP experiments, this thesis explored when and how object-state representation is established, maintained, and retrieved in language comprehension. We firstly examined whether different object-state representations could be established under two contrasting linguistic contexts (e.g., no change – “choose the ice cream” vs. change – “drop the ice cream”). Our findings showed that when linguistic context was provided, the representation that matched the consequences of described events was verified faster than the one that mismatched the expected outcome. Then, we studied the time course of establishing object-state representations with the visual world paradigm. Our results suggested that: a) the difference in looks towards the depicted versions of the situationally appropriate target object (an intact vs. a dropped ice cream) often manifested at the reference to the object but not prior to it; b) eye movements were primarily driven by semantic overlap between the visual display and the described object-state representation. Moreover, we found ERP evidence that was consistent with the need to keep track of, and retrieve object-state representations from episodic memory. We conclude that object-state representations were activated and retrieved during language processing. The work reported in this thesis highlights the need to take account of dynamics of event representation to capture the interplay between general semantic knowledge about objects and the episodic knowledge introduced by the sentential context in language comprehension.