The human process of spoken language recognition utilizes both lower-level acoustic-phonetic information and higher-level linguistic information. The present paper describes two experiments in which the gating paradigm was used to investigate the interaction and the trade-off between these sources of information in word identification and sentence recognition. Experiment I classifies the sentences into five broad categories according to their linguistic contents, and demonstrates the effects of these categorial differences upon word identification in sentences. By introducing the concept of commonness of the event described by a sentence to express the amount of semantic-pragmatic information, Experiment II shows the quantitative trade-off relationship between the acoustic-phonetic information and the semantic-pragmatic information in sentence recognition.