In natural dialogue, speakers show a tendency to mimic each other's utterances both in terms of  the lexical items used (cf., lexical priming) and of the particular syntactic structures used to construct their sentences (cf., syntactic priming). These effects are widely documented both in corpus and in experimental psycholinguistic studies. Although the reliability of these phenomena is already robustly established, there is still much discussion as to the mechanisms from which they arise. Recently, some authors have argued that structural resonance is but a side-effect from lexical resonance, which arises from the natural topic coherence in dialogue (e.g., Healey et al., 2010a,b; Pietsch et al., 2012). These authors claim that, when lexical factors are discounted, one actually observes structural divergence, that is, contiguous utterances from different speakers would tend to differ more than one would expect by mere chance. Beyond the possible confound between structural and lexical factors, there is debate as to whether resonance phenomena arise from either sharing of abstract syntactic structures (e.g., Reitter & Keller, 2007) or just from learning ordered chunks of lexical sequences (e.g., Chang, 2006). A final question concerns the duration of the resonance effects, which is crucial for the models that may account for the effects. Some authors have argued that the phenomenon is short-lived, lasting at most a few sentences (e.g., Branigan et al., 1999), while others have argued for longer lasting effects (e.g., Bock & Grif?n, 2000, Gries, 2006). I will present studies on English, German, and Japanese corpora that provide new insights these questions, and correct the methodological problems that 

are found in previous corpus study. I will show that corpus information does indeed provide crucial insights into the psychological mechanisms that are involved in language processing (cf., Gries, 2006)