Many studies on memory retrieval in language processing haveidentified similarity-based interference as a key determinant ofcomprehension. The broad consensus is that similarity-basedinterference reflects erroneous retrieval of a non-target itemthat matches some of the retrieval cues. However, themechanisms responsible for such effects remain debated.Activation-based models of retrieval (e.g., Lewis & Vasishth,2005) claim that any differences in processing difficulty due tointerference in standard RT measures and judgments reflectdifferences in the speed of retrieval (i.e., the amount of time ittakes to retrieve a memory item). But this claim is inconsistentwith empirical data showing that retrieval time is constant dueto the use of a direct-access procedure (e.g., McElree, 2000,2006). According to direct-access accounts, differences injudgments or RTs due to interference arise from differences inthe quality or availability of the candidate memoryrepresentations, rather than differences in retrieval speed. Toadjudicate between these accounts, we employed a novelmethodology that combined a high-powered (N = 200) two-alternative forced-choice study on interference effects withdrift diffusion modeling to disassociate the effects of retrievalspeed and representation quality. Results showed that thepresence of a distractor that matched some of the retrieval cueslowered asymptotic accuracy, reflecting an effect ofrepresentation quality, but did not affect retrieval speed,consistent with a direct-access procedure. These results suggestthat the differences observed in RTs and judgment studiesreflect differences in the ease of integrating the retrieved itemback into the current processing stream, rather than differencesin retrieval speed.

This study investigates interference effects in sentenceprocessing. A parade case involves agreement attraction,where the processing of a number mismatch between a verband its subject is eased by a number-matching lure (*Thekeytarget to the cabinetslure were rusty), relative to sentenceswhere neither noun matches the verb (*The key to the cabinetwere rusty). Existing accounts claim that this effect reflectserror-prone retrieval or misrepresentation of the target.Recently, a third account has been proposed which claims thatthe contrast between the two configurations reflects increaseddifficulty in the second sentence due to feature overwriting inthe encoding (both nouns are singular). We provide resultsfrom two self-paced reading experiments that isolate theeffects of feature overwriting and attraction by manipulatingthe presence of an agreement cue. Results showed a largerdifference within the configurations with a cue, which suggestthat attraction cannot be reduced to feature overwriting.