Although theories of sentence processing from pre-2008 assume a noise-free channel, a robust theory of sentence comprehension must consider the potential influence of imperfect linguistic input (Gibson, Bergen, and Piantadosi, 2013; Levy, 2008; Levy, Bicknell, Slattery, & Rayner, 2008; Levy, 2011). This study will explore how context affects comprehenders’ interpretation of communication in the presence of noise. In this investigation, we will examine the rate at which participants interpret sentences literally, which are made implausible by insertion and/or deletion of words, when those sentences are preceded by two error-free sentences that establish either supportive or non-supportive contexts, or when the sentences are stand-alone, with no preceding context. This study is an extension of previous studies (pre-registration available at osf.io/n3kgr for Study 1 and osf.io/zn579 for Study 2), each of which had 4 between-subject conditions, where subjects were presented with 1) sentences under Active/Passive construction preceded by non-supportive context, 2) sentences under Active/Passive construction preceded by supportive context, 3) sentences under double object (DO)/prepositional object (PO) construction preceded by non-supportive context, and 4) sentences under DO/PO construction preceded by supportive context, respectively. 60 participants were recruited for each condition in Study 1. The results were as predicted numerically but some of the results were not statistically significant, which suggested that the study might have been under-powered. Study 2 was identical to the Study 1 except that we doubled the number of participants (120 for each condition), and the results were as predicted both numerically and statistically significant. Another study (Study 3, available at osf.io/murz9) was also carried out with 6 between-subject condition: apart from the 4 aforementioned ones, subjects were also presented with 5) sentences under Active/Passive construction preceded by no context, and 6) sentences under DO/PO construction preceded by no context. 120 participants were recruited in each of the 6 conditions. The results were concordant with our theory of sentence processing both numerically and statistically. However, some of the predictions in the pre-registration for Studies 1, 2, and 3 regarding the active/passive constructions were incorrectly presented, due to an oversight error. In particular, we wrote that we were expecting 1) a lower literal interpretation rate for Active/Passive sentences in the supportive context condition compared with those in the non-supportive context condition, and 2) no difference in literal interpretation rate between Active sentences and Passive sentences in both supportive and non-supportive context conditions. Neither of these predictions was actually appropriate according to what we have learned from previous studies about the noisy-channel theory as it applies to these examples. First, it has now been well established that there are hardly any inferences in the implausible Active/Passive materials (Gibson et al., 2013; Poppels & Levy, 2016), probably because of the low probability of the edits (2 insertions / deletions according to Gibson et al., 2013; or exchanges across a main verb according to Poppels & Levy, 2016). Thus, we don’t expect much, if any, of an effect of context on the active-passive structures, because of the low probability of inferences in these cases. Furthermore, Studies 1 and 2 failed to find an effect of context; hence we don’t expect an effect of context in the current study for the active-passive constructions. Second, some studies have found a small effect of more inference for the implausible passive construction relative to the implausible active condition (e.g., Poppels & Levy, 2016). This is plausibly because of a structural prior: passives are much rarer than active structures, so people may make more inferences in the lower frequency structure (toward the higher frequency structure). We may replicate the Poppels & Levy observation of implausible passives having a higher inference rate than implausible actives. In this study (Study 4), we pre-register our correctly-written predictions and re-run the experiments.