While prediction has previously been considered a strong primitive in Subject-Object-Verb (SOV) languages, it is relatively unknown how prediction might manifest in morphologically complex languages with verbs that potentially undergo morphological decomposition into their constituent verb stems and affixes during online sentence comprehension. In this reading study, we looked at how such top-down (prediction) and bottom-up processes (decomposition) interact in an agglutinating language like Malayalam. We investigate two questions simultaneously (i) whether predicting a suffix actually confers any processing advantage during lexical access in real-time, similar to predicting a verb stem, and (ii) whether this can reveal something about the representations of verb stems and suffixes of words within the mental lexicon. We find that when predicted correctly, suffixes pattern similarly to verb stems, which seems to suggest that suffixes can have independent representations alongside verb stems, thereby aiding visual word recognition by acting as access points to activate lexical entries.