Leveraging Neural Language Models to Investigate Morphological and Semantic Processes During Natural Reading

The morpheme transposition paradigm has showed that stems are processed independently of their position within a word, whereas affixes are recognised only in their typical edge-aligned location (Spencer et al., 2023). However, this evidence has come from lexical decision tasks, leaving open the question of whether these positional constraints persist during natural, fluent reading. The present study addressed this gap through two eye-tracking experiments examining the morpheme transposition effect during sentence reading. Experiment 1 compared transposed compound words (e.g., berryblue from blueberry) and transposed suffixed words (e.g., anceavoid from avoidance) with their respective orthographic controls, embedded in predictable and unpredictable sentences. Results across both early and late eye-tracking measures revealed a robust morpheme transposition effect, such that transposed words were read faster than controls. Critically, this facilitation was significantly larger for compound words than for suffixed words, replicating the pattern of position-independent stem encoding and position-dependent affix encoding in an ecologically valid reading context. Sentence predictability produced main effects on several measures but did not interact with the morpheme transposition effect. Experiment 2 extended this investigation to the role of semantic transparency, comparing transposed semantically transparent compounds (e.g., cuptea from teacup) and semantically opaque compounds (e.g., linedead from deadline) embedded in predictable and unpredictable sentences. The interaction between morpheme transposition and semantic transparency emerged in gaze duration and refixation probability, such that the transposition effect was larger for transparent than opaque compounds. Again, predictability did not modulate these morphological effects. Together, these results provide the first evidence that the distinct positional encoding mechanisms of stems and affixes operate during naturalistic sentence reading and further reveal that semantic transparency plays a role in this process. The findings call for theoretical models of morphological processing to account for both positional constraints and semantic modulation under conditions of natural reading.Keywords: eye-tracking, morphological processing, morpheme transposition effect, semantic transparency, positional encoding