Asymmetries in the stem and suffix masked priming response: a large-scale online study

The research on morphological decomposition of the past 20 years has capitalized on the visual masked priming response elicited in pairs of words sharing the same root/stem (“stem priming”: driver-DRIVE), which has been shown to be robust across languages (a.o., English, French, Italian, Spanish), regardless of concatenativity of word formation (e.g., Arabic, Hebrew). On the other hand, the masked priming response to words sharing the same affix (“prefix priming”: disembark-DISPROVE; “suffix priming”: lovable-TAXABLE) has been reported not as consistently. This asymmetry seems to support a model of lexical access in which affixes are initially stripped in a first access stage and morphologically complex words are initially accessed via their stems (“prefix-stripping model of decomposition”: Forster & Davis, 1975). Here we point out two potential confounds at play. First, affix masked priming is not directly comparable to stem masked priming: the former cannot be elicited on its own (since affixes are bound morphemes by definition), whereas the latter can, but only in “word-based” languages such as English, in which bare words may surface as phonologically overlapping with the underlying stem. We believe this property of English (and similar languages) is rather idiosyncratic and not very common cross-linguistically; and, more importantly, may hinder the direct comparison between the affix and stem masked priming responses, and as a consequence complicate the detection of potential differences (or lack thereof). Second, previous studies were relatively underpowered and therefore unable to detect medium-to-small effects (<=15 ms). To tackle the first confound, this study elicited the online stem, prefix, and suffix priming response to comparable word pairs (i.e., all involving bimorphemic words), while taking into account the unavoidable and uncontrollable properties of each morpheme type. In experiment 1, we elicited the priming response to prefixes and suffixes (retouch-RESALE, jogger-PLANNER) and stems of prefixed and suffixed words (disuse-MISUSE; lovable-LOVELESS); and to identical (scorpion-SCORPION), orthographically-related (advertise-ADVENTURE), and semantically-related words (particle-ELECTRON). To control for potential confounds, in experiment 2 we elicited the stem priming response from prefixed (skillful-SKILL) and suffixed words (unleash-LEASH), along with the same identity, semantic and orthographic conditions defined above. To tackle the second confound, we ran a series of power simulations, which suggested that a sample size of 6,000 subjects (each experiment) would ensure 80% power for effects equal to or larger than 5 ms. Our results show a gradient in the masked priming response, in which the (a) identity priming had the biggest effects, followed by (b) the stem priming, and then (c) by the affix priming, with (b) and (c) significantly differing from one another. While these results support an affix-general stripping model of decomposition, they also show that affix priming is statistically similar to orthographic priming, thus suggsting that the affix priming response may not be a true morphological effect after all.