The morphological priming paradigm

In a typical morphological priming experiment, participants perform lexical decisions on target words such as walk, with their lexical decision times being measured. The target words are preceded by either a morphologically complex prime word such as walked, which is based on the same stem as the target word, or by a control prime which is morphologically unrelated to the target, such as brush. If the processing of a complex word such as walked involves the decomposition of the word into its morphological constituents, this should lead to the activation of the stem walk during processing of the prime word. This should facilitate subsequent recognition of the target word walk, with faster lexical decision times relative to the unrelated control condition, because for a prime-target pair such as walked-walk, the stem walk is already activated when the target word walk is encountered, while this is not the case for a prime-target pair such as brush-walk.

In order to conceal the true purpose of the experiment, morphological priming experiments typically contain a substantial number of filler trials. This is particularly important because L2 learners often possess considerably more elaborate explicit knowledge about the morphological properties of the language than native speakers, which might lead to strategic effects. In addition, in order to make the lexical-decision task meaningful, it is common to add additional filler trials in which the target word is a nonword (typically so that 50% of all trials contain nonword targets).

With regard to the dependent measure, morphological priming experiments are by no means limited to lexical decision times, but can also rely on measures such as naming latencies (e.g., Coughlin & Tremblay, Reference Coughlin and Tremblay2014), EEG (see Bosch & Leminen, to appear) or fMRI (e.g., De Grauwe, Lemhöfer, Willems & Schriefers, Reference De Grauwe, Lemhöfer, Willems and Schriefers2014). Primes and targets can be presented either visually or auditorily, making the paradigm suitable for research on both written and spoken word recognition (see Gor & Cook, Reference Gor and Cook2010, and Gor & Jackson, Reference Gor and Jackson2013, for two examples of L2 studies investigating spoken word recognition). Also, while the majority of studies have focussed on late L2 learners, the paradigm has also been employed to investigate other groups of bilinguals, such as heritage speakers (Gor & Cook, Reference Gor and Cook2010; Jacob & Kırkıcı, Reference Jacob and Kırkıcı2016).

A fundamental problem in morphological processing research is to tease apart the role of the morphological, semantic, and orthographic information contained in a word. For morphological priming, prime-target pairs such as walked-walk are not only morphologically related (in the sense that both words are based on the same stem), but are also similar with regard to other, non-morphological properties. For example, walked and walk are orthographically and phonologically similar, and are also related in meaning. As a result, a priming effect for a prime-target pair such as walked-walk is not necessarily morphological in nature, but could also be due to other, non-morphological similarities between prime and target, such as semantic relatedness or surface-form similarity. Thus, in order to be able to make the claim that priming effects are really caused by prime and target sharing the same stem, the experimental design has to include possibilities to rule out these alternative, non-morphological explanations for a priming effect. To achieve this, morphological priming experiments typically contain control conditions with prime-target pairs which are only semantically related (such as battle-fight) or only surface-form related (such as scandal-scan). Additionally, a substantial number of studies have also attempted to deal with the issue by adding an identity control condition, in which a target word such as walk occurs as both prime and target. The rationale behind this is that if a priming effect for a pair such as walked-walk is indeed caused by the fact that walked and walk are based on the same stem, this priming effect should be as strong as the priming effect for the identity pair walk-walk. The identity condition thus allows for a distinction between full priming (i.e., a priming effect which is as strong as the one observed for the identity control condition) and partial priming (i.e., a priming effect which is weaker than the one for the identity control condition), with full priming being considered particularly robust evidence that a priming effect is genuinely morphological in nature (Stanners, Neiser, Hernon & Hall, Reference Stanners, Neiser, Hernon and Hall1979, and much subsequent work). Yet another possibility to control for the influence of non-morphological factors is based on comparisons between different groups of morphologically related prime-target pairs. For example, a number of studies (e.g., Neubauer & Clahsen, Reference Neubauer and Clahsen2009; Feldman et al., Reference Feldman, Kostić, Basnight-Brown, Filipović Đurđević and Pastizzo2010) have compared morphological priming effects for regular (e.g., walked-walk) and irregular (e.g., threw-throw) forms. Regular forms such as walked differ from irregular forms such as threw with regard to whether or not the stem can be identified by stripping of an affix, while the semantic and orthographic similarities between prime and target can be kept similar for both groups of forms. Thus, a difference in priming effects can be potentially informative with regard to issues related to morphological decomposition. Finally, it is possible to employ specific experimental procedures which have been shown to be less affected by particular non-morphological factors. In a cross-modal priming experiment, for instance, prime and target words are presented in different modalities, typically with the prime word being presented auditorily and the target word visually. This method is considered to reduce effects of surface-form similarity. However, participants are consciously aware of the prime word and thus have full access to its semantic and conceptual properties, which typically causes robust semantic priming effects. In a masked-priming experiment, in contrast, prime words are presented on screen for only an extremely short amount of time (typically for an SOA between 30 and 70 ms, which can also be experimentally manipulated), and are also preceded by a visual mask (typically a number of hash marks). As a result, participants are usually not consciously aware of the prime words. This type of presentation has been shown to considerably reduce semantic priming effects, typically with either no or only very small facilitation for prime-target pairs such as battle-fight in the semantic control conditions (e.g., Rastle, Davis, Marslen-Wilson & Tyler, Reference Rastle, Davis, Marslen-Wilson and Tyler2000; Marslen-Wilson, Reference Marslen-Wilson and Gaskel2007). Finally, in a long-lag or delayed priming experiment, primes and targets are presented overtly and in the same modality, but are separated from each other, for example with a number of (typically morphologically simplex) filler words presented in between. This typically reduces semantic and form-based priming effects, but preserves priming effects for morphologically related words (e.g., Raveh & Rueckl, Reference Raveh and Rueckl2000; Rueckl & Aicher, Reference Rueckl and Aicher2008; De Grauwe et al., Reference De Grauwe, Lemhöfer, Willems and Schriefers2014).

The results from a particular morphological priming experiment can differ substantially depending on the exact experimental procedure employed in a study. For instance, an especially striking effect which is largely specific to masked priming comes from studies which have investigated priming effects for so-called pseudo-morphologically related prime-target pairs, such as corner-corn. While corner and corn are morphologically unrelated, corner superficially looks as if it consisted of the existing English stem corn and the existing English affix –er. In masked priming, but typically not in cross-modal priming (e.g., Longtin, Segui & Hallé, Reference Longtin, Segui and Hallé2003; Meunier & Longtin, Reference Meunier and Longtin2007) or long-lag priming (e.g., Rueckl & Aicher, Reference Rueckl and Aicher2008), prime-target pairs such as corner-corn yielded robust priming effects (e.g., Rastle, Davis & New, Reference Rastle, Davis and New2004, Marslen-Wilson, Bozic & Randall, Reference Marslen-Wilson, Bozic and Randall2008). Rastle and colleagues consider this so-called corner-corn effect as evidence for an early and semantically blind morpho-orthographic segmentation mechanism, which, at a pre-lexical level of processing, decomposes a word into stems and affixes. With regard to the difference between masked and cross-modal priming, Marslen-Wilson (Reference Marslen-Wilson and Gaskel2007) argues that the different priming paradigms address distinct levels of processing, with masked priming tapping into an early, automatic, and pre-lexical level and overt priming addressing a later, central-lexical level. While the idea of an early morpho-orthographic decomposition mechanism for complex words in L1 speakers is not entirely uncontroversial (e.g., Baayen, Reference Baayen, Jarema and Libben2007; Stockall & Marantz, Reference Stockall and Marantz2006), it is important to note that it is at least difficult to directly compare the results from masked and cross-modal priming experiments, even when the respective studies are otherwise (i.e., with regard to experimental materials and participants) very similar.

Morphological priming in bilinguals

In studies investigating morphological processing in second-language (L2) learners, the same priming experiment is typically conducted with a group of L2 speakers and a control group of native (L1) speakers of the language. Priming effects are then determined separately for each of the two groups, and subsequently compared with each other. This procedure avoids any direct comparisons of lexical decision times between the L1 and L2 groups, which is useful given that L2 speakers often show generally slower lexical decision times overall. However, note that slower overall reaction times in L2 speakers can in principle also have an influence on the size of a priming effect. For example, it is possible that priming effects in an L1 group are relatively small because the group already shows very fast lexical decision times overall, so the possibility to speed up even more when primed by a morphologically related prime might be limited compared with an overall slower L1 group. A way to control for this is to conduct a z-transformation of the lexical decision times separately for each group prior to comparing L1 and L2 priming effects, which sets any group differences in overall lexical decision times to zero.

With regard to the experimental materials used in such a study, the fact that the items have to be suitable for L2 speakers typically has profound consequences for item selection. If materials have originally been designed for a study on L1 processing, and are then used to replicate the study with L2 speakers, it is often the case that a large number of data points have to be excluded due to incorrect lexical decisions or extremely slow lexical decision times. The fact that the materials have to be appropriate not only for L1, but also for L2 speakers also restricts the possibility to compare the results from the L1 control group with other studies on morphological processing in L1, which do not face such additional restrictions during item selection. As a result, comparisons between the results from the L1 control group in an L2 processing study and those from an experiment designed to investigate L1 processing is not without problems.

Previous morphological priming studies comparing L1 versus L2 speakers have yielded detailed insight into similarities and differences between L1 and L2 morphological processing. For example, a number of masked priming studies (e.g., Silva & Clahsen, Reference Silva and Clahsen2008; Neubauer & Clahsen, Reference Neubauer and Clahsen2009; Clahsen & Neubauer, Reference Clahsen and Neubauer2010) have found substantial L1-L2 differences in morphological priming effects, with robust priming in L1, but either no or considerably reduced priming in L2 speakers. Note that the L2 groups in these studies nevertheless showed significant identity priming, suggesting that they were in principle sensitive to the masked prime words. Jacob, Fleischhauer, and Clahsen (Reference Jacob, Fleischhauer and Clahsen2013) found a similar L1-L2 difference in a cross-modal priming study investigating the processing of regular German past participles, with full priming for L1 speakers, but only partial priming in the L2 group. This difference turned out to be specific to regular participles, and did not emerge for otherwise similar irregular forms. The pattern of results observed in these studies gave rise to the idea that the processing of complex words in L2 speakers relies less on morphological decomposition and instead more on storage and retrieval of full word-form representations in the mental lexicon. Interestingly, other L2 masked-priming studies (Feldman, Kostić, Basnight-Brown, Filipović Đurđević & Pastizzo, Reference Feldman, Kostić, Basnight-Brown, Filipović Đurđević and Pastizzo2010; Diependaele, Duñabeitia, Morris & Keuleers, Reference Diependaele, Duñabeitia, Morris and Keuleers2011; Coughlin & Tremblay, Reference Coughlin and Tremblay2014; Reference Coughlin and Tremblay2014; Foote, Reference Foote2015), as well as L2 studies in which the primes were presented overtly (Basnight-Brown, Chen, Hua, Kostić & Feldman, Reference Basnight-Brown, Chen, Hua, Kostić and Feldman2007; De Grauwe et al., Reference De Grauwe, Lemhöfer, Willems and Schriefers2014), have instead found similar morphological priming effects in L1 and L2 speakers, and have interpreted this as evidence against fundamental L1-L2 differences in morphological processing. In order to account for this differential pattern of results, it has been suggested that L1-L2 differences might be restricted to particular morphological phenomena. For instance, Silva and Clahsen's (Reference Silva and Clahsen2008) masked-priming study determined morphological priming effects for both derived forms (with prime-target pairs such as darkness-dark) and inflected forms (with pairs such as walked-walk) in L1 speakers and two groups of (native Chinese and German) L2 speakers of English. For derived forms, only a gradual difference emerged, with a full priming effect for the L1 group and partial priming for the two L2 groups. For inflected forms, in contrast, while the L1 group displayed a full priming effect, both L2 groups showed no priming effects at all. Kırkıcı and Clahsen (Reference Kırkıcı and Clahsen2013) found a similar pattern of results in a masked-priming study on the processing of Turkish derived and inflected forms in L1 and L2 speakers. Again, the L2 group showed similar morphological priming effects as the L1 group for derivation, but no priming for inflection. Jacob, Heyer, and Veríssimo (Reference Jacob, Heyer and Veríssimo2017) directly compared masked-priming effects for German derived vs. inflected forms in a study in which the same target word was primed by either a derived nominalization or a past participle. Again, while L1 speakers showed similar priming effects for both phenomena, the L2 group showed a difference, with significant priming effects only for derivation, but not for inflection. At least some of the studies reporting similar morphological priming effects for L1 and L2 speakers (Diependaele et al., Reference Diependaele, Duñabeitia, Morris and Keuleers2011; De Grauwe et al., Reference De Grauwe, Lemhöfer, Willems and Schriefers2014) have investigated the processing of derived forms. Note, however, that several other masked-priming studies (Feldman et al., Reference Feldman, Kostić, Basnight-Brown, Filipović Đurđević and Pastizzo2010; Coughlin & Tremblay, Reference Coughlin and Tremblay2014, Voga, Anastassiadis-Symeonidis & Giraudo, Reference Voga, Anastassiadis-Symeonidis and Giraudo2014) have found significant morphological priming effects in L2 speakers even for inflected forms. Also, at least one masked-priming study (Clahsen & Neubauer, Reference Clahsen and Neubauer2010) reports substantial L1-L2 differences for derived forms.

Another possibility is that L1 and L2 speakers might show similar priming effects for morphologically complex words, but for different reasons. Recall that prime-target pairs such as walked-walk in a morphological priming experiment are not only morphologically related, but are necessarily also related with regard to surface form. It is at least conceivable that L2 speakers, during word recognition, might focus relatively more on surface form properties of a word, such as its orthography. As a result, even if L2 speakers do not decompose complex words, they might still show a priming effect for prime-target pairs such as walked-walk, but not because walked and walk are both based on the same stem, but simply because they share a number of letters. Feldman et al. (Reference Feldman, Kostić, Basnight-Brown, Filipović Đurđević and Pastizzo2010) initially considered this possiblility a serious problem, and argued that morphological priming paradigms in which primes and targets occur in the same modality, such as masked priming, might be unsuitable for L2 processing research. However, as discussed above, morphological and form-based priming effects can be disentangled through appropriate control conditions. A good example of this is Heyer & Clahsen's (Reference Heyer and Clahsen2015) masked-priming study, which compared morphological versus orthographic priming effects in native speakers and L2 learners of English. For morphologically related prime-target pairs such as darkness-dark, both subject groups showed significant priming effects. However, the L2 group, but not the L1 group, also showed significant priming effects for purely orthographically related prime-target pairs which were morphologically and semantically unrelated, such as scandal-scan. Indeed, L2-specific orthographic priming effects also emerged in at least some of the L2 processing studies which have reported similar morphological priming effects in L1 and L2 (Feldman et al., Reference Feldman, Kostić, Basnight-Brown, Filipović Đurđević and Pastizzo2010; Diependaele et al., Reference Diependaele, Duñabeitia, Morris and Keuleers2011). Heyer & Clahsen (Reference Heyer and Clahsen2015) thus argue that, even though the L2 processing of complex words might indeed rely less on morphological decomposition and more on storage of full-word-form representations, L2 speakers might nevertheless still show priming effects for complex words. However, unlike for the L1 group, these priming effects are actually unrelated to morphological structure, and are instead orthographically mediated. Note, however, that the above-mentioned L2-specific difference between derived and inflected forms is difficult to explain along these lines, given that the degree of orthographic overlap between primes and targets was very similar for derived and inflected items in all three studies in which this difference emerged (Silva & Clahsen, Reference Silva and Clahsen2008; Kırkıcı & Clahsen, Reference Kırkıcı and Clahsen2013; Jacob et al., Reference Jacob, Heyer and Veríssimo2017). Also, at least some L2 processing studies have not found evidence for an L2-specific orthographic priming effect in their orthographic control conditions. For instance, in Jacob et al. (Reference Jacob, Heyer and Veríssimo2017), L2 speakers showed significant priming for derived nominalisations, but no such effect for morphologically unrelated items in which primes and targets shared the same number of letters as in the derived condition.