No priming for Maltese word patterns in spoken word recognition

Adam Ussishkin (The University of Arizona)

Luke Galea (University of Malta)

Andrew Wedel (The University of Arizona)

Samantha Wray (Dartmouth College)

Abstract

As a Semitic language, Maltese exhibits root-and-pattern morphology, thus raising the question of whether roots and patterns provide cues that native speakers can exploit during word recognition. In this chapter, we focus on whether the pattern provides native speakers with cues that can facilitate spoken word recognition as has been previously found for visual word recognition (Galea 2011). We present a priming experiment designed to test whether the vocalic melody, prosodic CV structure, and the combination of these elements facilitates response times to auditorily presented target words, using the same auditory masked priming technique previously used for Maltese by Ussishkin et al. (2015). We find no such facilitation, thereby supporting models in which the pattern is not decomposed in the earliest stages of lexical access.

10.5281/zenodo.16933153

1 Introduction

An important debate in modeling the architecture and organization of the mental lexicon concerns the extent to which word-internal structure might play a role in word identification and recognition. Within psycholinguistics, most evidence supports those models of lexical access which incorporate morphological structure in contrast to models which posit only whole-word storage for the mental lexicon.

The majority of work which aims to elucidate the role of morphology in word recognition has focused on its relevance to the visual domain, and auditory word recognition has been less studied. Models stemming from these studies which incorporate a decompositional component that recognizes a role for constituent morphemes in a word include those proposed by Meunier and Segui (1999), Taft, Hambly and Kinoshita (1986), Wurm (2000), and Winther Balling and Baayen (2008). Those models which are characterized by a whole-word storage and non-decompositional access include Shortlist B (Norris and McQueen 2008) and the Discriminative Lexicon (Baayen, Chuang, Shafaei-Bajestan, and Blevins 2019), among others. However, these studies focus on languages with largely linear morphology such as English and Dutch. Multimorphemic words in these languages consist of contiguous morphemes linearly ordered, for example ’-ed’ is suffixed to regular English verbs to indicate the past tense (e.g. play - played). In the auditory domain, this means that the constituent morphemes of a complex word unfold sequentially in time with the acoustic signal.

Semitic languages, such as Hebrew and Arabic, exhibit a markedly different internal structure with which to further test models which integrate a morphological decomposition element as part of lexical access. In the Semitic language Maltese, for example, the perfective verb kiser ‘he broke’ and its imperfective counterpart jikser ‘he breaks’ are not related only by linear prefixation, but also by word-internal changes, in this case having to do with their differing prosodic structures (CVCVC vs. CVCCVC, respectively, where C = consonant and V = vowel). Nonconcatenative word formation involves embedding a consonant root (such as ksr in kiser) within a word pattern specifying vowels and prosodic structure, and is typical of Semitic languages. For consistency with other psycholinguistic studies, we will use the Hebrew term binyan (plural = binyanim) to refer to the word pattern.

Maltese is a Semitic language spoken in the Republic of Malta where it is one of three official languages alongside English and Maltese Sign Language. Roughly half the Maltese lexicon is Indo-European in origin and utilizes linear morphology, such as the Maltese word prodott ‘product’ and its plural prodotti ‘products’. The other half of the lexicon is of Semitic origin and exhibits root-and-pattern morphology typical of other Semitic languages (Mifsud 1995, Bovingdon and Dalli 2006) such as broken plural forms (e.g., kiteb ‘book’, kotba ‘books’; see Lucas and Čéplö (2020) for a more detailed corpus analysis of etymological sources for the Maltese lexicon). As in other Semitic languages, the root carries the most salient semantic content, whereas the word pattern contributes grammatical meaning. Verbal patterns are organized into binyanim which are typically, but not always, associated with specific morpho-syntactic functions, as shown in Table 1.

Of importance to the experiments reported below, a given binyan in Maltese does not necessarily contain only verbs with a unique vocalic melody. For binyanim I and VII, there are six possible vocalic melodies, and for binyanim II and V there are five. The following illustrative examples in Tables 2 and 3 are from Galea (2011).

Formal treatments of the vocalic melody, prosodic structure, and word pattern in Maltese verbs vary in ways that the experiments we report on below are designed to test between. There are two competing hypotheses regarding the status of the Maltese verbal word pattern, which we call the phonological hypothesis and the morphological hypothesis. Traditional formal accounts of Semitic (e.g., McCarthy 1981) claim that the vocalic melody is a morpheme (the morphological hypothesis). In contrast, Fabri (2009:15-16) claims that “the function of the vowel melody in Maltese in root-based stems is phonological, not morphological” (the phonological hypothesis). A similar, phonological treatment is offered in Camilleri (2014:8, 13). While the evidence adduced in these conflicting formal accounts seems amenable to both the morphological and phonological hypotheses, our goal here is to contribute to the debate regarding the status of the Maltese word pattern with behavioral psycholinguistic evidence in order to resolve the issue. As reviewed in detail below, each hypothesis makes distinctive predictions for auditory word recognition. The phonological hypothesis would be confirmed by evidence of competition or inhibition between Maltese verbs with overlapping phonological material, either by sharing a prosodic structure or a vocalic melody. The morphological hypothesis makes a contrasting prediction: if the word pattern is morphological, then we predict that Maltese verbs with overlapping prosodic structure or vocalic melody would facilitate each other, as found for verbs sharing a consonantal root (Ussishkin, Dawson, Wedel, and Schluter 2015). We first review relevant studies in the visual modality, and then turn to the auditory modality.

2 Visual processing

Much work has been done to address the role of Semitic roots and word patterns in lexical access in visual word recognition. The preeminent method for investigating whether these morphemes facilitate lexical access when presented in their orthographic forms has been done using visual masked priming with a lexical decision task (see Forster, Mohan, and Hector 2003 for a review of this technique). For visual word processing in Arabic, it has been found that in visual masked priming experiments, when prime-target pairs share either a root or a word pattern, response to the target is facilitated (Boudelaa, Pulvermuller, Hauk, Shryrov, and Marslen-Wilson 2010, Boudelaa and Marslen-Wilson 2001 et seq). Similarly, in Hebrew, it has been shown that prime-target pairs which share the root result in facilitated lexical access for both verbs and nouns. However, prime-target pairs which shared word patterns showed inconsistent facilitation: verbal patterns facilitated lexical access and nominal patterns did not (Deutsch, Frost, and Forster 1998, though see Deutsch and Malinovitch 2016 for more recent results showing Hebrew evidence for nominal patterns as a morpheme). For Maltese, facilitation has been shown for prime-target pairs which share a root (Twist 2006), and for those which share a word pattern (Galea 2011). Twist (2006) conducted the first lexical access study on Maltese. In a masked visual priming experiment using four conditions (see examples in Table 4), she investigated whether the consonantal root and the word pattern in Maltese facilitate lexical retrieval in a lexical decision task.

Prime and target pairs in the identity condition were the same. The root condition consisted of pairs sharing the same consonantal root (e.g., ksr in nkisser - KISER) but with different word patterns. The pattern condition consisted of prime-target pairs with the same word pattern but different roots. Finally, the control condition, in which prime and target shared neither a root nor a word pattern, served as a baseline priming condition. Twist (2006) found significant facilitatory priming effects for the identity and root conditions, but not for the pattern condition. However, no significant differences were found between the pattern and unrelated conditions. Twist (2006) concluded that for Semitic verbs, the consonantal root facilitates lexical access in visual word recognition in Maltese No such claim could be made for patterns, though, due to the lack of priming in the pattern condition.

However, the design of this study did not carefully control for the variation on vocalic melody within Maltese binyanim, as illustrated above. Because Twist’s (2006) materials included pairs like nasab ‘to entrap’ - GIREF ‘to scratch’ in the pattern priming condition, an alternative interpretation of the lack of pattern priming is that it resulted from the fact that some prime-target pairs, while technically in the same verbal binyan, lacked sufficient phonological overlap to induce facilitation. This is because in such pairs, while the CV-structure is held constant between the prime and the target (e.g., CVCVC in the above example for both nasab and GIREF), the vocalic melodies between the two members of the prime-target pair are not identical (a-a vs. i-e, for instance). to address this, Galea (2011) investigated whether the vocalic melody, the CV-structure, and the combination of these two elements might facilitate lexical access in a masked visual priming experiment. Four conditions were tested (see Table 4 below for examples).

In the vowel condition, the prime and target shared the vocalic melody, and thus had the same vowels with different prosodic or CV-structure; e.g. tnikket ‘to become worried’ - FILEĠ ‘to exhaust’, where both prime and target have the vowels i-e. In the CV condition, the common property shared between the prime and the target was their CV structure. The vocalic melody was different in the prime and target, but the number and arrangement of consonants and vowels was the same (e.g. samat ‘to scald’ - FILEĠ ‘to exhaust’, where both prime and target have the structure CVCVC). In the word pattern condition, primes and targets shared both their vocalic melody and their CV structure, which together form the word pattern (e.g. dineb ‘to sin’ - FILEĠ ‘to exhaust’, where both prime and target have the vocalic melody i-e and the CV-structure CVCVC). Additionally, there was also an unrelated, or control, condition, in which the prime and target share none of these properties.

All targets in all conditions were also matched with unrelated primes to form a baseline or control condition. Facilitation was found in the vowel and CV conditions, which Galea (2011) interpreted to mean that the vocalic melody and the CV-structure are extracted from word forms during lexical access. Furthermore, Galea (2011) argued that the priming effects for the vocalic melody were due to facilitatory form priming, whereas the priming effects for the CV-structure and the word pattern were interpreted as morphological in nature.

3 Auditory processing

While much psycholinguistic research on morphological processing has focused on the visual modality, studying auditory word recognition is crucial for fully understanding the role of morphological structure in Semitic languages. This is especially true given a key orthographic feature of Semitic languages: Semitic writing systems tend to exhibit an unbalanced usage of consonants and vowels; that is, that for both Hebrew and Arabic, consonants are obligatorily written for all writing types, but the writing of vowels is restricted to particular domains such as books for religious purposes or those intended for children or beginning readers. This results in an orthographic form which is equivalent to the root but contains little obvious visual information about the pattern.

Therefore, a model of word processing for Semitic languages which only takes visual evidence into account is incomplete, and studies on auditory word processing offer a more complete picture of the role of morphological effects by effectively avoiding orthographic confounds. What’s more, only auditory methods are able to distinguish between the morphological and the phonological hypotheses for Maltese word patterns, since typically in visual processing both morphological and form-based overlap result in facilitated lexical access. On the other hand, in auditory wordrecognition, form-based overlap in the absence of a morphological relationship can inhibit lexical access.

Across Semitic languages, auditory priming studies consistently show robust facilitation for roots, but mixed and inconsistent findings for word patterns. For example, in Tunisian Arabic, Boudelaa and Marslen-Wilson (2013) reported facilitation from both roots and patterns, while in Moroccan Arabic, Schluter (2013) found root facilitation but no pattern priming effects. In Hebrew, Frost, Deutsch, and Forster (2000) reported a small, non-significant inhibitory effect for word pattern priming.

to address orthographic confounds and access early stages of auditory processing, auditory masked priming has been used (Kouider and Dupoux 2005). This technique presents participants with durationally compressed primes embedded in masks composed of compressed and reversed speech which has been reduced in volume, making them effectively imperceptible at the conscious level, while still influencing lexical access. This method has shown consistent facilitatory priming effects in Semitic languages. For Moroccan Arabic, Schluter (2013) showed that lexical decision for prime-target pairs which share roots is facilitated, but not for those which shared patterns. The same results were shown for Maltese by Ussishkin et al. (2015).

Relevant to the current study, Maltese differs from other Semitic languages by using a fully transparent Latin-based orthography, where both consonants and vowels are consistently represented. The motivation for the current study then is not to avoid an orthographic confound, but rather to explore a separate phenomenon confined to the auditory domain at the exclusion of the visual domain: Specifically, we investigate whether prime-target pairs sharing word pattern elements (prosodic structure and/or vocalic melody) facilitate, inhibit, or have no effect on spoken word recognition.

Prior research by Boudelaa and Marslen-Wilson (2000, 2004, 2005) suggests that in Arabic, elements of the word pattern (CV structure and vocalic melody) may contribute to lexical representation, but with short-lived and timing-sensitive effects, unlike the more robust and temporally stable root priming effects. Therefore, root priming has earlier and longer lasting effects in extraction and processing than word-patterns.

For Maltese, Ussishkin et al. (2015) investigated the psycholinguistic reality of roots and patterns in Maltese spoken word recognition using two auditory priming techniques in a lexical decision task. With auditory primes, facilitatory priming effects were found for prime-target pairs sharing consonantal roots, but not for prime target pairs sharing CV structure. The same results obtained using auditory masked priming, a technique originally reported by Kouider and Dupoux (2005) for French and applied by Ussishkin et al. (2015) to Maltese. Based on these experiments, Ussishkin et al. (2015) conclude that Maltese roots, but not word patterns, facilitate lexical access.

In order to further explore the nature of the word pattern in Maltese, we report here on two experiments using auditory lexical decision with priming. Both experiments were designed as auditory replications of Galea’s (2011) visual masked priming experiment. Experiment 1 used audible primes, and Experiment 2 used auditory masked primes as previously reported for Maltese in Ussishkin et al. (2015).

4 Methods

4.1 Experiment 1: Lexical decisision with audible primes

4.1.1 Description

Experiment 1 involved presentation of prime-target pairs in which both the prime and target were presented auditorily without any alteration. For each trial, an audible prime was presented followed by an audible target. Primes and targets were presented with an interstimulus interval (ISI) of 150 ms. Participants were instructed to focus on target lexicality and to make no lexical judgments regarding the prime.

4.1.2 Participants

A total of 53 subjects participated in Experiment 1 (16 male participants, 37 female participants). The mean age of participants in Experiment 1 was 22.83 years, and the median age was 20 years. Participants in Experiments 1 and 2 reported normal vision and hearing. All participants reported being bilingual speakers of Maltese and English, which is typical of Maltese speakers. All participants reported that Maltese was their dominant native language. All participants in Experiments 1 and 2 were undergraduate students at the University of Malta. Participants were recruited by email, via the on-campus news service, and in class.

4.1.3 Materials

We used the same items in Experiment 1 as Galea (2011) used for visual priming. However, recall that instead of presenting stimuli using the visual modality, our experiment presented all stimuli auditorily. Each participant was presented with 44 prime-target pairs mainly consisting of verbs, counterbalanced across three experimental lists using a Latin square among the four following priming conditions:

Prosodic structure (henceforth CV): prime-target pairs in this condition shared one of 4 prosodic templates: CVCVC, CVCCVC, tCVCCVC, nCVCVC

Vocalic melody (Henceforth V): prime-target pairs in this condition shared vocalic constructions in one of the 6 configurations: i-e, a-a, e-a, o-o, a-e, e-e

Word pattern (henceforth W): prime-target pairs in this condition belonged to the same verbal binyan from one of 4 binyanim: 1, 2, 5, 7 These binyanim were selected because they have been shown to be the most densely populated binyanim of Maltese (Francom et al. 2010)

Control: prime-target pairs in this condition were not related, so the shared neither their consonantal roots, their vocalic melodies, nor their prosodic structure.

Examples of prime-target pairs across the four experimental conditions are shown in Table 6.

As seen in Table 6, the prime and target in the CV column have the same CV structure (in the example shown, the structure is CVCVC). The second column shows a prime-target pair which shares the vocalic melody i-e but does not share the CV structure. Both the CV structure and vocalic elements are shared across the prime and target in the W column which shows both the prime and target from binyan 2. Finally, the control condition column shows an unrelated prime-target pair that does not share CV structure, vocalic elements, or word pattern/binyan.

In addition to 44 real word targets, participants were also presented with 91 real word filler targets, matched for frequency based on the MLRS corpus (Gatt and Čéplö 2013) as well as 91 nonword filler targets. Nonwords were vetted by a native speaker of Maltese to ensure their nonlexical status. All targets were matched with a real word prime.

All words and nonwords were recorded in a sound-attenuated Whisper Room booth at a sampling rate of 44,100 Hz. The native speaker read each item from a laser-printed list in Maltese orthography, and was instructed to speak in a relaxed yet careful manner. Recordings were made while the speaker wore an omnidirectional head-mounted Isomax microphone made by Countryman Associates connected via a Symetrix Audio 302 pre-amplifier to an Alesis Masterlink 9600. The native speaker pronounced each item three times, and the best token of the three was selected for use in our experiments by a trained phonetician. This was defined as the token with the clearest enunciation, the most neutral intonation and no non-linguistic intrusions such as coughs. This token was subsequently demarcated and labelled in Praat (Boersma and Weenink 2011) using a text grid, and extracted and stored as its own.wav file using a Praat script. Complete lists of prime-target pairs can be found in the Appendix.

4.1.4 Procedure

Experiment 1 was conducted at the University of Malta Institute of Linguistics and Language Technology using a desktop computer running DMDX software (Forster & Forster 2003) which presented the stimuli as well as recorded responses, which participants gave by pressing a button on the keyboard. Two dependent measures were recorded for each stimulus: RT from target onset and lexical decision accuracy. During the experiment, participants were seated in front of a computer screen and wore a pair of Audio Technica ATH-M40f headphones. Stimuli were presented auditorily through the headphones and instructions in Maltese were presented visually on the screen. Participants were instructed to listen to prime-target pairs presented auditorily using the headphones, and decide as quickly and accurately as possible whether the target was a real word of Maltese or not, with a timeout of 1500 ms. Participants were instructed to indicate their decisions by pressing buttons on a standard computer keyboard labeled IVA (“YES”) or LE (“NO”). Feedback was provided following each trial: response accuracy (‘correct’ or ‘incorrect’) was displayed on the screen after each item to establish and maintain participant focus and engagement in the task, and items which timed out at 1500 ms with no response triggered a visual message that no response had been detected. The experiment began with a six-item practice session to ensure that the participant understood the task, and the experiment lasted for approximately 20 minutes from the display of the directions to the response to the last prime-target pair.

4.1.5 Results

Reaction times (RTs) were measured from target onset. Initial analyses revealed positive skew in the distribution of reaction times (which is typical), and so reaction times were transformed to the log scale prior to analysis. Trials on which the log reaction time was greater than 2.5 standard deviations from either the subject or target item mean were excluded. In addition, incorrect responses were excluded from the reaction time analyses. Finally, 8 out of a total of 55 subjects were excluded because they provided correct answers less than 80% of the time. In total, 15% of the trials were excluded. Reaction time data were analyzed using linear mixed effects models with subjects and targets as crossed random effects.

Results were analyzed using a linear mixed effects analysis of the relationship between RTs and factors ‘Pros’, coding for shared prosodic structure (‘CV’), and ‘Vow’, coding for shared vocalic pattern (‘V’). The priming conditions, Pros and Vow were entered into the model as fixed effects, and Pros and Vow were additionally allowed to interact to model shared prosody and vowel structure (‘W’). As random effects, intercepts for Participant and Target were included in the model. Random slopes were not included to allow model convergence. Mean reaction times are presented below in Table 7.

In Table 7, the condition without either shared prosodic or vowel pattern (top left) represents to the control condition. The condition with shared prosodic pattern only (top right) and shared vowel pattern only (bottom left) represent the CV and V prime-target patterns respectively. Finally, the condition in which both prosodic and vowel patterns are shared (bottom right) corresponds to the W condition as described above. None of the conditions in which the prime shared some structure with the target were statistically different than the control in which the prime and target were unrelated.

4.1.6 Discussion

The results of Experiment 1 fail to show a significant effect of any priming condition on either facilitation or inhibition of lexical access of the target. Given that the potential inhibitory effects of either shared CV structure and shared vocalic elements on target access were not confirmed to be significant in Experiment 1, we conducted a second experiment, but rather than using both audible primes and targets, the following experiment utilized masked primes in an effort to probe into early, automatic stages of auditory processing.

4.2 Experiment 2: lexical decision with masked primes

4.2.1 Description

Experiment 2 used prime-target pairs identical to those used in Experiment 1. However, whereas in Experiment 1, primes were audible, Experiment 2 used the auditory masked priming technique. This means that although the target is consciously audible and easily perceived by participants, the prime is not consciously perceived. Auditory masked priming was first reported by Kouider and Dupoux (2005) who showed that facilitatory effects obtained for repetition primes presented at 35% compression, and that at this compression rate participants were not consciously aware of the primes. Further evidence for the lack of conscious awareness of primes masked using this technique was shown by Kouider and Dupoux (2005) for French, Schluter (2013) for Moroccan Arabic, and Ussishkin et al. (2015) for Maltese. In Experiment 2, we used the same auditory masked priming technique to examine whether the word pattern elements facilitates access of the target (supporting a morphological status of the pattern) or inhibits it (supporting a phonological status).

4.2.2 Participants

A total of 79 subjects participated in Experiment 2 (21 male, 58 female). The mean age of participants in Experiment 1 was 23.22 years, and the median age was 20 years. As in Experiment 1, participants were recruited from the University of Malta, were Maltese-dominant bilingual speakers of Maltese and English, and reported normal vision and hearing. No subjects who participated in Experiment 1 participated in Experiment 2.

4.2.3 Materials

Materials in Experiment 2 were identical to those in Experiment 1. However, in in Experiment 2, primes were masked by durational compression to 35%, and were embedded in a sequence in which it follows one forward mask and precedes five backward masks. Prime compression retains the original pitch of the uncompressed version of the prime. Masks are created by durationally compressing and temporally reversing other words individually. Masks and prime are all volume attenuated by a reduction of 15 dB, and the target is superimposed on the sequence of masks and prime such that target onset coincides with prime offset. In contrast with the prime and masks, the target is presented at a normal volume. A visual representation of the anatomy of a prime-target pair used in auditory masked priming is shown in Figure 1: “mask” is backward to visually represent that mask words were temporally reversed; “mask” and “prime” are written with close font spacing to visually represent that mask and prime words were durationally compressed to 35% of their original duration.

4.2.4 Procedure

Experiment 2 was also conducted at the University of Malta Institute of Linguistics and Language Technology using a desktop computer running DMDX software (Forster & Forster 2003) which presented the stimuli as well as recorded responses, which participants gave by pressing a button on the keyboard. Two dependent measures were recorded for each stimulus: RT from target onset and lexical decision accuracy. During the experiment, participants were seated in front of a computer screen and wore a pair of Audio Technica ATH-M40f headphones. Stimuli were presented auditorily through the headphones and instructions in Maltese were presented visually on the screen. Participants were instructed that they would hear, on each trial, some noise during which there would be something intelligible (the target), which may or may not be a real word of Maltese. The instructions explicitly stated that hearing the intelligible portion might be difficult. They were instructed to decide as quickly and accurately as possible whether the target was a real word of Maltese or not, with a timeout of 1500 ms. Participants were instructed to indicate their decisions by pressing buttons on a standard computer keyboard labeled IVA (“YES”) or LE (“NO”). Other than reference to ‘noise’ in the visually presented instructions, subjects were not given any information about the content of the auditory masks or the compressed prime that accompanied each target. Feedback was provided following each trial: response accuracy (‘correct’ or ‘incorrect’) was displayed on the screen after each item to establish and maintain participant focus and engagement in the task, and items which timed out at 1500 ms with no response triggered a visual message that no response had been detected. The experiment began with a six-item practice session to ensure that the participant understood the task, and the experiment lasted for approximately 20 minutes from the display of the directions to the response to the last prime-target pair.

4.2.5 Results

As in Experiment 1, reaction times (RTs) were measured from target onset, and were log-transformed (we note, however that results were not substantially different using untransformed RTs). As in Experiment 1, trials on which the log reaction time was greater than 2.5 standard deviations from either the subject or target item mean were excluded, and incorrect responses were excluded from the reaction time analyses. 16 out of a total of 73 subjects were excluded because they provided correct answers less than 80% of the time. In total, 30% of the trials were excluded. Reaction time data were analyzed using linear mixed effects models with subjects and targets as crossed random effects.

As in Experiment 1, results were analyzed using a linear mixed effects analysis of the relationship between RTs and factors ‘Pros’, coding for shared prosodic structure (‘CV’), and ‘Vow’, coding for shared vocalic pattern (‘V’). The priming conditions, Pros and Vow were entered into the model as fixed effects, and Pros and Vow were additionally allowed to interact to model shared prosody and vowel structure (‘W’). As random effects, intercepts for Participant and Target were included in the model. Random slopes were not included to allow model convergence. Mean reaction times are presented below in Table 8.

In Table 8, the condition without either shared prosodic or vowel pattern (top left) represents to the control condition. The condition with shared prosodic pattern only (top right) and shared vowel pattern only (bottom left) represent the CV and V prime-target patterns respectively. Finally, the condition in which both prosodic and vowel patterns are shared (bottom right) corresponds to the W condition as described above. This data is presented below in Figure 2 as an interaction plot.

The ‘V’ condition in which the vowel pattern but not the prosodic pattern is shared between the Prime and Target has a significantly higher RT (t = 2.4) than the control condition (Prosodic pattern and vowel pattern both absent). RTs in the ‘CV’ condition in which the prosodic pattern but not the vowel pattern is shared trend higher, but do not reach statistical significance. The interaction condition ‘W’ in which both are shared has a significantly lower RT than the single pattern conditions (t = 2.7) and is numerically similar to the control with no patterns shared.

4.2.6 Discussion

The results of Experiment 2 show inhibition, not facilitation, for target words primed by a word with the same vocalic melody but a differing prosodic structure. This supports the hypothesis that the vocalic melody does not have morphological status in Semitic Maltese verbs. Similarly, the trend for a similar pattern of inhibition with respect to prosodic structure suggests that this structure is a form-based, non-morphological element of Maltese verbs as well.

5 General discussion and conclusion

With audible primes (Experiment 1), none of the priming conditions in which the prime and target shared any material showed a significant difference from the control condition. Recall that models treating any aspect of the word pattern as morphemic predict that participants should respond more quickly in those conditions, but this did not occur and therefore the results of Experiment 1 fail to support models in which the word pattern is morphological. On the other hand, with masked primes (Experiment 2), the only significant difference in response times was a slower response when targets were primed by another word with the same vocalic melody, though if the two shared their prosodic structure this effect was not found. Again, these results fail to support models treating the word pattern as morphological.

The significant inhibitory effect for prime-target pairs sharing only their vocalic melody provides support for the theory proposed by Fabri (2009), whose model proposes that the vocalic melody is a strictly phonological element. Unlike the parallel visual experiment reported by Galea, where the vocalic melody facilitated recognition of visually presented words, in auditory word recognition with masked primes, sharing the vocalic melody significantly inhibits word recognition. We conclude, therefore, that there is no priming for patterns in spoken word recognition in Maltese. This result provides additional motivation for focusing on the mechanisms and processes that shape spoken word recognition.

6 Appendix

6.1 List of real word stimuli (prime-target pairs) for Experiment 1

6.2 List of real word stimuli (prime-target pairs) for Experiment 2

7 References

Baayen, R. Harald, Yu Ying Chuang, Elnaz Shafaei-Bajestan, & James Blevins. 2019. “The discriminative lexicon: A unified computational model for the lexicon and lexical processing in comprehension and production grounded not in (de)composition but in linear discriminative learning.” Complexity 2019 (1). 4895891.

Boersma, Paul & David Weenink. 2011. Praat: Doing phonetics by computer (Version 5.2.20).

Boudelaa, Sami & William Marslen-Wilson. 2000. “Non-concatenative morphemes in language processing: Evidence from Modern Standard Arabic.” Proc. Spoken Word Access Processes (SWAP), 23–26.

Boudelaa, Sami & William Marslen-Wilson. 2001. “Morphological units in the Arabic mental lexicon.” Cognition 81 (1). 65–92.

Boudelaa, Sami & William Marslen-Wilson. 2004. “Abstract morphemes and lexical representation: The CV-Skeleton in Arabic.” Cognition 92 (3). 271–303.

Boudelaa, Sami & William Marslen-Wilson. 2005. “Discontinuous morphology in time: Incremental masked priming in Arabic.” Language and Cognitive Processes 20 (1). 207–260.

Boudelaa, Sami & William Marslen-Wilson. 2013. “Morphological structure in the Arabic mental lexicon: Parallels between standard and dialectal Arabic.” Language and Cognitive Processes 28 (10). 1453–1473.

Boudelaa, Sami, Friedemann Pulvermüller, Olaf Hauk, Yury Shtyrov, & William Marslen-Wilson. 2010. “Arabic morphology in the neural language system.” Journal of Cognitive Neuroscience 22 (5). 998–1010.

Bovingdon, Roderick & Angelo Dalli. 2006. In Dawn Archer & Paul Rayson (eds.), Statistical analysis of the source origin of Maltese, 63–76. Leiden: Brill.

Camilleri, Maris. 2014. The stem in inflectional verbal paradigms in Maltese. Unpublished thesis, University of Surrey (United Kingdom).

Davis, Colin, Jeesun Kim, & Angelo Barbaro. 2010. “Masked speech priming: Neighborhood size matters.” The Journal of the Acoustical Society of America 127 (4). 2110–2113.

Davis, Colin & Stephen Lupker. 2006. “Masked inhibitory priming in English: evidence for lexical inhibition.” Journal of Experimental Psychology: Human Perception and Performance 32 (3). 668.

Deutsch, Avital, Ram Frost, & Kenneth Forster. 1998. “Verbs and nouns are organized and accessed differently in the mental lexicon: Evidence from Hebrew.” Journal of Experimental Psychology: Learning, Memory, and Cognition 24 (5). 1238.

Deutsch, Avital & Tamar Malinovitch. 2016. “The role of the morpho-phonological word-pattern unit in single-word production in Hebrew.” Journal of Memory and Language 87. 1–15.

Dufour, Sophie & Ronald Peereman. 2004. “Phonological priming in auditory word recognition: Initial overlap facilitation effect varies as a function of target word frequency.”* Current Psychology Letters: Behaviour, Brain & Cognition* 14 (3). Available online at https://journals.openedition.org/cpl/437.

Fabri, Ray. 2009. “To agree or not to agree: suspension of formal agreement in Maltese.” In Ray Fabri (ed.), Maltese linguistics: A Snapshot. In memory of Joseph A. Cremona (1922–2003) (Il-Lingwa Tagħna - Our Language 1/2009), 35-61. Bochum: Universitätsverlag Dr. N. Brockmeyer.

Forster, Kenneth & Jonathan Forster. 2003. “DMDX: A Windows display program with millisecond accuracy.” Behavior Research Methods, Instruments, & Computers 35. 116–124.

Forster, Kenneth, Kathleen Mohan, & Jo Hector. 2003. “The mechanics of masked priming.” In Sachiko Kinoshita & Stephen J. Lupker (eds.), Masked Priming: The State of the Art, 3–37.

Frauenfelder, Uli, Mark Scholten, & Alain Content. 2001. “Bottom-up inhibition in lexical selection: Phonological mismatch effects in spoken word recognition.” Language and Cognitive Processes 16 (5–6). 583–607.

Frost, Ram, Avital Deutsch, & Kenneth Forster. 2000. “Decomposing morphologically complex words in a nonlinear morphology.” Journal of Experimental Psychology: Learning, Memory, and Cognition 26 (3).751.

Galea, Luke. 2011. Mental representations of Maltese Semitic verbs. Unpublished thesis, University of Essex.

Gatt, Albert & Slavomír Čéplö. 2013. “Digital corpora and other electronic resources for Maltese.” In Andrew Hardie & Robbie Love (eds.), Corpus Linguistics, 96–97. Lancaster: UCREL.

Goldinger, Stephen, Paul Luce, & David Pisoni. 1989. “Priming lexical neighbors of spoken words: Effects of competition and inhibition.” Journal of Memory and Language 28 (5). 501–518.

Goldinger, Stephen, Paul Luce, David Pisoni, & Joanna Marcario. 1992. “Form-based priming in spoken word recognition: The roles of competition and bias.” Journal of Experimental Psychology: Learning, Memory, and Cognition 18 (6). 1211.

Jaeger, T. Florian, Katrina Furth, & Caitlin Hilliard. 2012. “Incremental phonological encoding during unscripted sentence production.” Frontiers in Psychology 3. 481.

Kouider, Sid & Emmanuel Dupoux. 2005. “Subliminal speech priming.” Psychological Science 16 (8). 617–625.

Lorch, Robert, David Balota, & Edward Stamm. 1986. “Locus of inhibition effects in the priming of lexical decisions: Pre- or postlexical access?” Memory & Cognition 14. 95–103.

Lucas, Christopher & Slavomír Čéplö. 2020. “Maltese.” In Christopher Lucas & Stefano Manfredi (eds.) Arabic and contact-induced change, 265–302. Berlin: Language Science Press.

Luce, Paul & David Pisoni. 1998. “Recognizing spoken words: The neighborhood activation model.” Ear and Hearing 19 (1). 1–36.

Lupker, Stephen. 1984. “Semantic priming without association: A second look.” Journal of Verbal Learning and Verbal Behavior 23 (6). 709–733.

Marslen-Wilson, William. 1987. “Functional parallelism in spoken word recognition.” Cognition 25 (1–2). 71–102.

Marslen-Wilson, William & Alan Welsh. 1978. “Processing interactions and lexical access during word recognition in continuous speech.” Cognitive Psychology 10 (1). 29–63.

McCarthy, John. 1981. “A prosodic theory of nonconcatenative morphology.” Linguistic Inquiry 12 (3). 373–418.

McClelland, James & Jeffrey Elman. 1986. “The TRACE model of speech perception.” Cognitive Psychology 18 (1). 1–86.

Meunier, Fanny & Juan Segui. 1999. “Frequency effects in auditory word recognition: The case of suffixed words.” Journal of Memory and Language 41 (3). 327–344.

Mifsud, Manwel. 1995. Loan verbs in Maltese: A descriptive and comparative study. Leiden: Brill.

Norris, Dennis. 1994. “Shortlist: A connectionist model of continuous speech recognition.” Cognition 52 (3). 189–234.

Norris, Dennis & James McQueen. 2008. “Shortlist B: A Bayesian model of continuous speech recognition.” Psychological Review 115 (2). 357–395.

Radeau, Monique, José Morais, & Juan Segui. 1995. “Phonological priming between monosyllabic spoken words.” Journal of Experimental Psychology: Human Perception and Performance 21 (6). 1297.

Schluter, Kevin. 2013. Hearing words without structure: Subliminal speech priming and the organization of the Moroccan Arabic lexicon. Unpublished thesis, The University of Arizona.

Seidenberg, Mark, Gloria Waters, Marcia Barnes, & Michael Tanenhaus. 1984. “When does irregular spelling or pronunciation influence word recognition?” Journal of Verbal Learning and Verbal Behavior 23 (3). 383–404.

Slowiaczek, Louisa & MaryBeth Hamburger. 1992. “Prelexical facilitation and lexical interference in auditory word recognition.” Journal of Experimental Psychology: Learning, Memory, and Cognition 18 (6). 1239.

Slowiaczek, Louisa, Howard Nusbaum, & David Pisoni. 1987. “Phonological priming in auditory word recognition.” Journal of Experimental Psychology: Learning, Memory, and Cognition 13 (1). 64.

Taft, Marcus, Gail Hambly, & Sachiko Kinoshita. 1986. “Visual and auditory recognition of prefixed words.” The Quarterly Journal of Experimental Psychology Section A 38 (3). 351–365.

Twist, Alina. 2006. A psycholinguistic investigation of the verbal morphology of Maltese. Unpublished thesis, The University of Arizona.

Ussishkin, Adam, Colin Dawson, Andrew Wedel, & Kevin Schluter. 2015. “Auditory masked priming in Maltese spoken word recognition.” Language, Cognition and Neuroscience 30 (9). 1096–1115.

West, Richard & Keith Stanovich. 1982. “Source of inhibition in experiments on the effect of sentence context on word recognition.” Journal of Experimental Psychology: Learning, Memory, and Cognition 8 (5). 385.

Winther Balling, Laura & Harald Baayen. 2008. “Morphological effects in auditory word recognition: Evidence from Danish.” Language and Cognitive Processes 23 (7–8). 1159–1190.

Wurm, Lee. 2000. “Auditory processing of polymorphemic pseudowords.” Journal of Memory and Language 42 (2). 255–271.

Ziegler, Johannes & Mathilde Muneaux. 2007. “Orthographic facilitation and phonological inhibition in spoken word recognition: A developmental study.” Psychonomic Bulletin & Review 14. 75–80.