Second-language acquisition of the English past-tense: from rules to analogy
Abstract
188
Full text
Second-language acquisition of the English past-tense: From rules to analogy Lucie Jiránková— Luca Cilibrasi (Charles University, Prague) ABSTRACT The present study investigates the production of novel morphologically inflected forms in secondlanguage learners of English with Czech as L1. The study attempts to investigate which production model (singleor dual-route) best accounts for L2 learners’ morphological productivity when forming regular past forms of novel words. Additionally, it explores the possible interference effects of L1. 88 English L2 learners and 9 native speakers heard sentences in which anew activity was described with anovel word (The baby likes to dize. Look, there it is dizing. Everyday it dizes.) and past-tense forms were elicited (So yesterday it…). The results revealed that for native speakers the likelihood of averb being produced in aregular past-tense form was inversely related to its phonological similarity to existing irregular verbs (replicating previous studies). L2 speakers showed adevelopment in this direction: While for the A1 to B1 participants similarity to existing irregulars did not matter, B2 and C1 participants appeared to be sensitive to these similarities and behaved comparably to native speakers. In addition to the form analysis, the reaction-times results showed that the lowest language levels used their L1 as aperformance facilitator (with slower performance with novel words that do not respect the phonology of the participants’ L1), while proficient learners and native speakers were not sensitive to this property of the novel words. The results suggest that the L2 acquisition of the English past-tense is characterized by adevelopment from the mastery of mechanistic rules to the refinement of their application based on analogical patterns extracted from existing verbs, with Czech promoting the production at the earliest proficiency stages. KEYWORDS analogy, Czech, English, inflectional morphology, past-tense, rules, second-language acquisition DOI https://doi.org/10.14712/18059635.2021.2.4 1 INTRODUCTION The baby likes to dize. Look, there it is dizing. Everyday, it dizes. So yesterday it…dized? Or doze? The answer to this question poses anumber of difficulties. The nature of representations for inflectional morphology has fuelled along-term and intense psycholinguistic debate, both in the study of perception and in the study of production. The discussion of whether novel morphologically inflected forms are generally produced by the application of rules (e.g. Prasada & Pinker 1993) or by analogy based on stored examples (e.g. Bybee & Slobin 1982) is continuously addressed in the field. Special attention is being paid to the difference between regular and irregular inflection, particularly that of the English past tense, since English encompasses arelatively clear-
LUCIE JIRáNKOVá— LUCA CILIBRASI 189 cut distinction between the two forms and therefore suits perfectly to the purpose of disentangling between these two approaches. Regular past-tense inflection realized by the “add -ed to the verb stem” rule (as in jump— jumped or cook— cooked) applies to thousands of verbs and is often productively used in the generalization of new word-forms (as in rick— ricked in Berko, 1958), while irregular past-tense forms (such as fall— fell) apply only to circa 180 known irregular verbs and often fall victim to overgeneralization by children (as in *sitted instead of sat) (Pinker & Ullman 2002). These irregular forms have been traditionally assumed to be acquired and stored as compact units in our memory (Prasada & Pinker 1993). However, the simple idea that regular past-tense forms are obtained from rules and irregular past-tense forms are stored as units has proved problematic in several aspects: irregular verbs, for instance, form families of similar verbs such as cut and put whose past tense remains the same as the present form and thus allows for analogical generalization of novel verbs (see, for instance, Bybee & Moder 1983). English also includes many quasi-regular inflections (Cilibrasi et al. 2019: 750): such past-tense forms are “obtained following aproductive pattern that is, however, less frequent than that normally labelled as regular.” Examples of quasi-regular past-tense forms would be feel/felt, keep/kept, or build/built. The main question of the above-mentioned debate thus centres around whether “regular and irregular past tense forms are generated by two qualitatively distinct mechanisms or whether all forms are produced in asingle, associative process” (Westermann & Ruh 2012: 649). Despite the length of the debate, the question remains partly unanswered, and the views differ diametrically. Generativist theories often attribute generalization to rules, connectionist theories attribute it to association based on analogy, while hybrid theories attribute regular generalization to rules and irregular generalization to analogy, but they also suggest that frequent regular verbs may be stored in memory with their inflection (Prasada & Pinker 1993). As aconsequence, two models have been proposed for past-tense production: (i)the single-route model (see, for instance, Bybee & Moder 1983), which assumes that both regular and irregular past-tense forms are produced by analogy across word forms that already exist in our mental lexicon, and (ii) the dual-route model (see, for instance, Prasada & Pinker 1993), which assumes that the two past-tense forms are produced by two different processes: the generation of regular past-tense forms is governed by the application of abasic rule for past-tense inflection (i.e. adding –ed to the verb stem), while irregulars are stored as units and used in analogical generalization. It is evident that both models assume that the same mechanism is used in the generation of irregular past-tense forms, stating that if our mind does not retrieve the irregular form directly from our mental lexicon as aunit, analogy steps in and the irregular form in question is produced based on its phonological similarity to existing forms that are already stored in our memory. Investigating irregular verbs, therefore, does not help us decide which model is abetter description of how past-tense forms are produced, and regular inflection needs to be studied instead. The majority of psycholinguistic studies generally use novel words to test morphological productivity (e.g. Prasada & Pinker 1993; Albright & Hayes 2003; Ambridge 2010; Blything et al. 2018) since such aprocedure requires generalization and avoids
190 LINGUISTICA PRAGENSIA 2/2021 effects related to the lexical access (Blything et al. 2018). This assumption is also at the foundation of studies such as Albright and Hayes (2003) and Blything et al. (2018) who used an elicited production task to investigate which of the two models is better at describing morphological productivity in native speakers of English. This experiment builds on these two previous studies and aims to additionally explore (i) the development of morphological productivity of second-language learners of English and (ii) the effect of their L1 (Czech) on the L2 production. 2 THEORETICAL FRAMEWORK AND HYPOTHESES Albright and Hayes (2003), Ambridge (2010), and Blything et al. (2018) addressed these issues using acceptability judgement and novel-verb production tasks, testing adults and children with aclear intention to distinguish between the two aforementioned models. Albright and Hayes’ study (2003) created novel verbs that showed phonological similarity or dissimilarity to existing regular and irregular English verbs. Using an elicited production task, the authors found that novel words that resemble existing regular verbs were more likely to be produced and judged as regular past forms by adult English speakers and that novel words that resemble existing irregular verbs were similarly more likely produced and judged as irregular past forms, in line with the single-route model. Replicating the judgement section of Albright and Hayes’s study (2003), Ambridge (2010) tested children aged 6–7 and 9–10. Again, he found that the participants favourably judged novel irregular past forms that phonologically resembled existing irregular verbs. The older group of the English-speaking children also more often accepted regular past forms of novel verbs that highly resemble existing regular verbs, again in line with the single-route model. Blything et al.’s study (2018) builds on the two previous studies. Using the novel verbs created by Albright and Hayes (2003), the authors investigated the mechanisms that underlie morphological productivity, focusing on verbal morphology of the English past tense and recruiting groups of children from 3 up to 10 years of age. Their results have shown that “the likelihood of anovel verb being produced in aregular pasttense form is positively associated with its phonological similarity to existing regular verbs” (Blything et al., 2018: 3), in line with the single-route model. All three studies found that both past forms of novel verbs are generated by analogy across existing word forms stored in our mental lexicon and ruled by their similarity to these stored forms (Ambridge 2019). These findings provide us with tentative evidence that the single-route model might be better at describing the morphological production of novel forms in comparison to the dual-route model, and, consequently, that (structural) analogy prevails over rules in regular past-tense production. The authors investigated both adults and older children of 9–10 years of age, and also smaller children at the peak rate of over-generalization, not only paying attention to mature linguistic systems, but also to the systems still undergoing development (Blything et al.2018). This experiment aims to explore asimilar step in the acquisition of pasttense production by focusing on second-language learners of English, and it also attempts to describe any potential difference from native speakers of English.
LUCIE JIRáNKOVá— LUCA CILIBRASI 191 Previous studies on L2 morphological development show that the dual-route model is not capable of fully explaining and describing the production patterns of L2 learners. Studying frequency effects on regular verbs, Beck (1997) and Ellis & Schmidt (1998) reported findings inconsistent with the dual-route model. In their studies, the authors crucially showed that the frequency effects do not apply only to irregular verbs but also to regular ones, suggesting aprevalence of analogical processing. These results match those presented in Murphy (2000), who showed that the assumptions of the dual-route model are difficult to apply to L2 data. Her results are better explained by an alternative, associative model. Even though her L2 participants (similarly to native speakers) did add past-tense suffixes to novel regular verbs more than to the irregular ones, both native and non-native speakers used the novel verb’s phonological resemblance to existing (ir)regulars as adecisive factor in producing past-tense forms— afinding not predicted by the dual-route model. These results thus undermine the claims of the dual-route mechanism that supports acomplete dissociation of the processes behind the two past-tense forms (regular vs. irregular) and of the learning systems underlying language representation. Murphy (2000: 112) proposes “amore parsimonious account invoking asingle set of associative learning mechanisms” instead. In more recent years, Cuskley et al. (2015) examined the difference between native and L2 production of past tense using anovel-word task, showing that both groups of participants show sensitivity towards phonological similarities between the novel and existing verbs. Their results in fact show that it is more probable that the participants will produce (ir)regular past forms with novel verbs that phonologically resemble existing (ir)regular verbs, in line with the single-route model. In addition, the authors have also noted that instead of showing astraightforward preference for the default regular past-tense rule, the L2 learners were generally more prone to using sub-rules for the production of various types of irregular past-tense forms. Similarly to Murphy (2000), Agathopoulou (2009) found that adult Greek L2 learners of English and English native speakers produced past tense forms of novel verbs that cannot be fully captured by the dual-route model (due to an apparent similarity effect found in the elicited forms) and noted no qualitative difference between how native and nonnative speakers of English handled English (ir)regular verb morphology. These studies offer interesting insights into morphological processing in L2 learners. Building on them, the present research aims at exploring how these processing strategies are achieved by L2 learners of English with Czech as L1 and thus at describing the development of their L2 morphological system. In comparison to the English past-tense morphology, Czech creates past-tense forms through the combination of the past stem of averb and the addition of the so-called past-tense “l-forms,“ i.e.suffixes -l, -la,-lo,-li/y. These forms vary according to gender and number of the subject, offering thus larger variability in comparison to English inflection. In addition, number and person may be expressed by an auxiliary verb, which appears with the main verb and thus contributes to the morphological complexity of the sentence. Finally, the past stem can undergo stem alteration (Čechová 1996), depending on the verb employed (e.g. brát— bral or chtít— chtěl). Since Czech is alanguage with such acomplex inflectional system, it creates an interesting contrast with the notoriously poor inflectional system of English and therefore provides asuitable ground for in-
192 LINGUISTICA PRAGENSIA 2/2021 vestigating possible interference effects of Czech on the acquisition of English inflectional morphology. To achieve our aims, this study uses aproduction task with novel verb stimuli and tests participants with various proficiencies, including participants in the initial stages of learning, participants with rather mature systems of the target language, and the range in between. The inclusion of aprogression of language levels in our study is what strengthens our argument in comparison to other L2 studies that usually work with L2 learners as aunit or with two or three selected language proficiencies. Testing participants from the A1 to the C1 level allows us to properly map the development of inflectional morphology in L2 learners and describe its specifics. 2.1 HYPOTHESES Based on the previous research outlined above, our tentative hypothesis was that the dual-route model would not fully explain the production of L2 learners and that the native speakers would perform similarly to the participants tested in Albright and Hayes (2003) and Blything et al. (2018). Regarding language proficiency, the study also operated with the tentative hypothesis that the lower levels might be more inclined to use rules due to lesser experience with L2, while the higher language levels might be more inclined to use analogy (and therefore resemble more native speakers of English) due to having more significant experience with the language and the analogical rules underlying past-tense formation. As far as Czech interference effects are concerned, there were two possible outcomes: (i) the reaction times may be quicker for the novel words that are phonotactically legal in Czech (referred to as the A-set in the study) for the L2 learners since they will be in away familiar with the word structure from their mother tongue, and (ii) the novel words phonotactically legal in Czech (A-set) may, on the contrary, cause inhibitory effects and slow the reaction times since the L2 learners will be faced with both their L1 and L2 at the same time. 3 DATA AND METHODOLOGY 3.1 PARTICIPANTS 88 English second-language learners with Czech as their first language at A1–C1 proficiency levels1 and acontrol group of 9 native speakers2 were recruited. None of the 1 The participants were divided into proficiency levels based on aplacement test created on the basis of the English Grammar and Vocabulary Profiles that describe which lexical and grammatical aspects of English are typically learned at each proficiency level. 2 Acontrol group consisting of only 9 native speakers may not seem as asufficiently large group for any comparison with the L2 speakers; however, our group of native speakers truly served the control purpose and our findings could be further compared to the previous work by Albright and Hayes (2003) and Blything et al. (2018) who focused on adult native speakers of English, using the exact same task.
LUCIE JIRáNKOVá— LUCA CILIBRASI 193 participants had been diagnosed with any language impairment. The mean age was 37 years, standard deviation 9.4 years. 73 participants were female, 24 were male. All subjects were recruited from the faculty language school at the Faculty of Arts, Charles University, and testing occurred at this location in aprivate room. 3.2 STIMULI The study adopted 32 novel verbs that were deemed phonotactically legal in English3 (e.g. bize, drice, spling) from Albright and Hayes’ (2003) and Blything et al.’s (2018) studies. Some of the original novel words were excluded due to their existence in Present-Day English4 (since the novel words were created in 2003 and some of them have already entered the language as real words since) and the need for abalanced dataset. Albright and Hayes (2003) started with the creation of acontrol corpus of existing English verbs which were all sourced from the English part of the CELEX lexical database (Baayen et al., 1995), with aspecial focus on those verbs whose lemma frequency equalled or exceeded 10. If the verb had more than one past-form variant (such as dive-dived/dove), both alternatives were included separately. The resulting corpus of existing English verbs consisted of 4,035 regular and 218 irregular verbs. These verbs served as evidence of real phonotactic properties and phonological changes in past-tense forms and were used for counting the (dis)similarity of the created novel-verb stimuli to the existing English (ir)regular verb. In order to do that, the phonemic transcriptions of these verbs were fed into Albright and Hayes’ rulebased model (Albright & Hayes 2003). Albright and Hayes used the same model for the creation of the novel verbs. They created adataset of 2,344 novel-verb forms by “concatenating combinations of relatively common syllable onsets and syllable rhymes” (Albright & Hayes 2003: 135). The full list of novel-verb forms was then uploaded back into the model, which then produced (ir)regular past forms for each novel verb and also rated their similarity to existing (ir)regulars (Albright & Hayes 2003). For each novel word, Albright and Hayes (2003) and Blything et al. (2018) employ two ways of counting the novel word’s similarity to existing (ir)regulars— based on (i) the Generalized Context Model (hereby GCM; see Nosofsky 1990) and (ii) the Minimal Generalization Learner (hereby MGL; see Albright & Hayes 2003). GCM provides values of (i) “variegated” similarity, while MGL outputs (ii) “structured” similarity (Albright & Hayes 2003; Blything et al. 2018). To ensure findings are compatible with Albright and Hayes (2003) and Blything et al. (2018), the current study also employed ameasure of the novel verbs’ similarity to existing regular and irregular past-tense verbs. Since Albright and Hayes (2003) and Blything et al. (2018) introduce two different models of similarity measures (the 3 Phonotactically legal English novel words follow the restrictions on the permissible combinations of English phonemes (e.g. Pitt 1998). 4 Albright and Hayes created the novel words in 2003 and since then some of their monosyllabic novel verbs have come into existence as English words, e.g. rife (nowadays afrequently-used adjective), which is not asurprising phenomenon for amonosyllabic English novel word. The whole dataset was thus checked by three native speakers to eliminate such instances.
194 LINGUISTICA PRAGENSIA 2/2021 aforementioned MGL and the GCM), it was necessary to decide which of the two models to use in the data analysis. As Blything et al. (2018: 6) argue, GCM compares “phonological segments across entire words such that phonological similarities at the beginning, middle, and end of aword have equal weighting, and each comparison form can be similar to the novel verb in its own way.” The stimulus’ similarity to each related word is compared “feature-by-feature” (Ambridge 2019: 12) (e.g. the change from ito ain sit, spit, or swim when forming past tense) and is then divided by its feature-by-feature comparison to all existing verbs stored in out mental lexicon. The result yields the probability of agiven word being apart of the specific class. For that reason, variegated similarity is sometimes regarded as a“pure” analogy by some linguists (Albright & Hayes 2003: 122). On the contrary, MGL (Albright & Hayes 2003), working with the structured similarity, pays attention “only to phonological properties that are shared uniformly among comparison verbs (e.g. the verb cling shares its final two segments with string, sting, and fling, etc.)” (Blything et al. 2018: 10) and thus assess the similarity to the phonological properties based on the features “most relevant to averb’s past-tense form (e.g. the past tense of cling is clung because it shares afinal segment with string, sting, and fling)” (Blything et al. 2018: 10), creating explicit micro-rules for each sub-regularity (Ambridge 2019). Measures of variegated similarity of cling may be thus misled by irrelevant segments of aword. Blything et al. (2018: 7) conclude that as such, “structured similarity is compatible with more sophisticated conceptions of analogy such as schema-based approaches” (see, for instance, Langacker 2000). Considering possible shortcomings of the variegated similarity measure obtained from GCM, the analysis of the current study will use the structured similarity measure as aprimary decisive metric. Since the article also attempts to explore the effect of L1 on second-language production, cross-language interference effects have been taken into consideration. The 32 novel verbs were further divided into 16 phonotactically legal (Aset) and 16 phonotactically illegal (B set) novel words for Czech (for the full table of novel word stimuli, see the Appendix).5 Since there is no existing calculator of anovel word’s similarity to an existing Czech word, Šturm and Lukeš (2017)‘spaper “Fonotaktická analýza obsahu slabik na okrajích českých slov vmluvené apsané řeči” was used as areference. This study investigates Czech “syllable onsets and final codas” (Šturm & Lukeš 2017: 99) both in spoken and written Czech texts imported from the Czech National Corpus. Along with adetailed study on the complexity of Czech syllable onsets and codas, the authors have also compiled alist of all phonotactically legal Czech syllable onsets and codas.6 Using this reference list, and more specifically its spoken part (since the experiment was equipped with an aural prompt), our stimuli were further divided into 16 words that were phonotactically legal in Czech (i.e. both the syllable onset and coda were deemed phonotactically legal in Czech) and into 16 words phonotactically illegal in Czech (i.e., either the syllable onset or coda or both parts were deemed 5 Phonotactically legal Czech novel words follow the restrictions on the permissible combinations of Czech phonemes (e.g. Pitt 1998— an experimental study done on English but whose findings are replicable also on Czech). 6 Available at https://fonetika.ff.cuni.cz/vyzkum/materialy/fonotaktika/.
LUCIE JIRáNKOVá— LUCA CILIBRASI 195 phonotactically illegal in Czech). One of the major downsides of this division (and the study in general) is that Šturm and Lukeš’s paper does not investigate syllable nuclei, and our stimuli division therefore could not provide afull and complex picture of Czech phonotactic (il)legality. However, given the lack of any automatized calculator that would allow stimuli importation and export the word’s similarity or dissimilarity to existing Czech words, this phonotactical analysis was the only relevant reference point that could be used for the division. Therefore, even though we are fully aware of its drawbacks, using this paper was the only suitable reference for the inclusion of Czech interference effects available. Any assumptions on the effect of Czech, however, need to be read with this in mind. The novel words were then incorporated into frame sentences sourced from Blything et al. (2018). The original sentences needed to be adapted lexically (using the English Vocabulary Profile7) for the lowest-level participants, i.e. the A1 level, to ensure the study’s applicability and participants’ understanding. Each frame sequence contained four sentences presenting the novel word in abare-infinitive, presentprogressive, and present-simple form of the following template: “The (agent) likes to VERB. Look, there he is VERBing. Everyday he VERBs. So yesterday he…” (e.g.The baby likes to dize. Look, there it is dizing. Everyday it dizes. So yesterday it…). To avoid semantic effects of the surrounding words (as documented, for instance, in Ramscar, 2002) on the choice of the past-tense form, each novel word was inserted into three different frame sentences (e.g. This person/the waitress/my father likes to dize. Look, there s/he is dizing. Everyday s/he dizes. So yesterday s/he…). The pairing of verbs and frames was randomized and different for each participant. 3.3 PROCEDURE In an elicited production paradigm, each participant was presented with the 32 novel verbs inserted into frame sentences in amixed (both withinand betweensubjects) design. The participants were told they would hear sentences in which someone would perform an activity described with anovel word. They were then instructed to finish the sentence when the recording stopped. The participants were thus prompted to say what the agent “did yesterday”, and the past-tense forms elicitation was ensured. The recorded measures obtained from this experiment included the produced forms and reaction times to the presented novel word (i.e. the time between the end of the stimulus sentence and the participant’s response). 4 RESULTS 4.1 ANALYSIS OF THE PRODUCED FORMS This analysis attempts to answer the question of whether the produced form is somehow influenced by the novel word’s similarity to existing (ir)regulars or not. 3,094 trials were recorded in total. Responses were coded adopting the response scale used by Blything et al. (2018), i.e., based on whether the recorded output was (i) aregular 7 Accessible at https://www.englishprofile.org/wordlists/evp.
196 LINGUISTICA PRAGENSIA 2/2021 past-tense form (e.g. blafed), (ii) an irregular past-tense form8 (e.g. blofe), (iii) aform without any formal change (e.g. blafe), (iv) athird-person singular of present simple (e.g. blafes), (v) aprogressive form (e.g. blafing), or (vi) aform unclassifiable under (i)–(v) (e.g. blafest). The mean proportion of each output form by language level is given in Table 1 below. Regular (vs.irregular only) Regular (vs.all) Irregular (vs.all) No change (vs. all) Third-person present (vs.all) Past progressive (vs.all) Unclassified (vs. all) A1 level 94% (vs. 6%) 86% 5% 1.6% 0% 0.14% 6% A2 level 91% (vs. 9%) 82% 8% 3% 0.6% 0% 7% B1 level 87% (vs. 13%) 68% 12% 2% 0.3% 0% 2% B2 level 83% (vs. 17%) 79% 16% 2% 0% 0.3% 2% C1 level 82% (vs. 18%) 79% 17.5% 2% 0% 0.8% 1% Native speakers 66% (vs. 34%) 65.5% 33% 0.7% 0% 0% 0.3% L2 learners 87% (vs. 13%) 82% 12% 2% 0.2% 0.5% 3% All participants 85% (vs. 15%) 80% 14% 2% 0.2% 0.5% 3% Table 1.Mean percentage of regular, irregular, no-change, third-person singular of present-simple tense, past progressive, and unclassified forms produced by each language level (the table outline is sourced from Blything et al., 2018). Since this study is concerned with the development of the past-tense system, our analysis focused solely on the investigation of regular and irregular forms, and responses (iii)–(v) were excluded from the analysis. To investigate whether the learner data showed similar outcomes to Albright & Hayes (2003) and Blything et al. (2018), the data were analysed with abinomial linear mixed-effects model (using the 8 Contrary to the approach adopted in Blything et al. (2018), we have decided to include not only irregular forms of the novel words with the most favourable irregular score in Albright and Hayes’s (2003) study (as Blything et al. (2018) did), but also all the other possible irregular past forms cited in Quirk et al. (1985: 104). As aconsequence, the irregular past forms of verbs drit, gude, and nold were homophonous with the stem form analogically to no-change forms such as cut, put, or hit. Such past forms of these three novel words were then coded as “irregular” rather than “no change”.
LUCIE JIRáNKOVá— LUCA CILIBRASI 203 Czech had no effect on their performance. These four language groups thus made no difference between the novel words in the two sets and consequently between the two languages: This suggests that there were no interference effects at play and Czech functioned neither as afacilitator nor as an inhibitor. 4.3 QUALITATIVE ANALYSIS The data obtained also called for qualitative analysis to explore afew interesting features and their changes over the process of second-language acquisition. In the qualitative analysis, attention was paid to (i) different past forms produced by the participants, their mean proportion, and distribution over language levels; (ii) the pronunciation of the bound past-tense morpheme and its changes during L2 development; and (iii) the addition of superfluous morphemes to the past-tense endings. 4.3.1 PRODUCED FORMS Since this study is focused on the development of the past-tense system, it also raises the question of which forms are produced more frequently at each language level, if this production changes over the course of L2 acquisition, and if it is fuelled by any specific strategies. The analysis of past forms showed that at all language levels the production of regular forms highly surpasses the production of irregulars, and that the production of regular forms declines with proficiency, while the production of irregular forms rises (see Figure 3 below). Three occurring types of irregular past forms have been identified in the collected data: 1. an internal change (IC; exemplified by the change of /eɪ/ into /əʊ/ in to chake— choke) Mean reaction times (s) Mean proportion of regular inflections Set (divided by language levels) Similarity to existing irregulars 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% A1 A2 B1 B2 C1 NS A B A B A B A B A B A B Past forms divided by the level 0 10 20 30 40 60 90 100 50 70 80 R IR R IR R IR R IR R IR R IR Mean percentage of production (%) Mean percentage of production (%)Mean percentage of production (%) Types of past forms divided by the level 0 10 20 30 40 50 60 70 IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC Types of pronunciation divided by the level 0 10 20 30 40 70 60 50 90 regular full regular full regular full regular full regular full regular full 80 100 A1 A2 B1 B2 C1 NS A1 A2 B1 B2 C1 NS A1 A2 B1 B2 C1 NS A1 A2 B1 B2 C1 NS Figure 2.Mean reaction times to novel words of the two sets divided by language levels.
204 LINGUISTICA PRAGENSIA 2/2021 2. an internal change accompanied by the addition of the past bound morpheme -d, -t, or -ed (IC + -ed; exemplified by to teep— tept or to bredge— bridged) 3. an irregular no-change form (IR NC; exemplified by to drit— drit or to gude— gude) The investigation of the types of irregular past forms showed that the internal change accompanied by the addition of apast bound morpheme -ed (e.g. to bredge into bridged) seemed to be adominant production pattern at the lower levels of the L2 learners and its use slowly declined with higher proficiency. Similar instances of this type have already been mentioned in the literature. Bybee and Slobin (1982), for example, suggested that the final t/d may serve as an aid in acquiring the vowel change, i.e. as an important clue that the form is in fact expressing the past tense. In the earliest stages, children (and in our case the lower-level L2 students) have the knowledge of the regular -ed rule. Using it does not force them to rely so heavily on the surrounding context (as in the case of asole internal past-tense change, e.g. in sing-sang or break-broke) to distinguish the past tense. The production of the past form both with the internal change and the -t/d suffixation may thus be understood as an intermediate mastery of the internal-change production. L2 speakers may also be influenced by such models as sleep/slept or keep/kept (Rumelhart & McClelland 1986). Rumelhart and McClelland (1986) noted that three of their four responses with this type of past-tense change were verbs ending in /p/, similarly to the sleep/keep models above. They understand this as the participants’ “sensitivity to the regular and subregular patterns of the English past tense” (Rumelhart & McClelland 1986: 34). At the lowest levels, this production was closely followed by irregular no-change forms (e.g. drit— drit); this strategy, too, shows asteady decline with higher proficiency. On the contrary, the internal change (e.g. chake into choke) was not frequent at the lowest levels and gathered noticeable strength with higher proficiency. The distribution is visualized in Figure 4 below: Mean reaction times (s) Mean proportion of regular inflections Set (divided by language levels) Similarity to existing irregulars 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% A1 A2 B1 B2 C1 NS A B A B A B A B A B A B Past forms divided by the level 0 10 20 30 40 60 90 100 50 70 80 R IR R IR R IR R IR R IR R IR Mean percentage of production (%) Mean percentage of production (%) Mean percentage of production (%) Types of past forms divided by the level 0 10 20 30 40 50 60 70 IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC Types of pronunciation divided by the level 0 10 20 30 40 70 60 50 90 regular full regular full regular full regular full regular full regular full 80 100 A1 A2 B1 B2 C1 NS A1 A2 B1 B2 C1 NS A1 A2 B1 B2 C1 NS A1 A2 B1 B2 C1 NS Figure 3.Mean percentage of production as afunction of produced past forms divided into language levels (R refers to aregular past form, IR refers to an irregular past form).
LUCIE JIRáNKOVá— LUCA CILIBRASI 205 The analysis of the no-change past forms (e.g. drit, gude, or nold) then showed that the recurrent no-change novel words have acommon denominator, afinal alveolar phoneme, and this choice may thus be phonetically conditioned. The reason for the no-change past form may be (i) the analogy with other existing no-change forms such as cut, put, or hit, or (ii) the participants’ unintentional feeling that the novel word already ends with an alveolar phoneme and no other past bound morpheme is needed, since the bound morpheme blends with the final alveolar phoneme of the verb stem in their minds. This type of irregular past-tense formation has also been previously mentioned in literature. Stemberger (1981) has shown that inflectional languages commonly avoid adding abound inflectional morpheme to astem that already seems to contain such an affix. This would explain why the unchanged past-tense forms originally ended in an alveolar phoneme. Bybee and Slobin (1982) conclude that it is not accidental that words whose past-tense form remains unchanged end their stem in either t or d (e.g. hit or shed), since the speakers may simply use aschema […t/d] verb/past in which the form of the stem and its past form would be identical. Berko (1958) has also described this phenomenon in her classic study, noting that the tested children showed aprominent tendency not to add the regular past-tense suffix to novel verbs that ended in an alveolar. She understands this finding as an incomplete acquisition of the /ɪd/ allomorph. Bybee and Slobin (1982) counter this interpretation with the claim that such findings have more to do with the phonological shape of the novel form than with the late acquisition of the allomorph. Anisfeld and Gordon (1968) have shown that children are more likely to accept bare novel forms as their own past forms if the final phoneme of the stem shares some features with t or d. Slobin (1971) pointed out that children also tend not to regularize verbs whose past forms remain unchanged from the stem. Consequently, Kuczaj (1978) suggested that children are even more likely to accept unchanged past-forms of such verbs (and reject any such regularized forms, e.g. hitted). Both Berko (1958) and Bybee and Slobin Figure 4.Mean percentage of production as afunction of past-form types divided by language level (IC + -ed stands for an internal change accompanied by the addition of apast bound morpheme, IR NC stands for an irregular no-change form, and IC stands for an internal change). Mean reaction times (s) Mean proportion of regular inflections Set (divided by language levels) Similarity to existing irregulars 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% A1 A2 B1 B2 C1 NS A B A B A B A B A B A B Past forms divided by the level 0 10 20 30 40 60 90 100 50 70 80 R IR R IR R IR R IR R IR R IR Mean percentage of production (%) Mean percentage of production (%)Mean percentage of production (%) Types of past forms divided by the level 0 10 20 30 40 50 60 70 IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC Types of pronunciation divided by the level 0 10 20 30 40 70 60 50 90 regular full regular full regular full regular full regular full regular full 80 100 A1 A2 B1 B2 C1 NS A1 A2 B1 B2 C1 NS A1 A2 B1 B2 C1 NS A1 A2 B1 B2 C1 NS
206 LINGUISTICA PRAGENSIA 2/2021 (1982) then find asignificant difference in this application between children and adults, who add the suffix more consistently. Bybee and Moder (1983) hypothesized that speakers tend to understand the verbs with the -d or -t stem ending as irregular in some respect and therefore supply past-tense forms identical to the stem. In their computer simulation of past-tense production, Rumelhart and McClelland’s model (1986) also produced some strong no-change forms with verbs ending either in -d or -t. It is interesting to note that in the early stages of their simulation, the model was more prone to regularization (i.e. adding -ed to the stem); however, in the later stages, the no-change past-forms appeared as apart of the learning process. The authors also make an interesting note by pointing out that the words ending in t/d in the previous studies were all monosyllabic and it would be beneficial to see if the same phenomenon can be observed also in multisyllabic words such as devote or decide. 4.3.2 PRONUNCIATION OF THE FINAL -ED MORPHEME Another interesting feature appears to be the full pronunciation of the final bound morpheme -ed as /ɪd/— in places where the phonological surrounding asks for aregular /t/ or /d/ pronunciation (after any phoneme except for the alveolar ones11)— for instance, the full pronunciation of daped as /deɪpɪd/ instead of /deɪpt/. The mean percentage of its production was calculated alongside the percentage of aregular pronunciation to see the ratio of the regular vs. full pronunciation. The production was also divided by language level to see if this phenomenon possibly evolves over the process of language acquisition. Avisualization of its distribution over levels can be seen in Figure 5 below. The graphical visualization shows that the mean percentage of the full pronunciation in places where regular pronunciation should be is the highest (and also the most prominent) in the lowest A1 level and its distribution steadily declines with higher proficiency. Not surprisingly, native speakers did not produce any instance of full pronunciation. Similarly, the regular (i.e. correct, not full) pronunciation rises with higher proficiency, reaching almost 100% production at the C1 level. Yet, it is interesting to notice that even the highly proficient C1 participants produced some past forms with the full pronunciation (while the native speakers did not). Another question thus arises of whether its production be phonetically conditioned for L2 learners, i.e. whether the full pronunciation could be triggered by the final phoneme of the preceding stem. After analysing the phonological area preceding the past bound morpheme, i.e.the quality of the final phoneme of the stem, six types of word-final phonemes were identified: plosives (i.e. /p/, /k/, /b/, and /g/— excluding /t/ and /d/); nasals (i.e. /m/, /n/, and /ŋ/); liquids (only /r/, /l/ did not appear); fricatives (i.e. /s/, /z/, /f/, /ʒ/ and /ʃ/); affricates (i.e. /dʒ/ and /tʃ/); and also collectively sibilants (/s/, /z/, /ʒ/, /ʃ/, / dʒ/, and /tʃ/), which overall covered an ample share of full pronunciation. The high 11 Instances of the full pronunciation after dentals were not investigated as full pronunciation and counted under regular pronunciation since the full pronunciation of -ed after dentals is considered phonologically correct.
LUCIE JIRáNKOVá— LUCA CILIBRASI 207 proportion of full pronunciation with stem-final sibilants could be explained by an unintended priming that originates from the uttered sentence preceding the one in the past: for instance, in The baby likes to dize. Look, there it is dizing. Everyday it dizes. So yesterday it…, the full pronunciation of -es in dizes /daɪzɪz/ (which is phonologically correct after asibilant fricative) could have primed the subsequent pronunciation of dized as /daɪzɪd/. The participants might have simply re-applied the full pronunciation to awrong phonological position. The other phonological surroundings (i.e.plosives, nasals, and liquid /r/) also led to full pronunciations, though more rarely. These instances are very interesting in that they cannot be easily explained referring to repetition; one of the possible explanations is that they are caused by hypercorrection. The L2 speaker may generally believe, through amisunderstanding of the rules, that the full form is more preferable or formal (see Selinker 1972). Another possible explanation is that the particular consonantal sequence may be hard to pronounce for the L2 speaker (since it is phonotactically illegal in their L2), and the learners may resort to the full form to make the pronunciation easier. 4.3.2 INSERTED S/T/D PHONEMES In the process of transcribing the produced past-forms, an interesting and re-occurring phenomenon has arisen— the insertion of -s-, -t-, or -dbetween the stem and the bound morpheme of the past tense, e.g. glipsed or pankted. Its frequency is relatively low but there seems to be aprogressive development from ahigher frequency at the lowest levels to aminimal frequency at the C1 level. However, again, also the C1level participants seem to be inserting the /s/ phoneme in between the stem and the bound morpheme on some occasions, and it is thus interesting to look at apossible stem-final phonetic conditioning in the L2 learners (since the native speakers produced none of these insertion forms). Similarly to the full pronunciation phenomenon, four stem-final phoneme types were identified: plosives, nasals, liquid /r/, and Mean reaction times (s) Mean proportion of regular inflections Set (divided by language levels) Similarity to existing irregulars 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% A1 A2 B1 B2 C1 NS A B A B A B A B A B A B Past forms divided by the level 0 10 20 30 40 60 90 100 50 70 80 R IR R IR R IR R IR R IR R IR Mean percentage of production (%) Mean percentage of production (%) Mean percentage of production (%) Types of past forms divided by the level 0 10 20 30 40 50 60 70 IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC IC + -ed ICIR NC Types of pronunciation divided by the level 0 10 20 30 40 70 60 50 90 regular full regular full regular full regular full regular full regular full 80 100 A1 A2 B1 B2 C1 NS A1 A2 B1 B2 C1 NS A1 A2 B1 B2 C1 NS A1 A2 B1 B2 C1 NS Figure 5.Mean percentage of production of regular vs. full pronunciation of the morpheme -ed as afunction of the language level.
208 LINGUISTICA PRAGENSIA 2/2021 fricatives. There seems to be arather prominent frequency of stem-final plosives and nasals that might trigger this specific phoneme insertion. The insertion of the -sphoneme after these phonemes could be potentially explained by yet another unintentional priming (e.g. glipsed after Everyday he glips) when the participant might have incorrectly used the inflected form glips as the stem. This insertion might be also implicitly conditioned phonetically by anumber of similar phonemes attached one after another. In such cases, the speaker may use such insertion forms to differentiate the phonemes, aphenomenon known as dissimilation. Such instances have also been previously described in classic literature as double past markers (in the case of -tinsertions) and using affixes for both present and past (in the case of -sinsertions). In their computer simulation, Rumelhart and McClelland (1986), too, found the double past marking on seven responses to the input word. Even though they attribute such forms to errors that children and adults occasionally make, they also notice that the doubling occurred for those verbs whose stem ended in /p/ or /k/ and whose correct past-tense form should be created by the addition of/t/ and whose stem ended in /p/ or /k/ (e.g. /dript’d/). Interestingly enough, Berko (1958) also noted that several of the tested children retained the 3rd-person-singular /s/ or /z/ phoneme in addition to the past-tense suffix, as in /spoʊzd/. 5 DISCUSSION AND CONCLUSION The results of this experiment extend previous research (e.g. Albright & Hayes 2003, Ambridge 2010, and Blything et al. 2018) to second-language learners of English and bring new insights into the acquisition of L2 morphology. The analysis of the produced forms has shown that for the native speakers the likelihood of averb being produced in aregular past-tense form is negatively associated with its phonological similarity to existing irregular verbs, as predicted by the single-route model. However, L2 learners have shown lesser dependency on the verbal similarity to irregulars when deciding on its past form. Further analysis has shown that the L2 acquisition of the English past-tense is characterized by adevelopment from the mastery of mechanistic rules (in line with the dual-route mechanism at the A1, A2, and B1 levels) to the refinement of their application when the learner starts spotting analogical patterns of existing verbs (in line with the single-route analogical mechanism at the B2 and C1 levels, similarly to native speakers). The second-language speakers thus come closer to native speakers with the higher proficiency of B2 and C1 levels, while lower levels seem to adopt an alternative strategy. The analysis of the reaction times has also shown that, for the lowest language levels, Czech implicitly functions as afacilitator in their performance, with participants being faster with novel words that are phonotactically legal in both Czech and English. L2 speakers with higher proficiency and native speakers, instead, did not show any effect of Czech phonotactic probabilities. Finally, the qualitative analysis has shown some specific pronunciation issues, such as the use of full pronunciation of bound morphemes where asingle phoneme is expected, or the insertion of additional phonemes in the bound morpheme, possibly due to priming. Recent production tasks using novel words (e.g. Albright & Hayes
LUCIE JIRáNKOVá— LUCA CILIBRASI 209 2003; Ambridge 2010; and Blything et al. 2018) have provided some evidence that the production of the English past tense shows signs of analogical processes associated with the single-route model, without the universal application of the default pasttense rule. So far, the role of the similarity to existing (ir)regular verbs in the production of novel verbs has been observed both in native English-speaking children (Ambridge 2010; Blything et al. 2018), in adults (Albright & Hayes 2003) and also in L2 learners of English (Lee 1994; Beck 1997; Ellis & Schmidt 1998; Murphy 2000; Agathopoulou 2009; Cuskley et al. 2015). Our study further investigates whether evidence for the analogical morphological productivity extends to second-language learners of English with Czech as L1 and whether the process behind the production differs among various language levels, starting with aproficiency sample that has just started learning English and moving to the language sample closest to English native speakers. The data suggest that the lowest proficiency groups (A1, A2, and B1) generate regular past forms by the default rule of adding -ed to the novel verb’s stem, seemingly irrespective of the novel verb’s phonological similarity to existing (ir)regular verbs. In contrast, the more proficient groups (B2 and C1) and native speakers of English rely on amorphological system in which the production of regular past forms depends on the novel verb’s phonological similarity to existing verbs. These findings extend previous studies that spoke in support of analogical processes associated with morphological productivity (e.g. Albright & Hayes 2003, Ambridge 2010, or Blything et al. 2018) to proficient second-language learners of English. Blything et al. (2018: 14) use their findings to show that “overregularization errors made by children […] need not be attributed to arule-based mechanism […] and, indeed, are better explained in terms of analogy across stored exemplars.” Similarly, our results suggest that this claim could be extended to the L2 learners at the B2 and C1 levels. In their study, Blything et al. (2018) make athick dividing line between the effects of the singleand dual-route model, trying to distinguish the former from the latter. However, given the results of our research that point to agradual progression from rule application to the use of analogy, and in line with previous research (e.g.Cilibrasi et al. 2019), this paper attempts to propose aless dividing conclusion: redundancy. Redundant models (Schreuder et al. 1999) hold that inflectional processing generally involves two types of analyses that seem to operate simultaneously: single- (rote-based parsing) and dual-route mechanisms (rule-based parsing). These analyses operate in parallel, but each system has avarying importance connected to how frequently agiven item is generally used: “frequent items tend to be parsed with the rote system, while infrequent items tend to be parsed with the rule system” (Cilibrasi et al. 2019: 769). Amechanism of this kind may be at play in second-language learners of English. Albright and Hayes (2003: 120) themselves argue for anew (third) “model of morphology that makes use of multiple, stochastic rules”, and our results suggest that these rules may equally apply to L2 learners, with the combined contribution of simple rules and generalizations based on analogy. These two mechanisms may be present at all levels but have varying importance depending on proficiency. In conclusion, this experiment has enabled us to look more closely into the processes underlying morphological productivity in second-language learners of Eng-
210 LINGUISTICA PRAGENSIA 2/2021 lish and also into its potential development with higher proficiency. The most important contribution of this study lies in showing that the L2 acquisition of past-tense morphology is characterized by agradual progression from the application of default rules to the application of analogical patterns. Acknowledgements This project has been financially supported by aSpecific Academic Research grant (SVV), “Jazyk anástroje pro jeho zkoumání”. The authors would also like to thank Albright and Hayes and Blything et al. who agreed to the use of their existing stimuli, frame sentences, and task design for the purposes of this study. REFERENCES Agathopoulou, E. (2009) The regular-irregular dissociation in non-native English past tense. Selected Papers from the 18th ISTAL, 37–47. Albright, A. and B.Hayes (2003) Rules vs. analogy in English past tenses: Acomputational/experimental study. Cognition 90/2, 119–161. https://doi. org/10.1016/S0010-0277(03)00146-X Ambridge, B. (2010) Children’s judgments of regular and irregular novel past-tense forms: New data on the English past-tense debate. Developmental Psychology 46/6, 1497–1504. https://doi.org/10.1037/a0020668 Ambridge, B. (2019) Against stored abstractions: Aradical exemplar model of language acquisition. First Language, 1–51. https://doi. org/10.1177/0142723719869731 Ambrus, L., C.Sunnqvist, M.Asp, S.Westling and A. Westrin (2020) Coping and suicide risk in high risk psychiatric patients. Journal of Mental Health 29/1, 27–32. Anisfeld, M. and M.Gordon (1968) On the psychophonological structure of English inflectional rules. Journal of Verbal Learning and Verbal Behavior 7 973–979. https://doi. org/10.1016/S0022-5371(68)80055-6 Baayen, R.H., R.Piepenbrock and L.Gulikers (1995) The CELEX lexical database. Baayen, R.H., D.J.Davidson and D.M.Bates (2008) Mixed-effects modeling with crossed random effects for subjects and items. Journal of Memory and Language 59/4, 390–412. https://doi.org/10.1016/j.jml.2007.12.005. Barr, D.J., R.Levy, C.Scheepers and H.J.Tily (2013) Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language 68/3, 255–278. https://doi.org/10.1016/j. jml.2012.11.001 Bates, D., M.Mächler, B.Bolker and S.Walker (2015) Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software 67/1, 1-51. https://doi.org/10.18637/jss.v067.i01 Beck, M.-L. (1997) Regular verbs, past tense and frequency: Tracking down apotential source of NS/NNS competence differences. Second Language Research 13/2, 93–115. Berko, J. (1958) The Child’s Learning of English Morphology. Word 14/2–3, 150–177. https://doi. org/10.1080/00437956.1958.11659661. Blything, R.P., B.Ambridge and E.V.M.Lieven (2018) Children’s Acquisition of the English Past-Tense: Evidence for aSingle-Route Account From Novel Verb Production Data. Cognitive Science 42, 1–19. https:// doi.org/10.1111/cogs.12581. Bybee, J.L. and D. I. Slobin (1982) Rules and Schemas in the Development and Use of the English past Tense. Language 58/2, 265–289. https://doi.org/10.2307/414099. Bybee, J.L. and C.L.Moder (1983) Morphological Classes as Natural Categories. Language 59/2, 251. https://doi. org/10.2307/413574. Cilibrasi, L., Stojanovik, V., Riddell, P., & D.Saddy (2019) Sensitivity to Inflectional
LUCIE JIRáNKOVá— LUCA CILIBRASI 211 Morphemes in the Absence of Meaning: Evidence from aNovel Task. Journal of Psycholinguistic Research 48/3, 747–767. https:// doi.org/10.1007/s10936-019-09629-y. Cuskley, C., Colaiori, F., Castellano, C., Loreto, V., Pugliese, M., and F. Tria (2015) The adoption of linguistic rules in native and non-native speakers: Evidence from aWug task. Journal of Memory and Language 84, 205–223. Čechová, M. (1996) Čeština: Řeč ajazyk. Praha: ISV nakl. Ellis, N.C. and R.Schmidt (1998) Rules or associations in the acquisition of morphology? The frequency by regularity interaction in human and PDP learning of morphosyntax. Language and Cognitive Processes 13/2–3, 307–336. Kuczaj, S.A. (1978) Why do children fail to overgeneralize the progressive inflection? Journal of Child Language 5/1, 167–171. https:// doi.org/10.1017/S0305000900002026. Kuznetsova, A., P.B.Brockhoff and R.H.B.Christensen (2015) LmerTest: Tests in Linear Mixed Effects Models. Journal of Statistical Software 2/13, 1–26. Langacker, R. (2000) Adynamic usage-based model. In: Kemmer, S. and M.Barlow (eds), Usage Based Models of Language, 1–64. Stanford: CSLI Publications. https://ci.nii.ac.jp/ naid/10030056583/. Lee, B.P.H. (1994) Generalization of regular and irregular inflectional patterns: Towards alanguage processing model for both native and non-native speakers of English. Cambridge University Working Papers in English and Applied Linguistics 1, 15–33. Moore, D.S. and G.P.McCabe (1999) Introduction to the practice of statistics (3rd ed). New York: W.H. Freeman. Murphy, V.A. (2000) Inflectional morphology and second language learning systems: An investigation of the dual-mechanism model and L2 morphology. Unpublished PhD thesis. Montreal: McGill University. Nosofsky, R.M. (1990) Relations between exemplar-similarity and likelihood models of classification. Journal of Mathematical Psychology 34/4, 393–418. https://doi. org/10.1016/0022-2496(90)90020-A. Pérez, A., H.S.S.L.Joseph, T.Bajo and K.Nation (2016) Evaluation and revision of inferential comprehension in narrative texts: An eye movement study. Language, Cognition and Neuroscience 31/4, 549–566. https://doi.org/10. 1080/23273798.2015.1115883. Pinker, S. and M.T.Ullman (2002) The past and future of the past tense. Trends in Cognitive Sciences 6/11, 456–463. https://doi.org/10.1016/ S1364-6613(02)01990-3 Pitt, M.A. (1998) Phonological processes and the perception of phonotactically illegal consonant clusters. Perception & Psychophysics 60/6, 941–951. https://doi.org/10.3758/ BF03211930. Prasada, S. and S.Pinker (1993) Generalisation of regular and irregular morphological patterns. Language and Cognitive Processes 8/1, 1–56. https://doi. org/10.1080/01690969308406948. Quirk, R., S.Greenbaum, G.Leech and J.Svartvik (eds) (1985) AComprehensive grammar of the English language. London: Longman. R Core Team (2013). R: Alanguage and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. http://www.R-project.org/ Ramscar, M. (2002) The role of meaning in inflection: Why the past tense does not require arule. Cognitive Psychology 45/1, 45–94. https://doi.org/10.1016/S00100285(02)00001-4. Rumelhart, D.E., and J. L. McClelland (1986) On learning the past tenses of English verbs. In: McClelland, J.L., D.E.Rummelhart and the PDP Research Group (eds), Parallel distributed processing: Explorations in the microstructure of cognition. Volume 2: Psychological and biological models, 216–271. Cambridge (MA): The MIT Press. Selinker, L. (1972) Interlanguage. International Review of Applied Linguistics 10, 209–231. Schreuder, R., N. de Jong, A.Krott and H.Baayen (1999) Rules and rote: Beyond the linguistic either-or fallacy. Behavioral and Brain Sciences
212 LINGUISTICA PRAGENSIA 2/2021 22/6, 1038–1039. https://doi.org/10.1017/ S0140525X9947222X Slobin, D.I. (1971) Developmental Psycholinguistics. In W.O.Dingwall (ed.), Asurvey of linguistic science. College Park, MD: University of Maryland Press. Stemberger, J.P. (1981) Morphological Haplology. Language 57/4, 791–817. https://doi. org/10.2307/414242 Šturm, P. and D. Lukeš (2017) Fonotaktická analýza obsahu slabik na okrajích českých slov vmluvené apsané řeči. Slovo aSlovesnost, 78/2, 99–118. Westermann, G. and N.Ruh (2012) Aneuroconstructivist model of past tense development and processing. Psychological Review 119/3, 649–667. https://doi.org/10.1037/ a0028258. Lucie Jiránková Department of English Language and ELT Methodology Faculty of Arts, Charles University, Prague nám. Jana Palacha 2, 11638 Prague ORCID ID: 0000-0002-1480-6860 [email protected] Luca Cilibrasi Department of English Language and ELT Methodology / Department of General Linguistics Faculty of Arts, Charles University, Prague nám. Jana Palacha 2, 11638 Prague ORCID ID: 0000-0001-8262-0542 [email protected]