CC BY
22
Лингвистика М.Д. Васильева
Московский государственный университет им. М.В. Ломоносова, 119991 г. Москва, Российская Федерация
Задача постепенной демаскировки как метод исследования зрительного распознавания слов: случай падежных форм существительных1
Роль падежа в восприятии существительных остается одной из нерешенных проблем в литературе, посвященной зрительному распознаванию слов. Так, например, идут споры о том, одинаковым ли образом устроена обработка различных падежных форм. В настоящей работе мы обращаемся к обработке падежного словоизменения русских существительных. Цель состояла в том, чтобы оценить совокупное влияние таких факторов, как падеж, неоднозначность и контекст, на зрительное восприятие существительных в единственном числе. Мы исследуем ранние стадии распознавания слова, используя метод постепенной демаскировки. Задача постепенной демаскировки дает результаты, сопоставимые с результатами, полученными ранее в задаче лексического решения. Это указывает на то, что различия в обработке словоформ разных падежей носят универсальный характер и не являются специфичными для конкретного задания.
Ключевые слова: зрительное распознавание слов, морфологическая обработка, задача постепенной демаскировки слов, русское словоизменение, падеж, омонимия.
1 Автор выражает благодарность Л. Лычко и А. Кулаковой за помощь в сборе данных, а также М.В. Фаликман и О.В. Федоровой за ценные замечания и комментарии.
M.D. Vasilyeva
Lomonosov Moscow State University, Moscow, 119991, Russian Federation
Progressive demasking task
as a method to study visual word recognition:
the case of inflected nouns2
The role of case in noun processing remains one of the unsolved issues in the visual word recognition literature. For instance, it is disputed whether different case forms have equal processing cost. The present work focuses on the processing of Russian nominal inflection. The goal is to assess how joint factors such as case, ambiguity and context affect visual recognition of nouns in the singular form. Here we explore the early stages of word identification using the progressive demasking task. Progressive demasking reveals comparable results to the lexical decision task, supporting the idea that case processing differences are general and not task-specific.
Key words: visual word recognition, morphological processing, progressive demasking task, Russian inflection, case, ambiguity.
1. Introduction
Research on morphological processing constitutes a vast domain in the visual word recognition literature. The core questions are how the morphological structure is encoded in the mental lexicon and how morphological complexity affects lexical access.
In the present study, we extend our work on the processing of Russian nominal inflection [Vasilyeva, to appear]. The prior study involved only the lexical decision task (LDT), a method that is widely employed in the visual word recognition literature (see, e.g. [Baayen, 2014]). In the LDT, the participant is presented with sequences of letters and has to judge whether they are legal words. Nevertheless, the LDT has several drawbacks. Response latencies measured in the LDT reflect not only the time course of the lexical access, but also the time spent for decision making [Diependaele et al., 2012]. Furthermore, the processing of word stimuli may
2 The author would like to thank Leonora Lytchko and Anna Kulakova for their assistance with participant recruitment as well as Maria Falikman and Olga Fedorova for their comments on the earlier version of this paper.
be affected by the properties of constructed non-words (see, e.g. [Baayen, 2014]). Thus, Vasilyeva (to appear)'s findings might be confounded.
One way of moving beyond the LDT preponderance, as suggested in [Baayen, 2014], is to dispense with the investigation of isolatedly presented words and shift to the recognition of context-embedded words. We acknowledge the ecological validity of this approach, yet it is not particularly suitable for studying the role of case in word recognition. Effects observed for wordforms presented in isolation do not show up for wordforms embedded in sentences (e.g., [Hyona, Vainio, 2002]). We believe that the traditional LDT research might still give considerable insight, if it is complemented with another method that also involves the presentation of the stimulus in isolation, but lacks the decision making component and does not require non-word construction.
One such task is the progressive demasking task (PDT) developed by Grainger and Segui (1990). The PDT is a variant of the continuous threshold latency identification [Feustel et al., 1983] and might be regarded as a visual modality analogue of the gating task [Grosjean, 1980] that is used in the auditory word recognition research. In the PDT, the stimulus (signal) and the mask (noise) are rapidly presented one after the other in cycles of constant length. With each successive cycle, the signal becomes more prominent (the presentation time of the stimulus increases), while the noise is slowly reduced (the presentation time of the mask gets shorter). Under these conditions, the participant has an illusion that the word is gradually emerging from the mask. The participant's task is to press a button as soon as s/he identifies the word.
Since the word recognition in the PDT is slowed down or rather stretched out, starting with [Grainger, Segui, 1990], this method was assumed to capture early stages of visual word recognition and, thus, to be particularly appropriate for the investigation of perceptual properties (such as length, surface (wordform) frequency, phonological and orthographic neighborhood). Subsequent research revealed its sensitivity to semantic characteristics of the word (e.g., [Yap et al., 2012]) and to its morphological complexity (e.g., [Laine et al., 1999]). Importantly, the PDT, unlike the LDT, is robust to postlexical (semantic) interference: for instance, the effect of the morphological family size (by a morphological family we understand all the words sharing the same root), which is usually interpreted as being semantic in nature, is present in the LDT, but absent in the PDT [Schreuder, Baayen, 1997].
Though word identification in the PDT happens in the visual modality, parallels to auditory word processing are also drawn [de Jong et al., 2000]: as in both cases the word unfolds over time, multiple candidates matching the incoming signal are activated and evtntually only one competitor is chosen.
In this vein, when the results of the auditory and visual LDT diverge, the PDT patterns together with the auditory LDT [Baayen et al., 2003].
If we turn to nominal recognition, case forms of the same lexeme are the most likely competitors in the identification process. The nominative as the citation form is detected the fastest ([Laine et al., 1999] for Finnish), but whether other (oblique) case forms are identified with equal speed is so far unknown. The nominative superiority effect is also observable in the LDT (see, e.g. [Feldman, Fowler, 1987] for Serbian; [Laine et al., 1999] for Finnish; [Gor et al., 2017] for Russian), while comparisons of oblique case processing give contradictory results (see, [Feldman, Fowler, 1987] for a null effect in Serbian; [Clahsen et al., 2001] for an effect in German, etc.).
In [Vasilyeva, to appear], reproducing once more the nominative superiority effect, we argue for differences in Russian oblique case processing (here we will discuss only the data for singular). Two inflectional classes of Russian nouns (feminine -a and masculine -0) were considered in the LDT (case endings are provided in Table 1). According to our findings, oblique wordforms can be divided at least into two groups: instrumental and -u wordforms ('acc.f'3 and 'dat.m') are processed faster than -e wordforms ('dat/loc.f' and 'loc.m'); feminine genitive belongs to the easier group, while masculine genitive - to the more difficult group. Moreover, feminine nouns are processed faster than masculine nouns in instrumental and the -e form ('dat/loc.f' and 'loc.m').
Table 1
Case endings of Russian feminine -a and masculine -0 nouns in the singular form
NOM GEN DAT ACC INS LOC
Feminine -a -e -u -oJ -e
Masculine -0 -a -u -0 -om -e
The objective of the present study is to explore the potential of the PDT in the investigation of the Russian case processing. Additionally, we want to estimate whether inflectional and relative entropy measures play the same role in the PDT as in the LDT. The inflectional entropy encodes the amount of information that is proper to the lexeme's paradigm, while the relative entropy shows how dissimilar two frequency distributions are -that of the target word and that of its inflectional class; the smaller this
3 Abbreviations: f - feminine, acc - accusative, dat - dative, gen - genitive, ins - instrumental, loc - locative, m - masculine, nom - nominative.
dissimilarity is, the easier it is to process the noun (e.g., [Milin et al., 2009] for Serbian nouns; [Vasilyeva, to appear] for Russian feminine nouns). Formulas for calculating inflectional and relative entropy are provided in (1) and (2), respectively, where the wordform w. belongs to the paradigm P of a lexeme w, f stands for frequency, e. stands for the inflectional exponent, IC stands for the inflectional class.
(1) H (P >--Z fW •
w f ( w) f (w) . .. . f (w1) f (w1 )/f (w)
(2) D (PIC ) = -£ fwv lQg2 f( V f () •
v ' Wi„P f (w) f (e, Vf (e)
To our knowledge, the impact of these measures on continuous word identification was not investigated. In principle, both measures should be associated with later stages of word processing similarly to the morphological family size effect, and their influence, thus, will not emerge in the PDT.
2. Method
Participants. 48 undergraduate or postgraduate students (32 female, 16 male), aged 17-28 years old (mean age 20), all Russian native speakers, right-handed, with normal or corrected to normal vision, took part in the study.
Stimuli. We used the same stimulus set as in [Vasilyeva, to appear]: 54 feminine and 54 masculine base nouns matched for lemma frequency.4 All stimuli did not undergo any stem alternations and had fixed stress on the stem. Length in the nominative form differed from 4 to 6 letters (each group comprised one third of words with each length). Characteristics of the stimuli are presented in Table 2.
4 We used the following words in our experiment (lemma frequency counts are given in brackets):
- feminine nouns (mean: 47.39): anketa 'questionnaire' (14.4), arfa 'hap' (2.6), astra 'aster' (3.8), aura 'aura' (5.1), beseda 'conversation' (87.5), bukva 'letter. character' (63.5), data 'date' (49.5), doza 'dose' (22.4), dyuna 'dune' (2), fleita 'flute' (5.8), gazeta 'newspaper' (237.5), gitara 'guitar' (22.2), kareta 'carriage' (9.4), karta 'map' (103), kassa 'cashier's desk' (20.9), klumba 'flower-bed' (8.7), klyaksa 'blot' (4.5), kofta 'jacket' (7.7), lampa 'lamp' (34), lapa 'paw' (39.7), lenta 'ribbon' (35.9), lira 'lyre' (8), lyustra 'lustre' (9.9), mera 'measure' (284.3), minuta 'minute' (344.2), moneta 'coin' (17.5), norma 'norm' (111.3), orbita 'orbite' (15),pal'ma 'palm tree' (14.3),pasta 'paste' (6.3), pochva 'soil' (56.2), poza 'pose' (29.8), raketa 'rocket' (62.9), rama 'frame' (21.2), rana 'wound' (29.4), rasa 'race' (5.9), rifma 'rhyme' (8.5), roza 'rose' (42.7), shakhta 'pit' (20.7), shina 'tire' (15.3), shirma 'folding-screen' (5.3), shkola 'school' (316), shlyapa 'hat' (34.2), shuba 'furcoat' (18.7), shvabra 'mop' (3.4), summa 'sum' (130.6), trassa 'route' (32.5), travma 'trauma' (19.6), tsifra 'numeric' (62.2), tsitata 'citation' (21.5), tykva 'pumpkin' (5), vaza 'vase' (14.3), yakhta 'yacht' (9.5), yurta 'yurt' (2.7);
Table 2
Word stimulus characteristics
Feminine M (SD) Masculine M (SD)
Length in letters nominative singular all singular case forms 5.0 (0.8) 5.2 (0.9) 5.0 (0.8) 5.8 (1.1)
Length in syllables nominative singular all singular case forms 2.2 (0.4) 2.2 (0.4) 1.8 (0.6) 2.4 (0.7)
Mean Levenstein distance to the nearest 20 lexeme-neighbors5 2.0 (0.4) 2.2 (0.4)
Lemma frequency6 47.4 (76.5) 47.4 (68.9)
Wordform frequency in singular 7.3 (19.3) 6.9 (16.7)
Inflectional entropy7 2.5 (0.2) 2.5 (0.3)
Relative entropy 0.3 (0.2) 0.3 (0.3)
Nouns were presented in all six cases in singular. A latin-square design was employed with the number of lists corresponding to the number of case forms, so that no participant saw the same lexeme twice. As our words were presented without context, case labels for ambiguous endings (-y and -e for feminine
- masculine nouns (mean: 47.37): al'bom 'album' (23.7), ananas 'pineapple' (3.6), aromat 'aroma' (22.9), aspect 'aspect' (35.6), atom 'atom' (20.5), banan 'banana' (7.3), baton 'loaf (of bread)' (5.3), bufet 'buffet' (20), buton 'bud' (4.6), desert 'dessert' (4), diplom 'diploma' (25.8), divan 'sofa' (60.1), dzhip 'jeep' (14.7), fontan 'fountain' (18.4), frukt 'fruit' (21.6), gimn 'hymn' (14.8), ideal 'ideal' (36), kanat 'rope' (9), kapriz 'caprice' (7.1), kedr 'cedar' (6.1), khalat 'bathrobe' (36.1), klad 'treasure' (7.5), komod 'dresser' (5.2), kontur 'contour' (15.3), kostyum 'costume. suit' (81.3), kurort 'resort' (12.8), metall 'metal' (57.5), moment 'moment' (306.8), nrav 'temper' (17.8), ofis 'office' (34.1),period 'period' (204.2), plan 'plan' (235.3), pled 'plaid' (4.8), reis 'flight. voyage' (22), remont 'reparation' (64.2), ritm 'rhythm' (30.6), romb 'rhombus' (1.8), rulon 'roll' (4.3), servis 'service' (14.6), sezon 'season' (69.2), shram 'scar' (10.7), shtraf 'forfeit' (32.3), simvol 'symbol' (46.4), sous 'sauce' (10.8), syuzhet 'storyline' (56.6), teatr 'theater' (305.3), tekst 'text' (146.2), temp 'tempo' (49), tovar 'item of goods' (115.5), tsikl 'cycle' (43.6), virus 'virus' (106.5), vulkan 'volcano' (6), yarus 'tier. layer' (6.5), zhanr 'genre' (36).
5 It was calculated in the vwr package [Keuleers, 2013].
6 The source of lexeme frequency counts was the frequency dictionary [Lyashevskaya, Sharoff, 2009]. Wordform frequency was measured in the Russian national corpora (www.ruscorpora.ru). To avoid zero frequencies, before the statistical analysis, we added one to all counts, as proposed in [Brysbaert, Diependaele, 2013].
7 In order to calculate these measures, we took exponent frequency from the database [Slioussar, Samojlova, 2014].
nouns, -0 for masculine nouns) are somewhat arbitrary. Hence, we do not expect any differences between locative and dative -e, nor nominative and accusative -0, and we use the mean over the two conditions in the subsequent analyses of contrasts.
Procedure. Each participant was tested individually and completed only one of the six experimental lists. Stimuli were presented using the PsychoPy software [Peirce, 2009]. In each trial, the mask (a row of hash marks) and the target stimulus were presented sequentially in cycles. The duration of the cycle was held constant (210 ms). In the first cycle the duration of the mask was 195 ms, and the duration of the stimulus was 15 ms. In subsequent cycles, the duration of the mask decreased by 15 ms, and the duration of the stimulus increased by 15 ms. The cycles continued until the participant hit the spacebar indicating that s/he has recognized the word. After that, s/he had to type the word so that we could check the accuracy of identification.
3. Results
Accuracy data were not analyzed (incorrect responses constituted less than 5% of the data). Prior to the analysis of reaction times (RTs), incorrect and too slow (> 3000 ms) responses were removed. We applied logtransformation to reduce the positive skew. Remaining outliers were cut off via interquartile trimming by participants, items (lexemes), gender and case.8 Raw RTs to nouns in different cases are provided in Table 3.
Linear mixed-effects modeling for the RT analysis was implemented in the package lme4 [Bates et al., 2014] in the statistical software R [R Core Team, 2014]. T-values, /»-values and standard errors were determined in the package lmerTest [Kuznetsova et al., 2015]. Fixed and random effects were included only if they significantly improved the model's fit in a backward stepwise model selection procedure. The best model was selected by applying Chi-square log-likelihood ratio tests with regular maximum likelihood parameter estimation.
Subject and lexeme were treated as random effects. Trial order (z-trans-formation on log numbers) was included to account for longitudinal task effects such as fatigue or habituation; experimental list was included to rule out potential counterbalancing issues. Log-transformed lexeme and wordform frequency, length in letters and syllables, mean Levenstein distance to the nearest 20 lexeme-neighbors, inflectional and relative entropy measures were additionally included as covariates. We transformed all counts except for trial into five principal components explaining 93.5% of the variance,
8 This method presupposes that only those RTs are kept which fall in the following range: Q1 - (2.5 X IQR) < RT < Q3 + (2.5 x IQR), where Q1 stands for first quartile, Q3 for third quartile, and IQR = Q3 - Q1 for interquartile range.
in order to avoid potential multicollinearity [Baayen, 2008]. The first principal component (PC1) captured orthographic characteristics of the stimulus. The second component (PC2) is inversely related to frequency. The third component (PC3) is inversely related to the relative entropy and positively related to the inflectional entropy.
Table 3
Mean RTs to feminine and masculine nouns in different cases (SD is provided in brackets)
Feminine Masculine
Nominative 1518 (247) 1536 (256)
Accusative 1561 (284)
Genitive 1539 (279) 1602 (297)
Instrumental 1559 (270) 1611 (311)
Dative 1611 (290) 1592 (269)
Locative 1642 (314)
Paired contrasts for case and gender were effectuated in the package lsmeans [Lenth et al., 2016], /»-values were FDR-adjusted [Benjamini, Hochberg, 1995].
The final model included subject and lexeme as random effects and the following fixed effects: trial (x2(1) = 71.47, p < 0.001), PC2 (x2(1) = 8.11, p = 0.004), case (x2(10) = 142.63, p < 0.001), gender (x2(6) = 32.26, p < 0.001) and a case by gender interaction (x2(5) = 29.99, p < 0.001) (see Table 4). All other predictors and interactions turned out to be insignificant. Trial had a facilitative effect on RTs (B = -0.016, /(4819) = -8.48,p < 0.001). PC2 (inversed frequency) had an inhibitory effect on RTs (B = 0.008, /(148) = 2.86, p = 0.005, respectively).
Paired contrasts analysis (Table 5) revealed significantly lower RTs to nominative wordforms than oblique wordforms of both noun types (this is not confirmed for feminine genitive wordforms) and several differences between oblique case forms, see (3).
(3) a. feminine nouns:
genitive ~ accusative ~ instrumental < -e 'dat/loc.f b. masculine nouns:
dative ~ instrumental ~ genitive, instrumental < locative, dative, genitive < locative
Table 4
Final model for RTs
Random effects:
Groups Name Variance SD
Subject (Intercept) 0.001 0.036
Lexeme (Intercept) 0.013 0.113
Residual 0.016 0.127
Fixed effects:
B SE df í-value Pr (>| í |)
(Intercept) 7.311 0.018 69 403.95 <.001
Trial -0.016 0.002 4819 -8.48 <.001
PC2 0.008 0.003 148 2.86 .005
Case[Acc] 0.026 0.009 4871 2.96 .003
Case[DAT] 0.061 0.009 4811 6.89 <.001
Case[GEN] 0.017 0.009 4835 1.94 .052
Case[iNs] 0.023 0.009 4789 2.63 .009
Case[Loc] 0.053 0.009 4812 5.98 <.001
Gender[M] 0.014 0.011 439 1.26 .210
Case[Acc] : Gender[M] -0.030 0.013 4838 -2.39 .017
Case[DAT] : Gender[M] -0.028 0.013 4854 -2.22 .026
Case[GEN] : Gender[M] 0.020 0.013 4835 1.58 .114
Case[iNs] : Gender[M] 0.017 0.012 4782 1.37 .171
Case[Loc] : Gender[M] 0.006 0.013 4811 0.51 .611
Feminine -e wordforms ('dat/loc.f') are identified significantly later than feminine genitive, accusative and instrumental wordforms; the latter do not differ between each other.
Masculine locative wordforms are identified later than genitive and dative wordforms; there is only a tendency for the masculine locative to be identified later than the instrumental. Masculine genitive, dative and instrumental wordforms, similar to the respective feminine wordforms do not differ among themselves.
Participants identified feminine instrumental wordforms significantly earlier than masculine instrumental wordforms; all other contrasts for gender were non-significant.
Table 5
Results of the paired contrasts analysis (analyses with p < 0.05 are given in bold)
A df t P
GEN -0.017 4835 -1.94 0.090
NOM vs. ACC -0.026 4871 -2.95 0.008
INS -0.023 4789 -2.63 0.018
-e 'dat/loc' -0.057 4821 -7.45 < 0.001
F ACC -0.009 4819 -0.10 0.411
GEN vs. INS -0.006 4879 -0.63 0.596
-e 'dat/loc' -0.040 4881 -5.08 < 0.001
ACC vs. INS 0.003 4839 0.35 0.739
-e 'dat/loc' -0.031 4818 -3.98 < 0.001
INS vs. -e 'dat/loc' -0.034 4789 -4.39 < 0.001
GEN -0.039 4834 -5.10 < 0.001
-0 'nom/acc' vs. DAT -0.035 4846 -4.54 < 0.001
INS -0.042 4779 -5.51 < 0.001
LOC -0.061 4785 -7.97 < 0.001
M DAT 0.004 4880 0.47 0.694
GEN vs. INS -0.003 4817 -0.33 0.739
LOC -0.022 4848 -2.49 0.026
DAT vs. INS -0.007 4839 -0.80 0.499
LOC -0.026 4808 -2.96 0.008
INS vs. LOC -0.019 4788 -2.17 0.056
feminine vs. masculine -0.012 105 -1.49 0.200
-oj ins.f' vs. -om 'ins.m' -0.031 449 -2.77 0.014
-e 'dat/loc.f' vs. 'loc.m' -0.016 330 -1.59 0.173
-u 'acc.f' vs. 'dat.m' -0.021 441 -1.84 0.108
-e 'dat/loc.f' vs. -u 'dat.m' 0.010 324 0.97 0.411
-a 'nom.f' vs. -0 'nom/acc. m' -0.012 315 -1.19 0.321
4. Discussion
The results of the current PDT experiment largely replicate the LDT data [Vasilyeva, to appear]. As in the prior study, we observe differences between case forms in the presence of the frequency effect. In line with Finnish PDT data [Laine et al., 1999], Russian nominative wordforms are identified earlier than oblique case forms.
Apart from the nominative superiority effect, oblique differences arise in the PDT. As in the LDT [Vasilyeva, to appear], locative wordforms trigger longer processing time than -u forms ('dat/loc.f' and 'loc.m') and instrumental forms; in instrumental, feminine nouns are identified earlier than masculine nouns.
In contrast to the LDT data, the masculine locative identification is not hindered in comparison to the feminine locative. In [Vasilyeva, to appear], we hypothesize that the two -e-s ('dat/loc.f' and 'loc.m') share a grammatical feature representation (as suggested in some formal accounts of Russian morphology, e.g. [Wiese, 2004]). During the lexical access stage, context and case frequency serve as conflicting cues: the frequency cue points to the locative [Slioussar, Samojlova, 2014], while the context cue (absence of a preposition) points to the dative. The competition between the two candidates triggers longer processing time compared to other oblique word-forms. In the LDT, the context cue appears to win over the frequency cue and lure the participant into the erroneous analysis of masculine -e wordforms as datives and a subsequent reanalysis. In the PDT, this is not so, thus appealing to the context must belong to the decision making stage of the LDT.
Another discrepancy with the LDT results concerns genitive processing. In the LDT, these wordforms, similar to locatives, yield the slowest responses among masculine nouns. In the PDT, the masculine genitives behave differently: they are identified faster than the locative and do not entail additional processing cost compared to other masculine obliques. In [Vasilyeva, to appear], we propose an analysis for the -a endings that mirrors the analysis for ambiguous -e-s, namely a shared representation for feminine nominative and masculine genitive. Yet this analysis fails to capture the finding that according to the error analysis, masculine genitives turn out to be recognized more accurately than masculine locatives. As the PDT results resemble the LDT accuracy data in this respect and run counter to the LDT analysis of response latencies, the shared representation hypothesis for two -a-s is questionable. Another explanation is to be sought. The LDT as a decision making task implies recurrence to certain strategies (e.g., [Baayen, 2014]), and the processing cost associated with masculine genitives in the LDT could be strategic-based. According to word gender studies (e.g., [Holmes, Segui, 2004]), sublexical cues play a substantial role in the decision-making process,
for instance nouns having word endings typical for their gender are classified as words faster and more accurately than nouns having gender atypical endings. In the nominative, -a is a strong indicator of feminine gender. Processing of masculine genitive wordforms, also ending in -a, might be hindered due to this mismatch between the gender of the noun and the gender supplied by the sublexical cue.
Parallel to the auditory LDT [Gor et al., 2017], but in contrast to the visual LDT [Vasilyeva, to appear], phonologically overt feminine nominatives do not differ from phonologically zero masculine nominatives. This goes in line with Baayen et al. (2003)'s idea that the PDT, despite its visual modality, is closer to speech recognition.
As predicted, in contrast to the LDT data, effects of the inflectional and relative entropy are absent in the PDT (analogously to the absence of the morphological family size effect in the PDT, as reported in [Schreuder, Baayen, 1997]), supporting our claim that they operate at later stages of word processing.
The PDT allows to rethink the data obtained in the LDT for Russian case inflected nouns.
References
Baayen, 2014 - Baayen R.H. Experimental and psycholinguistic approaches to studying derivation. Handbook of derivational morphology. Lieber R., Stekauer P. (eds.). Oxford, 2014. Pp. 95-117.
Baayen et al., 2003 - Baayen R.H., McQueen J., Dijkstra T., Schreuder R. Frequency effects in regular inflectional morphology: Revisiting Dutch plurals. Morphological structure in language processing. Baayen R.H., Schreuder R. (eds.). Berlin, 2003. Pp. 355-390.
Bates et al., 2014 - Bates D., Maechler M., Bolker B., Walker S. Lme4: Linear mixed-effects models using Eigen and S4. R package version 1.1-7.2014. URL: http:// CRAN.R-project.org/package=lme4.
Benjamini, Hochberg, 1995 — Benjamini Y., Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the royal statistical society. Series B (Methodological). 1995. Vol. 57. No. 1. Pp. 289-300.
Brysbaert, Diependaele, 2013 - Brysbaert M., Diependaele K. Dealing with zero word frequencies: a review of the existing rules of thumb and a suggestion for an evidence-based choice. Behavior research methods. 2013. Vol. 45. No. 2. Pp. 422-430.
Clahsen et al., 2001 - Clahsen H., Eisenbeiss S., Hadler M., Sonnenstuhl I. The mental representation of inflected words: an experimental study of adjectives and verbs in German. Language. 2001. Vol. 77. Pp. 510-543.
De Jong et al., 2000 - De Jong N.H., Schreuder R., Baayen H.R. The morphological family size effect and morphology. Language and cognitive processes. 2000. Vol. 15. No. 4-5. Pp. 329-365.
Diependaele et al., 2012 - Diependaele K., Brysbaert M., Neri P. How noisy is lexical decision? Frontiers in Psychology. 2012. Vol. 3. Pp. 348.
Feldman, Fowler, 1987 - Feldman L.B., Fowler C.A. The inflected noun system in Serbo-Croatian: lexical representation of morphological structure. Memory & Cognition. 1987. Vol. 15. Pp. 1-12.
Feustel et al., 1983 - Feustel T.C., Shiffrin R.M., Salasoo A. Episodic and lexical contributions to the repetition effect in word identification. Journal of Experimental Psychology: General. 1983. Vol. 112. No. 3. Pp. 309-346.
Gor et al., 2017 - Gor K., Chrabaszcz A., Cook S. Processing of native and nonna-tive inflected words: Beyond affix stripping. Journal of Memory and Language. 2017. Vol. 93. Pp. 315-332.
Grainger, Segui, 1990 - Grainger J., Segui J. Neighborhood frequency effects in visual word recognition: A comparison of lexical decision and masked identification latencies. Attention, Perception, Psychophysics. 1990. Vol. 47. No. 2. Pp. 191-198.
Grosjean, 1980 - Grosjean F. Spoken word recognition processes and the gating paradigm. Attention, Perception, Psychophysics. 1980. Vol. 28. No. 4. Pp. 267-283.
Holmes, Segui, 2004 - Holmes V.M., Segui J. Sublexical and lexical influences on gender assignment in French. Journal of Psycholinguistic Research. 2004. Vol. 33. No. 6. Pp. 425-457.
Hyona et al., 2002 - Hyona J., Vainio S., Laine M. A morphological effect obtains for isolated words but not for words in sentence context. European Journal of Cognitive Psychology. 2002. Vol. 14. No. 4. Pp. 417-433.
Keuleers, 2013 - Keuleers E. Vwr: Useful functions for visual word recognition research. R package version 0.3.0. 2013. URL: http://CRAN.R-project.org/ package=vwr.
Kuznetsova et al., 2015 - Kuznetsova A., Brockhoff P.B., Christensen R.H.B. Package 'lmerTest'. R package version 2. 2015. URL: http://cran.uib.no/web/ packages/lmerTest/lmerTest.pdf
Laine et al., 1999 - Laine M., Vainio S., Hyona J. Lexical access routes to nouns in a morphologically rich language. Journal of Memory and Language. 1999. Vol. 40. Pp. 109-135.
Lenth, 2016 - Lenth R.V. Least-squares means: the R package lsmeans. Journal of statistical software. 2016. Vol. 69. No. 1. Pp. 1-33.
Lyashevskaya, Sharoff, 2009 - Ляшевская О., Шаров С. Частотный словарь современного русского языка (на материалах Национального корпуса русского языка). М., 2009. [Lyashevskaya O., Sharoff S. Chastotnyi slovar' sovremennogo russkogo yazyka (na materialakh Natsional'nogo korpusa russkogo yazyka) [Frequency Dictionary of the Modern Russian Language (on the base of the National Corpus of the Russian Language)]. Moscow, 2009.]
Milin et al., 2009 - Milin P.P., Durdevic D.F., Moscoso del Prado M.F. The simultaneous effects of inflectional paradigms and classes on lexical recognition: evidence from Serbian. Journal of Memory and Language. 2009. Vol. 60. No. 1. Pp. 50-64.
Peirce, 2009 - Peirce J.W. Generating stimuli for neuroscience using PsychoPy. Frontiers in Neuroinformatics. 2009. Vol. 2. No. 1-8.
Schreuder, Baayen, 1997 - Schreuder R., Baayen R.H. How complex simplex words can be. Journal ofMemory and Language. 1997. Vol. 37. No. 1. Pp. 118-139.
R Core Team, 2014 - R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing. Vienna, Austria, 2014. URL: http://www.R-project.org/
Samojlova, Slioussar, 2014 - Samojlova M., Slioussar N. Frequencies of different grammatical features and inflectional affixes of Russian nouns: A database. URL: http://www.slioussar.ru/freqdatabase.html.
Vasilyeva, to appear - Vasilyeva M.D. Russian case inflection: Processing costs and benefits. Advances in formal Slavic linguistics 2016. D. Lenertova, R. Meyer, R. Simik, L. Szucsich (eds.). Language Science Press, to appear.
Wiese, 2004 - Wiese B. Categories and paradigms. On underspecification in Russian declension. Explorations in nominal inflection. G. Müller, L. Gunkel, G. Zifonum (eds.). Berlin, 2004. Pp. 321-372.
Yap et al., 2012 - Yap M.J., Pexman P.M., Wellsby M., Hargreaves I.S., Huff M.J. An abundance of riches: cross-task comparisons of semantic richness effects in visual word recognition. Frontiers in Human Neuroscience. 2012. Vol. 6. Article 72.
Статья поступила в редакцию 09.11.2017 The article was received on 09.11.2017
Васильева Мария Дмитриевна - аспирант кафедры теоретической и прикладной лингвистики филологического факультета, Московский государственный университет им. М.В. Ломоносова
Vasilyeva Maria D. - postgraduate student at the Department of Theoretical and Applied Linguistics of the Faculty of Philology, Lomonosov Moscow State University
E-mail: linellea@yandex.ru
