An ER-fMRI study of Russian inflectional morphology
Introduction
Inflectional morphology is at the center of an important debate in cognitive science, concerning the general principles according to which the mental lexicon is organized. So-called “Dual-system” (DS) approach distinguishes regular and irregular morphological forms. The former are computed by rules, the latter are stored in the memory. In the alternative “Single-system” (SS) approach, all forms are generated and processed by a single integrated system.
Initially, English past tense morphology was the testing ground for both approaches. According to the “Words and Rules” model, a version of DS approach proposed by Pinker, 1991, Pinker, 1999, regular past tense forms are generated and processed by a symbolic rule that is part of the productive, combinatorial system of grammar. Irregular forms are learned by rote and stored in the lexicon, from where they can be retrieved through associative memory mechanisms. The DS approach was also advocated e.g. in (Marslen-Wilson and Tyler, 1997, Pinker and Prince, 1988, Ullman, 2004). On the contrary, a connectionist network model from (Rumelhart & McClelland, 1986) represents a single system without any symbolic rules. All past tense forms are generated and processed by associative mechanisms that take into account phonological similarity and token and type frequencies of different elements. The SS approach was further developed e.g. in (MacWhinney and Leinbach, 1991, McClelland and Patterson, 2002, Plunkett and Marchman, 1993). The range of data used to test SS and DS theories has been very diverse: behavioral and neurophysiological experiments where participants generated forms from various real and nonce verbs, language acquisition and language deficit studies, and computer simulations. The results have always been controversial.
However, English past tense morphology is exceptionally simple: there is only one productive class that includes the vast majority of verbs and a small number of irregular verbs. So various authors investigated verb and noun inflection in other languages where the situation is more complex. German, Icelandic, Norwegian, Italian, Spanish, Arabic and Hebrew were among them (e.g. Berent et al., 1999, Clahsen, 1999, Clahsen et al., 2002, Hahn and Nakisa, 2000, Orsolini et al., 1998, Orsolini and Marslen-Wilson, 1997, Plunkett and Nakisa, 1997, Ragnasdóttir et al., 1999). Studies on Russian are discussed in Section 1.3. The findings offered new challenges for both approaches, and some of them cannot be easily accounted for by either approach. We illustrate this on the example of Russian below.
Thus, widening the pool of languages was extremely important for the SS vs. DS debate. For this reason, it appears to be problematic that existing functional imaging studies of past tense generation rely only on English, German and, in one case, Spanish data (Beretta et al., 2003, de Diego-Balaguer et al., 2006, Desai et al., 2006, Dhond et al., 2003, Indefrey et al., 1997, Jaeger et al., 1996, Joanisse and Seidenberg, 2005, Oh et al., 2011, Sach et al., 2004, Sahin et al., 2006, Ullman et al., 1997a). In most studies, irregulars are associated with a larger number of activated regions, but the list of these regions, as well as proposed explanations differ greatly. There are also several electrophysiological studies dedicated to past tense formation in English and German (e.g. Lavric et al., 2001, Marslen-Wilson and Tyler, 1998, Münte et al., 1999, Newman et al., 1999, Newman et al., 2007), which show variable results.
The present paper aims to fill this gap. We conducted an fMRI study based on Russian language where participants were asked to generate present tense forms from different real and nonce (nonword) verbs and to pluralize real and nonce nouns. We tried to avoid numerous pitfalls identified by the critics of the previous studies. We used two tasks (inflecting verbs and nouns) in a random order, which minimizes the risk of priming and strategy effects, and large sets of stimuli matched for word frequency and phonological complexity.
Russian verbs have two stems: the present/future tense stem and the past tense stem. Correlation between them determines the verb class. Out of several existing approaches, we will rely on the one developed by Jakobson and his followers (Davidson et al., 1996, Jakobson, 1948, Townsend, 1975), according to which Russian has 11 verb classes and several so-called anomalous verbs. Ten classes are identified by their suffixes (verbal classifiers). The eleventh class has a zero suffix, and is subdivided into subclasses depending on the quality of the root-final consonant (Jakobson and Townsend counted them as 13 separate classes). It includes many conjugational patterns and contains well under 100 basic stems.
Conjugational patterns of different classes include truncations or additions of the final consonant or vowel and may also include stress shifts, suffix alternations, alternations of stem vowels and stem-final consonants. Russian has two conjugation types in the present and future tense, i.e. two different sets of endings, and to which one a verb belongs is determined by its class. Importantly, the verb class is often unrecoverable from a particular form of the verb. For example, čitát’ ‘to read’ belongs to the AJ class, and its 3Pl present tense form is čitá-j-ut (-j- suffix is added, first conjugation type). Pisát’ ‘to write’ belongs to the A class, and its 3Pl present tense form is píš-ut (-a- suffix is truncated, first conjugation type, final consonant alternation, stress shift). Drožát’ ‘to tremble’ belongs to the ZHA class, and its 3Pl present tense form is drož-át (-a- suffix is truncated, second conjugation type).
Thus, Russian verb system is very complex. And, crucially, there is no obvious division into regular and irregular verbs. Unlike in English, there is no single productive pattern that can be applied to any stem irrespective of its phonological characteristics. Five verb classes are productive, but dramatically differ in type frequency. The Grammatical Dictionary of the Russian Language (Zaliznyak, 1977) contains 27,409 verbs. We counted the number of verbs in these five classes: 11,735 in the AJ class, 6875 in the I class, 2815 in the OVA class, 1377 in the NU class and 638 in the EJ class.
Russian noun system is much less complex than Russian verb system. Nouns are inflected for number and case and are classified into different declensions depending on their gender and on the set of their number and case endings. There are three main declensions, the forth declension with adjectival endings, several exceptional nouns and a number of uninflected nouns. First and second declensions usually have a choice of two endings for a particular form depending on the last consonant of the stem (all stems in the third declension end in palatal or sibilant consonants and use one set of endings). In addition to that, inside every declension there are small groups of nouns with various irregularities: minor stem changes or unusual endings in some forms. The majority of Russian nouns do not change their stem.
Some endings are unique for a particular declension, but most of them are shared by two or even three main declensions. In particular, Nom.Pl forms, which we looked at in this study, can have the following endings: −i (used with palatal, sibilant and velar stems of masculine and feminine nouns in all three declensions), −y (used with the other stems of masculine and feminine nouns in the first and second declension), −ja (used with palatal, sibilant and velar stems of neuter and some masculine nouns in the first declension), −a (used with the other stems of neuter and some masculine nouns in the first declension), −e (used in a very small group of animate masculine nouns in the first declension).
The predictions of the SS and DS theories were tested in numerous experiments with Russian verbs (e.g. Chernigovskaya et al., 2007, Gor, 2003, Gor, 2010, Gor and Chernigovskaya, 2001, Gor and Chernigovskaya, 2003, Gor and Chernigovskaya, 2005, Gor and Jackson, 2013, Gor et al., 2009, Svistunova, 2008, Tkachenko and Chernigovskaya, 2010). Adult native speakers, L1 and L2 learners and subjects with various neurological and developmental deficits were examined. In the majority of these experiments, participants were provided with infinitives or past tense forms of real or nonce verbs and prompted to generate 1Sg and 3Pl present tense forms.
Healthy adult native speakers showed a strong tendency to overgeneralize the AJ class pattern (AJ class is the most frequent). In particular, they applied it to the nonce verbs ending in −ili (only two real verbs and their derivates have this conjugational pattern, all the others belong to the productive and highly frequent I class) and to the ones ending in −yli (no real verbs have this conjugational pattern). Thus, despite the fact that Russian has several productive and highly frequent classes, one conjugational pattern is used as the default one. This is in conflict with the SS theory.
Four-year-old children also heavily rely on the AJ class pattern. But gradually, other patterns become more active. For example, around the age of five children stop making mistakes with OVA class verbs and actively overgeneralize this pattern. Overgeneralizations that do not respect phonological properties of the stem are a hallmark of a rule in the DS approach, and several rules that have different potential to be overgeneralized depending on various frequency-related and phonological factors contradict its very essence. The generalizations made in the studies of English-speaking subjects with SLI (specific language impairment), aphasiac deficits and Alzheimer disease (e.g. Ullman et al., 1997b) that supported the DS approach also were not borne out in Russian. The group of authors working on Russian argues that Yang’s (2002) model relying on multiple rules of different status may be better suited to account for their findings. A similar model for Russian is developed in (Gor, 2003).
The data from most previous functional imaging studies were interpreted in favor of the DS theory (Beretta et al., 2003, Dhond et al., 2003, Indefrey et al., 1997, Jaeger et al., 1996, Oh et al., 2011, Sahin et al., 2006, Ullman et al., 1997a). De Diego-Balaguer et al. (2006) appeal both to DS and to SS approaches. In all these studies, regular and irregular verbs were associated with different sets of activated regions: Jaeger et al. (1996) compared past tense generation with a baseline of reading, while in all subsequent studies, direct regular vs. irregular comparisons were made. But these sets did not coincide across studies. One can only say that irregulars usually produced more extensive activation.
According to the version of the DS approach developed by Ullman (2004), memory-based processing of irregular verbs depends on medial-temporal and temporo-parietal regions, while rule-based processing of regular verbs depends on the basal ganglia, Broca’s area, and neighboring anterior regions. No other principally different model was proposed, although many authors disagree with Ullman in details. However, only two above-mentioned studies found more activation in Broca’s area for regulars than for irregulars (Dhond et al., 2003, Oh et al., 2011). In addition to that, increased left IFG activation for regulars was observed in an fMRI study where processing of spoken regular and irregular forms was compared in a same-different judgment task (Tyler, Stamatakis, Post, Randall, & Marslen-Wilson, 2005). Several other studies observed the reverse pattern: Broca’s area was more activated by irregulars (Beretta et al., 2003, de Diego-Balaguer et al., 2006, Sahin et al., 2006). Sahin et al. (2006) suggest that this can be explained by conflict monitoring between the regular rule and an irregular form or by inhibition of the regular rule application.
In several studies the SS approach was preferred. Sach et al. (2004) observed only frequency effects, but no difference between regular and irregular verbs. Joanisse and Seidenberg (2005) found greater bilateral IFG activation for regulars than for irregulars, but then demonstrated that irregulars that were phonologically similar to regulars (e.g., slept, fled, sold) produced the same level of activation as regulars did, and significantly more activation than irregulars that were not phonologically similar to regulars did (e.g., took, gave). They conclude that observed activation patterns are better predicted by phonological properties of stimuli than by their regularity.
Desai et al. (2006) found that, when word frequency and phonological complexity are controlled for, the ‘regular > irregular’ comparison revealed no activated regions, while the ‘irregular > regular’ comparison was associated with greater bilateral activation of the posterior IFG (BA 44), the precentral gyrus, the anterior insula, the intraparietal sulcus (IPS), the basal ganglia, as well as some other small foci of activation. The authors note that these areas are commonly associated with executive control and attentional processes and are also activated by regular past tense generation compared to reading. They conclude that the observed activation differences reflect greater processing load posed by irregulars, which rely on less frequent inflectional patterns than regular verbs and therefore have greater attentional and response selection demands.
In the present study, real and nonce verbs and nouns were used as stimuli (see Appendix). Verbs were visually presented in the infinitive form, and nouns were presented in the Nom.Sg form. Subjects were asked to generate aloud as fast as possible the 1Sg present tense form if they see a verb or the Nom.Pl form if they see a noun. All oral responses were recorded simultaneously with fMRI data acquisition. Their correctness was assessed offline. When a response was no longer appropriate to the target’s category, the corresponding trial was discarded in the subsequent fMRI analyses.
The first group of 35 verbs belonged to the AJ class, the second group contained 35 verbs from several small non-productive classes. For the sake of brevity, we will further call the latter verbs irregular. The AJ class is productive and the most frequent. So we reasoned that if any differences between these two groups were found, we could further compare them to other groups, and if not, no other differences would be expected because these are the two poles of the Russian verb system (i.e. in the latter case, we will not be able to further exploit the diversity of this system, but this will become clear only when we have the results of the present study, because no similar fMRI experiments on morphologically rich languages have been done before).
Verbs in these two groups were matched for frequency using The Frequency Dictionary of Modern Russian Language (Lyashevskaya & Sharoff, 2009) and for length (see Appendix). Only unprefixed imperfective verbs were used. We also created two sets of 35 nonce verbs. They mimicked the general characteristics of the corresponding real verb groups (length and phonological properties of the stem). However, we tried to avoid close resemblance to particular real words, because this would be an extra factor that is hard to control for.
Two groups of 35 nouns were matched with the verb groups for frequency and length. All nouns were masculine, belonged to the first declension and had the Nom.Pl form ending in −y. In one group, the last vowel of the stem is dropped in many forms including the Nom.Pl form: e.g. koster ‘fire’ – kostry. In the other group, the stem remains the same in all forms, as is standard for Russian nouns: e.g. šofer ‘driver’ – šofery. For the sake of brevity, we will further call the first group irregular, although this is a relatively minor irregularity. As we noted above, Russian nouns have less diverse inflectional paradigms than Russian verbs, so this was the only irregular feature we could find in the necessary frequency range. Our goal was to analyze whether the effects of regularity would be comparable in verbs and nouns. We also had two corresponding groups of 35 nonce nouns.
Let us note that vowels are dropped only in a subgroup of noun stems ending in particular vowel and consonant clusters (e.g. −er, −or, −el, −ol etc.). We chose stems with such clusters both for irregular and for regular nouns not to make the first group much more phonologically homogenous than the second. In the majority of cases whether the vowel is dropped can be predicted from the combination of consonants before this vowel and from the position of the stress. In case of nonce nouns, participants did not know what the intended stress was, so different Nom.Pl forms could be licitly derived from them.
In total, there were eight types of stimuli: regular verbs (RV), irregular verbs (IV), regular nonce verbs (RNV), irregular nonce verbs (INV), regular nouns (RN), irregular nouns (IN), regular nonce nouns (RNN) and irregular nonce nouns (INN). We analyze them below using Regularity and Lexicality factors (we looked at verbs and nouns separately rather than putting them together and treating word category as the third factor primarily because the type of irregularity we could find for nouns was very minor compared to what we had in case of verbs). 21 Subjects took part in our study, so in total we had 735 responses in each category.
Section snippets
Results and discussion
Let us start with behavioral results. Participants made fewer mistakes with regular verbs than with irregular verbs (22 out of 735 and 71 out of 735, or 3.0% and 9.7%, respectively) and fewer mistakes with nonce regular verbs than with nonce irregular verbs (96 out of 735 and 320 out of 735, or 13.1% and 45.7%, respectively). A two-way RM ANOVA revealed significant main effects of Lexicality (F(1, 20) = 83.23; p < 0.001; η2 = 0.81) and Regularity (F(1, 20) = 32.29; p < 0.001; η2 = 0.62), but no significant
Subjects
Subjects were 21 native speakers of Russian (13 women), 19–32 years of age, with no history of neurological or psychological disorders. All participants were right-handed, as assessed by the Edinburgh Handedness Inventory (Oldfield, 1971). Subjects were given no information about the specific purpose of the study. All subjects gave their written informed consent prior to the study and were paid for their participation. All procedures were in accordance with the declaration of Helsinki and were
Acknowledgment
The study was partially supported by the Grant #12-06-00706 from the Russian Foundation for Humanitarian Sciences (RFHR) and by the Grant #0.38.518.2013 from St. Petersburg State University.
References (60)
- et al.
Default nominal inflection in Hebrew: Evidence for mental variables
Cognition
(1999) - et al.
An ER-fMRI investigation of morphological inflection in German reveals that the brain makes a distinction between regular and irregular forms
Brain and Language
(2003) - et al.
Reconciling time, space and function: A new dorsal-ventral stream model of sentence comprehension
Brain and Language
(2013) - et al.
Spatiotemporal maps of past-tense verb inflection
Neuroimage
(2003) - et al.
A new SPM toolbox for combining probabilistic cytoarchitectonic maps and functional imaging data
NeuroImage
(2005) - et al.
Stochastic designs in event-related fMRI
NeuroImage
(1999) - et al.
German inflection: Single route or dual route?
Cognitive Psychology
(2000) - et al.
An information theoretical approach to prefrontal executive function
Trends in Cognitive Sciences
(2007) - et al.
A double-dissociation of English past-tense production revealed by event-related potentials and low-resolution electromagnetic tomography (LORETA)
Clinical Neurophysiology
(2001) - et al.
Implementations are not conceptualizations: Revising the verb learning model
Cognition
(1991)
Rules, representations, and the English past tense
Trends in Cognitive Sciences
Rules or connections in past-tense inflections: What does the evidence rule out?
Trends in Cognitive Sciences
Decomposition of morphologically complex words in English: Evidence from event-related brain potentials
Cognitive Brain Research
An ERP study of regular and irregular English past tense inflection
NeuroImage
The past tense debate: Is phonological complexity the key to the puzzle?
Neuroimage
The assessment and analysis of handedness: The Edinburgh inventory
Neuropsychologia
On language and connectionism: Analysis of a parallel distributed processing model of language acquisition
Cognition
From rote learning to system building: Acquiring verb morphology in children and connectionist nets
Cognition
Abstract grammatical processing of nouns and verbs in Broca’s area: Evidence from fMRI
Cortex
Temporal and frontal systems in speech comprehension: An fMRI study of past tense processing
Neuropsychologia
Contributions of memory circuits to language: The declarative/procedural model
Cognition
Neural correlates of overcoming interference from instructed and implemented stimulus–response associations
The Journal of Neuroscience
Osobennosti poroždenija glagol’nyx form pacientami s bolezn’ju Alcgejmera (‘Generation of verb forms in Alzheimer’s disease’)
Lexical entries and rules of language: A multidisciplinary study of German inflection
Behavioral and Brain Sciences
The development of regular and irregular verb inflection in Spanish child language
Journal of Child Language
Russian: Stage one: Live from Moscow!
Neural circuits subserving the retrieval of stems and grammatical features in regular and irregular verbs
Human Brain Mapping
FMRI of past tense processing: The effects of phonological complexity and task difficulty
Journal of Cognitive Neuroscience
The rules and probabilities model of native and second language morphological processing
Beyond the obvious: Do second language learners process inflectional morphology?
Language Learning
Cited by (22)
Neural correlates of morphological processing: An activation likelihood estimation meta-analysis
2022, CortexCitation Excerpt :However, other research has suggested that this grammatical distinction between regular and irregular morphology disappears when phonological complexity, which will be discussed below, is matched between regulars and irregulars, yielding comparable performance in individuals with aphasia, for instance (Bird, Lambon Ralph, Seidenberg, McClelland, & Patterson, 2003). Neuroimaging studies in typical populations investigating neural substrates of regular and irregular inflectional morphology were conducted mostly in English utilizing past tense forms (Jaeger et al., 1996; Joanisse & Seidenberg, 2005; Klimovich-Gray & Bozic, 2019; Oh et al., 2011; Pliatsikas et al., 2014; Sahin et al., 2006; Tyler et al., 2005) and, to a lesser extent, plural forms of nouns (Sahin et al., 2006), but also in other languages including German using past participles (Beretta et al., 2003; Prehn et al., 2017; Sach, Seitz, & Indefrey, 2004) and plural forms of nouns (Beretta et al., 2003), Russian employing present tense forms of verbs and plural forms of nouns (Slioussar et al., 2014), and Italian utilizing past participles (De Martino et al., 2020). However, it is not clear to what extent regularity corresponds to a similar distinction in these languages.
Neural responses to Finnish inflected forms during overt and covert production: The role of stem frequency and stem allomorphy
2021, Journal of NeurolinguisticsCitation Excerpt :This corroborates findings on inflected word recognition, and suggests that BA 44 of the LIFG is involved in combinatorial processing of morphological units (Marslen-Wilson & Tyler, 2007; Ullman 2001, 2016; Vannest et al., 2005), both in recognition and production. Other brain regions consistently reported in the production of inflected words are the left cerebellum (de Diego-Balaguer et al., 2006; Oh et al., 2011), right cerebellum (Nevat et al., 2017; Oh et al., 2011; Pliatsikas et al., 2014; Slioussar et al., 2014), left putamen (Desai et al., 2006), and left and bilateral thalamus (Desai et al., 2006; Tyler et al., 2004). While the caudate nucleus of the basal ganglia has been associated with lexical operations, the putamen has been tied to the application of grammatical rules (Teichmann et al., 2008).
MORPHOSYNTAX in the BILINGUAL MENTAL LEXICON: An Experimental Study of Strong Stems in German
2014, Studies in Second Language AcquisitionExploring the nature of morphological regularity: an fMRI study on Russian
2023, Language, Cognition and Neuroscience