Mechanisms of Verbal Morphology Processing in Heritage Speakers of Russian

Abstract

The goal of the study is to analyze the morphological processing of real and novel verb forms by heritage speakers of Russian in order to determine whether it differs from that of native (L1) speakers¹ and second language (L2) learners; if so, how it is different; and which factors may guide the acquisition process. The experiment involved three groups of subjects: 28 adult native speakers, 14 adult heritage speakers, and 17 beginning American learners of Russian. The results demonstrate that (1) novel form production in heritage processing, as in native and L2 processing, is rule-based, and that rule application—i.e. the generalization rate of conjugational patterns—depends on the input-based mechanism of statistical probabilities (to be defined below), and (2) that heritage speakers' mental representations of morphological structures are unstable and their morphological processing is different from either adult native or L2 processing.

1. Introduction

The focus of this paper is Heritage Russian as spoken in the United States. According to Polinsky (2000) and Andrews (1993, 1999), its linguistic features arise from interrupted first language acquisition and/or acquisition of L1 in parallel or in competition with L2, from impoverished L1 input and/or from the lack of a formal education in Russian.

The scholarly literature attests to the linguistic peculiarities of these speakers: simplification of the case system, inconsistent use of adjective endings, incorrect aspectual forms of verbs, a general increase in the use of analytical forms, a deficient lexicon, serious difficulties with lexical access and retrieval loss of stylistic registers, and phonological and tonal changes (Polinsky, 1996, 1997; Andrews, 1998; Zemskaja, 2001; Kagan & Dillon, 2001). However, morphological processing in heritage learners has so far received relatively little attention.

This study presents a preliminary investigation into the mental representations of morphological structures in these speakers and seeks to determine whether heritage language processing at the morphological level differs from native and L2 morphology processing, how it is different, and which factors might facilitate the acquisition process.

2. Theoretical Framework

The morphological representation in grammar and lexicon reflects basic cognitive organizational principles (Bybee 1995). The psycholinguistic and neurolinguistic literature usually focuses on three major questions: whether regularly inflected forms exist as whole-word representations or as stem plus affix, whether inflected forms are generated by rule or by analogy, and how morphological patterns generalize (what factors determine the productivity of affixes).

There are currently two main models that explain the processing of regular and irregular inflections in different ways: the dual-mechanism account and the single-mechanism account. The dual-system account is related to a generative or symbolic framework (Pinker & Prince, 1988; Marcus et al., 1992, 1995; Prasada & Pinker, 1993; Clahsen, 1999; Jaeger et al., 1996; Ullman, 1999). The basic assumption is that two qualitatively distinct mechanisms are responsible for the production of regular and irregular forms: regular inflections are produced by a mental symbolic rule, whereas irregulars are stored in an associative memory (or lexicon). Based on this dichotomy, the dual-mechanism model claims to account for differences in the processing of regular and irregular inflections. Regular forms are applied productively to novel forms independently of their type and token frequency (e.g. fax - faxed)² . Irregular inflections, on the other hand, are retrieved from associative memory, are frequency-sensitive, and form neighborhoods based on phonological similarity both in existing families (read - read, lead - led, breed - bred) and in extension to novel forms (cleed - cled).

Proponents of the single-system view (MacWhinney & Leinbach, 1991; Plunkett & Marchman, 1991; Rumelhart & McClelland, 1986) (Bybee 1985, 1988, 1995; Langacker 1987, 1988) treat all inflections as a network of mappings from base to inflectional forms arising from regularities found within the network. They argue that both regular and irregular verbs are processed by a single mechanism in associative memory. Differences between regular and irregular morphology are not attributed to different processing types as in the dual-system model, but to differences in type frequency, in particular to the greater type frequency of regular patterns.

There are problems in each account, and various later studies have attempted to address specific criticisms, to demonstrate opposite effects or to integrate the two approaches (Stemberger & MacWhinney, 1988; Prasada & Pinker, 1993; Alegre & Gordon, 1999; Plunkett & Marchman, 1991; Marcus et al., 1992; Pinker & Prince, 1988, 1991; Pinker, 1995; Skousen, 1989; Marslen-Wilson & Tyleron, 1997; Baayen et al., 1997; Bertram et al., 1999; Bertram et al., 2000.) For a review of the literature and discussion, see Gor, 2003). According to Baayen et al. (1997), for example, symbolic rule application and search in associative memory proceed simultaneously, and whichever route is faster wins the competition.

To explain variations between child and adult languages and the gradualness of language development in children, Yang (2002), following upon Baayen et al. (1997) as described above, introduced a variational competition-based learning model (Rule Competition Model). An individual's variable linguistic behavior can be modeled as a statistical distribution of multiple grammars. When an input sentence is presented, a grammar is selected; the probability with which it is selected is determined by its weight. The grammar is then used by the learner to analyze the sentence. If the analysis is successful (i.e. the sentence is successfully parsed), the selected grammar is rewarded and all the other grammars are indirectly punished; otherwise, the selected grammar is punished, and all the other grammars are indirectly rewarded. Learning is the adaptive change in the weights of grammars in response to utterances successfully presented to the learner (Yang, 2002). In child L1 acquisition, competing grammars are probabilistically accessed by the learner, resulting in variation (non-uniformity) in child language. After the language is fully acquired, a speaker's linguistic knowledge becomes stable. Gor (2003), accepting the main postulations of the model, extends the principles from child L1 acquisition to adult L2 acquisition. Her Rules and Probabilities Model postulates that rules are used probabilistically in L2 processing, reflect L2 input type frequencies and interact with the complexity of the paradigm³ in verb classes. I will apply this model to the current work.

3. The One-stem System of Russian Verb Conjugation

Before presenting and discussing our experimental material, we need to provide a short overview of the Russian verb system. For a full discussion, see Chernigovskaya and Gor (2000), Gor and Chernigovskaya (2003), and Gor (2003). According to the one-stem verb system (Jakobson 1948; Townsend, 1975), Russian has 11 verb classes, each with its own suffix, or verbal classifier⁴ and different types of stem changes (suffix alternations⁵ , consonant mutation and stress shift). Table 1, reproduced from Gor (2003), contains information about the type frequencies and productivity of the stems included in the experiment. Examples of verbs belonging to each class are given in the Appendix. The figures in the first row immediately below the headings are type frequencies in the language overall and are based on the Grammatical Dictionary of the Russian Language (Zaliznjak, 1980), with 23,877 verbs. These counts contain all verbs belonging to a particular conjugational class, both prefixed and unprefixed. In addition, type frequencies of unprefixed stems for the smaller unproductive classes are cited from Townsend (1975) and Davidson et al. (1996). The second and third rows contain two types of data on input type frequencies specifically obtained for L2 learners in Gor's experiment (2003). There is no information on input type frequencies for heritage learners. The specific problem of whether they should resemble native-language (NL) type frequencies, due to the input heritage learners receive at home from Russian-speaking parents, relatives and people in the community, is not addressed.

Table 1. Type Frequencies of the Verb classes Included in the Experiments: Native and Second Language Input

^a The first figure corresponds to the number of verbs in the active vocabulary, and the second figure (in parentheses) to all the verbs from the active and passive vocabulary combined.

4. Method

4.1. Research Questions

The study compares heritage speakers with native speakers and L2 learners on the accuracy of stem recognition and verb form generation and seeks to answer the following questions:

1. Do heritage speaker (HS) recognize stems, i.e. verb classes, and conjugation patterns in the same way as native speakers (NS) and/or second-language learners (L2L)?
2. What are the tendencies in HS processing of real and novel verbs?
3. What does HS morphological processing depend on?

I hypothesize (1) that HS morphological processing is different from both native and non-native processing, (2) that HS recognize verb stems and generalize conjugational patterns less easily than native speakers but with greater facility than American learners, and (3) that their processing is rule-based and reflects input-based statistical probabilities (Yang 2002; Gor 2003).

4.2. The Experimental Material

The material for the experiment is adopted from Gor and Chernigovskaya (2003) and consists of 80 verbs: 50 real verbs and 30 nonce verbs created by manipulating the initial segment of 30 highly frequent real Russian verbs. The verbs belong to 9 classes and subclasses based on the one-stem verb system. The list of verbs is presented in the Appendix. In the focus of the experiment are four regular verbal classes: high-frequency productive class -aj (non-past-tense form čit-aj-u 'I read' versus infinitive čit-a-t' 'to read'), small non-productive class -a- (piš-u 'I write' versus pis-a-t' 'to write'), high-frequency productive classes -i- (noš-u 'I carry' versus nos-i-t' 'to carry') and -ova- (ris-uj-u 'I draw' vs. ris-ova-t' 'to draw'). Each class, -aj-, -a-, -i- and -ova-, is represented by 5 real verbs of high token frequency, 5 low-frequency real verbs, and 5 nonce verbs. Other stimuli belong to -ej- and -e- (represented by 5 high-frequency real verbs and 5 nonce verbs each); -ij- and -avaj- (represented only by 2 verbs each, one real and one nonce), and (-o)j- (with 3 real verbs and 3 nonce verbs) classes⁶.

Three pairs of stems have similar past tense and infinitives but different conjugational patterns in the non-past tense, because the -j- is truncated: -aj- and -a-, -ej- and -e-, and (-i)j- and -i-. According to the Rules and Competition Model (Yang, 2002) and the Rules and Probabilities Model (Gor, 2003), the underlying stem can be recovered based on morphological cues and/or rule (or type) statistical probabilities. Thus, in the generation of novel verb forms, one can expect the high-frequency, or productive types with less complex conjugation rules, to be generalized to the small unproductive classes. The application of symbolic rules (or the generalization rate of conjugational patterns) depends on their statistical probabilities. This last point is particularly important in our analysis of nonce verbs, since the infinitive alone rarely provides definitive evidence of the conjugation. Here, then, it is not a matter of absolute or theoretical correctness but of the subjects ability to recognize the more probable pattern.

4.3. The Experimental Procedure

All verb forms were generated by the same technique used by Gor and Chernigovskaya (2003): verb stimuli in the form of the infinitive were embedded into simple carrying sentences, with the exchanges between the experimenter and the subject forming a quasi-dialogue.

Each subject was asked to generate 160 forms involving the 1st person singular and the 3rd person plural forms of the non-past tense. The total number of verb forms generated and analyzed was 9,440 (3540 nonce/5900 real). All answers were tape-recorded and then transcribed. When subjects produced two responses to the same item (self-correction), the response included in the analysis was the last one.

4.4. Participants

Group 1 consisted of adult native Russian speakers (NS) (N=28). Stimuli were presented to this group both orally and visually. Group 2 comprised heritage speakers (HS) of Russian (N=14), high school and university students who had been living in the U.S. since childhood (mean age of arrival 6;9.6 (i.e., 6 years 9.6 months); mean length of residence in the U.S 12;8). All subjects still spoke Russian at home with their parents. They fell into two subgroups based on whether they had received formal training in Russian or not:

• 7 subjects had had less than a year of formal education in Russian or none at all. Though the stimuli to this group were presented both orally and visually, they preferred not to follow the written text because of their poor reading skills in Russian.
• 7 subjects had had a few years of elementary education in Russia before emigration. They could read and write in Russian, and the stimuli were presented both orally and visually.
• Group 3 consisted of American learners (L2L) at the end of the first year of studying Russian at the University of Maryland (N=17). The stimuli were presented both orally and visually.

5. Results and Discussion

5.1. Stem Recognition Rate for the Four Most Frequent Stem Types

Results for Real Verbs:

1. HS recognition of the most frequent -aj- stem is native-like and almost twice as accurate as L2L in both high-frequency and low-frequency verbs.
2. The smaller and non-productive -a- stem is recognized much less often by both HS and L2L, yet HS show more accuracy than the L2L.
3. Results for verbs with the second most frequent stem, the productive -i-, show that both HS and L2L recognize them more easily than other stems (except for -aj-. However, L2L do not differentiate between high- and low-frequency verbs, while the performance of HS drops on low-frequency оnes by almost 10%
4. -ova- verbs are recognized least often by both L2L and HS groups, yet HS show the same sensitivity to token frequency and are outperformed by L2L on low-frequency verbs.

Results for Nonce Verbs:

1. NS generalize -i- and -ova- patterns more than they do paired -aj- and -a- patterns, which are in competition. The same tendency can be seen in HS and L2L.
2. Recognition of nonce verbs by L2L follows the same pattern as for real verbs.
3. Comparison of stem recognition for real and nonce verbs confirms our finding that L2L are not sensitive to token frequencies.
4. Now it is L2L whose morphological processing more closely follows that of NS (with the exception of the aj- pattern). This finding is discussed in more detail below.

5.2 Nonce Verbs: Type Frequency

Figure 2 shows type frequencies generated during the experiment (stem choice rate) and Russian language type frequencies based on Zaliznjak's (1980) dictionary containing 23,877 verbs. The dictionary frequencies should not be compared to the response frequencies, because the verbs types included in the experiment did not reflect the distribution in the dictionary. However, an analysis of response type frequencies with the dictionary type frequencies in mind (the blue bar in the figure) shows that for NS and HS the generation of nonce verbs reflects existing frequencies of pattern types (their productivity): -aj-, -i- and -ova-. For L2L, however, stem distribution is more uniform for frequent types but reflects input deficiency for other types. Textbook input frequencies (see Table 1) level the difference between the productive -aj- and -i- patterns and the productive -ova- and non-productive -a- patterns; moreover, since productive -ej- type is represented almost exclusively in passive vocabulary, L2L overgeneralize the non-productive -e- pattern.

5.3. Nonce Verbs: Type Frequency Change

Figure 3 shows the absolute change in type frequency across all stems represented in the experimental material. To calculate the absolute type frequency change (F_A ), a simple formula was used: F_A = F_E - R, where F_E is the type frequency in the experimental stimuli, and R is the type frequency in responses.

Figure 3. Type frequency for nonce verbs for native, heritage and American learners

Note: Russian language type frequencies based on Grammatičeskij slovar' russkogo jazyka [Grammatical Dictionary of the Russian Language] (Zaliznjak, 1980) are shown in blue.

For all groups, frequencies of -aj- and -i- types increase, whereas frequencies of -ova- and -oj- decrease, especially for HS in the case of -ova-, compared with initial stimuli frequencies.

The redistribution of initial frequencies in responses across all stems (verb types) used in the experiment exhibits positive correlation with the dictionary type frequencies: the most frequent stems -aj- and -i- exhibit strong productive power and are overgeneralized; infrequent stem types are less productive and are undergeneralized, producing mixed responses or a decrease in type frequency. The correlation with the dictionary type frequency is the highest for NS, lower for HS and lowest for L2L.

Correlation Coefficients:

There are two new stems, used by all groups, which are illegal⁷ in Russian: -(y)j- and -(u)j-stems. HS and L2L preferred -(u)j- stem, which seems to be an artifact of a developmental tendency similar to that of Russian children acquiring the -ova- stem (Gor, 2003). As for the -(y)j- stem, it was generalized to the -(o)j- class mostly by native speakers, who did not recognize the -(o)j- stem in nonce verbs.

5.4. Nonce Verbs: Stem Recognition Rate

Table 2. Stem Recognition for Native Speakers Nonce Verbs

Table 3. Stem Recognition by American Learners - Nonce Verbs

Table 4. Stem Recognition by Heritage Speakers - Nonce Verbs

We can see from the Tables and from Figure 4 that:

1. NS recognized the -aj- pattern (the most frequent and productive) in 71% of responses. The rate goes down for HS -aj- (63%) and L2 learners (53%). NS generalized it to 38% of the -a- verbs, whereas HS generalized it to 56% and L2L to 48% of the -a- verbs.
2. Type frequency indeed influences productivity in the nonce verb production. The -a- pattern (small and nonproductive) is recognized least often by all groups (NS52%, HS 36% and L2L 47%), with HS having more problems than L2L.
3. The -ej- pattern (small but productive) is recognized by both NS and HS (76% and 68%) but very poorly by L2L (14%), who generalize -ej- verbs as -e-verbs at 74% (because the -ej- type is not part of their active lexicon).
4. The small and non-productive -e- class is better recognized by NS and L2L (82% and 88%) than by HS (55%), who generalize almost half of them as -ej- type (42%).
5. L2L recognize the low type frequency -ij- stem more often than other groups (88% vs. NS 43% and HS 29%). This indicates that native and heritage speakers had major problems dealing with the small unproductive -(i)j- stem. In half of the responses of HS it was conjugated like the -i- class. Low identification and generalization rates for the -(i)j- stem are not surprising, given its low type frequency.
6. The high-frequency and productive -i- pattern is recognized most easily by all groups (NS 92%, HS 74% and L2L 92%). HS, however, recognized it less often than L2L and generalized to the -aj- and -ij- classes at about 10% each.
7. The frequent and productive -ova- pattern is well recognized by both NS (91%) and L2L (78%). The latter, due to intensive formal instruction, react to the morphological cues better than HS with no training in Russian. HS recognized it at only 52%, and in 20% of the instances they conjugated it as an -aj- verb without suffix alternation, a developmental tendency also found among Russian children (Gor & Chernigovskaya, 2003).
8. HS fared best on two lowest frequency unproductive stems: -avaj- (64%)compared to NS 45% and L2L 18%, and -oj- ( 27%) compared to NS 14% and L2L 6%.

To summarize, our findings are as follows. Stem generalization rates do indeed depend on type frequency, or the statistical probability that a particular pattern will be used for verbs with particular suffix vowels. All groups apply symbolic rules to verb processing, and in processing new verbs all speakers use the Vowel + j default rule (Chernigovskaya & Gor, 2000). HS perform more consistently on real verbs and correctly generalize stems for verbs they know (token frequency effect). However, they are inconsistent in applying rules in novel verb production, which leads to the leveling of classes. It seems that because of interrupted or incomplete first language acquisition, the weights of rules are not set and rules are applied with a probability of less than 1, which may also explain the existence of doublets (double forms) in the production of real verbs.

Comparison of Literate and the Illiterate Groups of Heritage Speakers

I was interested to see if the level of literacy in Russian of HS plays any role in their morphology processing. The comparison of the literate and illiterate groups revealed that the former consistently outperformed the latter on most verb types. The reverse effect was shown only for aj-stems, which was generalized more by the illiterate group (a result also observed in children by Gor and Chernigovskaya (2003). This finding may indicate that illiterate incomplete acquirers, who have no orthographic representations of morphological patterns to rely on in processing, show more signs of instability and more rules competition (Fig.5.)

6. Conclusions

The results confirm my first hypothesis that heritage speakers of Russian are different in morphological processing from both native and non-native speakers. However, my second hypothesis, that HS are better than American learners in nonce verb processing, was not confirmed. HS behave similarly to NS on real verbs, and their generation of nonce verbs resembles native rates in terms of type frequencies. This might suggest that their language competence (or input-based knowledge of statistical probabilities) is closer to NS than to L2L.

However, HS stem recognition in nonce verb reveals that their mechanisms of morphological processing are less developed than those of beginning L2L, who receive extensive explicit instruction on verb conjugation. HS grammars are unstable: the weights of rules are imprecisely set and rules are applied incorrectly. These developmental tendencies also are found among Russian pre-school children (Gor, 2003; Romanova & Svistunova, 2004). This helps explain incorrect generalizations by HS of low-frequency real verbs (with which they may be unfamiliar) and their poor performance on nonce verbs.

Literate HS, who are familiar with orthographic conventions and morphological regularities in the spelling system, develop better mechanisms of morphological processing than illiterate HS, who lack phonological and morphological awareness. Russian spelling is mostly based on morphological principles and preserves the identity of the morpheme. In pronunciation, however, unstressed endings are reduced and sound similar to one another, though they may carry important linguistic information. For this information to be noticed and generalized, speakers must be exposed to statistical regularity effects in orthographic representations of morphology. Our finding suggests that literacy in the first language may be a major factor influencing the development of input-based rules and probability mechanisms in morphological processing (Romanova, 2005).

The results of the study suggest that rules are used probabilistically in adult HS processing, which lends support to the Rules and Probabilities Model (Gor, 2003) and that incorrect generalizations are a result of the instability of rule weights. If learning a grammar is, indeed, the adaptive change in the weights of symbolic rules in response to input frequencies, it is crucial to discover which factors (such as the level of literacy in the first language) may successfully contribute to the development of HS grammatical competence.

It is hoped that the findings of this study will contribute to the growing body of research on this unique group of speakers in particular, and on language acquisition in general.

References

Alegre, M., & Gordon, P. (1999). Frequency effects and the representational status of regular inflections. Journal of Memory and Language, 40, 41-61.

Andrews, D. (1993). American-immigrant Russian: Socio-cultural perspectives on borrowings from English in the language of the Third Wave. Language Quarterly, 31,101-126.

Andrews, D. (1999). Sociocultural perspectives on language change in diaspora. Soviet immigrants in the United States. Philadelphia, PA: John Benjamins.

Baayen, H., Dijkstra, T., & Schreuder, R. (1997). Singulars and plurals in Dutch: Evidence for a dual parallel route model. Journal of Memory and Language, 37, 94119.

Bertram, R., Laine, M., & Karvinen, K. (1999). The interplay of word formation type, affixal homonymy, and productivity in lexical processing: Evidence from a morphologically rich language. Journal of Psycholinguistic Research, 28, 213-226.

Bertram, R., Schreuder, R., & Baayen, H. (2000). The balance of storage and computation in morphological processing: The role of word formation type, affixal homonymy, and productivity. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 489-511.

Bybee, J. L. (1995). Regular morphology and the lexicon. Language and Cognitive Processes, 10, 425-455.

Chernigovskaya, T. & Gor, K. (2000). The complexity of paradigm and input frequencies in native and second language verbal processing: evidence from Russian. Language and Language Behavior, 3 (2), 20-37.

Clahsen, H. (1999). Lexical entries and rules of language: A multidisciplinary study of German inflection. Behavioral and Brain Sciences, 22, 9911060.

Davidson, D. E., Gor, K. S., & Lekic, M. D. (1996). Russian: Stage one: Live from Moscow! Dubuque, IA: Kendall/Hunt Publishing Company.

de Bot, K. (1999). Foreword. In L.Hansen (Ed.), Second language attrition in Japanese contexts, pp. vii-viii. Oxford: Oxford University Press.

Derwing, B., & Skousen, R. (1994). Productivity and the English past tense: testing Skousens analogy model. In Lima, Corrigan & Iverson (Eds.), The reality of linguistic rules, pp. 193-218. Amsterdam and Philadelphia, PA: Benjamins.

Gor, K. (2003). The Rules and Probabilities model of native and second language morphological processing. St. Petersburg: St. Petersburg University Press.

Gor, K., & Chernigovskaya, T. (2004). Generation of complex verbal morphology in first and second language acquisition: Evidence from Russian. Nordlyd, 31(6), 819-833.

Hansen, L., & Ritz-Kurashige, A. (1999). Investigating second language attrition: An introduction. In L.Hansen (Ed.), Second language attrition in Japanese contexts, pp. 3-21. Oxford: Oxford University Press.

Jakobson, R. O. (1948). Russian conjugation. Word, 4,155-167.

Kagan, O., & Dillon, K. (2001). A new perspective on teaching Russian: focus on the heritage learner. Slavic and East European Journal, 45(3), 507-518.

MacWhinney, B., & Leinbach, J. (1991). Implementations are not conceptualizations: Revisiting the verb learning model. Cognition, 40, 121-157.

Markus, G. F., Brinkmann, U., Clahsen, H., Weise, R., & Pinker, S. (1995). German inflection: The exception that proves the rule. Cognitive Psychology, 29, 189-256.

Markus, G. F., Pinker, S., Ullman, M., Hollander, M., Rosen, T. J., & Xu, F. (1992). Overregularization in language acquisition. Monographs of the Society for Research in Child Development, 57 (4), Serial No. 228. Chicago: University of Chicago Press.

Pinker, S. (1999). Words and rules: The ingredients of language. New York, NY: Basic Books.

Pinker, S., & Prince, A. (1988). On language and connectionism: Analysis of a parallel distributed processing model of language acquisition. Cognition, 28, 73-193.

Plunkett, K., & Marchman, V. (1991). U-shaped learning and frequency effects in a multi-layered perception: Implications for child language acquisition. Cognition, 38, 43-102.

Polinsky, M. (1997). American Russian: Language loss meets language acquisition. In W. Browne, J-Y. Kim, B. H. Partee, & R. A. Rothstein (Eds.), Formal approaches to Slavic linguistics #4: The Cornell meeting, pp. 370-406. Ann Arbor: Michigan Slavic Publications.

Polinsky, M. (2000). A composite linguistic profile of a speaker of Russian in the U.S. In O. Kagan & B. Rifkin, with S. Bauckus (Eds.), The learning and teaching of Slavic languages and cultures, pp. 437-465. Bloomington, IN: Slavica.

Prasada, S., & Pinker, S. (1993). Generalization of regular and irregular morphological patterns. Language and Cognitive Processes, 8, 1-56.

Romanova, N., & Svistunova, T. (2004, October). Verbal morphology processing by Russian heritage speakers: The Peter Pan syndrome. Paper presented at Second Language Research Forum, University Park, PA.

Romanova, N. (2005, October). Acquisition of morphological awareness by Russian heritage speakers: Beware the Jabberwock. Paper presented at Second Language Research Forum, New York, NY.

Rumelhart, D. E., & McClelland, J. L. (1986). On learning the past tenses of English verbs. In McClelland, J. L., & Rumelhart, D. E. (Eds.), Parallel distributed processing: Explorations in the microstructures of cognition, vol. 2, pp. 216-271. Cambridge, MA: Bradford/MIT Press.

Skousen, R. (1989). Analogical modeling of language. Dordrecht: Kluwer Academic Publishers.

Stemberger, J.P., & MacWhinney, B. (1988). Are inflected forms stored in the lexicon? In M. Hammond & M. Noonan (Eds.), Theoretical Morphology: Approaches in Modern Linguistics, pp. 101-116. San Diego, CA: Academic Press.

Townsend, C. E. (1975). Russian word formation. Cambridge, MA: Slavica Publishers.

Ullman, M. T. (1999). Acceptability ratings of regular and irregular past-tense forms: Evidence for a dual-system model of language from word frequency and phonological neighborhood effects. Language and Cognitive Processes, 14, 47-67.

Yang, C. D. (2002) Knowledge and learning in natural language. Oxford University Press.

Zaliznjak, A. A. (2003). Grammatičeskij slovar' russkogo jazyka: slovoizmenenie [A grammatical dictionary of Russian: inflection]. Moscow: Russkie slovari.

Acknowledgements

The paper is based on my Master's thesis written under the supervision of Kira Gor, University of Maryland. I am grateful to Ken Sheppard, David Andrews, and two anonymous reviewers for their helpful comments on earlier drafts of this paper. All errors are my responsibility.

Notes

1. By native I mean a fully acquired first language. (back)

2. Type frequency (class size) refers to the total number of words that exhibit a certain pattern or paradigm. Token frequency, on the other hand, refers to how often particular words occur in speech. For example, the regular English past tense ed has a very high type frequency because it applies to thousands of verbs. Type frequency strengthens the representational schema of the pattern and makes it more accessible for use with new words. The vowel change in swam and rang, on the other hand, has a low type frequency and is less likely to extend to new items. However, these words have a high token frequency, and this supports the preservation of the irregular pattern. (back)

3. The complexity of the paradigm is understood as the number and type of rules shaping the conjugational pattern of individual verbs (Chernigovskaya & Gor, 2000). (back)

4. Ten verb classes are identified by one of the following suffixes: -aj-, -ej-, -a-, -e-, -i-, -o-, -ova-, -avaj-, -nu-, and - a-, and the eleventh class has a zero suffix and is divided into smaller subclasses. The suffix determines all the parameters of the conjugational paradigm, including the choice of the thematic vowel in the inflections of either 1st or 2nd conjugation. (back)

5. If the stem ends in a vowel, and the ending begins with a vowel, the first vowel is truncated. The same is true for consonants: the first one is deleted. Past tense and infinitive endings begin with a consonant, and non-past tense endings begin with a vowel; therefore, stem-final vowels will be deleted in non-past tense forms, and consonants will be deleted in past tense and infinitive forms. (back)

6. While all the verb classes included in the experiment belong to the suffixed classes, the last two, (i)j- and (o)j-, are subclasses of non-suffixed stems, and each has an idiosyncratic (irregular) feature in the conjugational pattern. (back)

7. Illegal morphemes are those which do not occur in a language or are not "allowed" by symbolic rules. (back)

Download file: N_R_HLJ_6_1.pdf