J . Child Long. 32 (zoos), 805-825. O 2005 Cambridge University Press doi: ~ o . ~ o r ~ / S o ~ o ~ o o o g o ~ o o ~ ~ ~ z Printed in the United Kingdom

Beginning and end in the acquisition of the perfective aspect in Russian*

SABZNE STOLL

Max Planck Institute for Evolutionary Anthropology, Leipzig

(Received 6 February 2002. Amvsed 7 April 2005)

A B S T R A C T

The goal of this research is to determine the relevant factors that aid in the acquisition of the perfective aspect in Russian. Results confirm the findings of previous research, which say that aspect is not learned as a uniform category, but rather interrelates with the acquisition of Aktionsarten. This study focuses on the factors responsible for the difference in the rate of the acquisition of two complementary Akrionsarten in the perfective aspect: telic verbs (verbs including a result/goal of the denoted event) and ingressive verbs (verbs including the beginning of the event).

Since the usage of Aktionsarten strongly depends on the surrounding discourse, two experiments that varied in their discourse complexity were conducted. One study looked at the production of isolated utter- ances (thirty-nine children aged 3;o to 6; 1 I ) and the other study fo- cused on complex texts (fifty-two children aged 3;o to b; 11) . It was found that while telics are used independently of discourse context, ingressives depend strongly on contextual information.

These results suggest that discourse complexity and narrative competence define the acquisitional process for ingressives, yet are irrelevant in the acquisition of telics.

[*I This research was supported by the Max Planck Institute for Psycholinguistics, Nijmegen. I warmly thank the children and the kindergartens of St. Petershurg that participated in the study. I am further indebted to Natalja Guseva for help in adminis- trating the experiments, Tatjana Krug1jako.r-a for transcribing the data, and Galina Dobrova for logistic help. Special thanks go to Balthasar Bickel, Johanna Nichols, Dan Slobin and Alan Timberlake for continuous support, many inspiring discussions, and very helpful comments at a11 stages of this research. Further, I would like to thank Elena Lieven and two anonymous reviewers for very helpful comments. Great thanks also go to Daniel Stahl for help with the statistics. Address for correspondence: Max Planck Institute for Evolutionary Anthropology, Deutscher Platz 6, 04103 Leipzig, Germany. e-mail : stoll@eva.rnpg.de

In recent decades the acquisition of aspect has been studied intensely for a variety of languages. Both grammatical aspect and lexical aspect (here Aktionsart) have been investigated. One of the main findings of this research was the correlation between the use of grammatical aspect and tense and the use of Aktionsarten and tense (for a summary of this research see Li & Shirai, 2000; Weist, 2002). Numerous studies have shown that in the early phases of language acquisition, telic Aktionsart and perfective aspect usually correlate with the past tense, whereas durative Aktionsarten, (i.e. stative and activity verbs) and imperfective aspect correlate with the present tense (cf. e.g. Bronckart & Sinclair, 1973 for French; Antinucci & Miller, 1976 for Italian; Bloom, Lifter & Hafitz, 1980; Shirai & Andersen, 1995; Wagner, 2001 for English; Stephany, 1985 for Greek; Li, 1989 for Mandarin; Shirai, 1998 for Japanese; Aksu-Koq, 1988 for Turkish; Weist, Wysocka, Witkowska-Stadnik & Buczkowska, 1984 for Polish and Gagarina, 2000 for Russian). This was a very important discovery in that it demonstrated that lexical semantics could play an important role in the acquisition of a grammatical category like aspect or tense. However, it is unknown which factors or principles are relevant for the generalization of the just mentioned semantic restrictions to a full-fledged independent grammatical category. Any theory trying to explain grammatical aspect has to show what principles or factors are responsible for the extension from lexical semantic classes to a grammatical category.

Generativist accounts usually assume a nativist position on functional categories and aspect is considered to be a functional category (Sano & Hyams, 1994). There is a dispute, however, as to whether functional categories mature (Borer & Wexler, 1987) or whether they are continuously present beginning with chiId grammar (Pinker, 1984). Under a maturationist account (Wexler, rggg), language acquisition consists of the maturation or growth of a sequence of grammars each relevant at different stages of development. These different stage theories are constrained by UG, which places the outer limits of maturation. The continuity assumption by contrast claims that there is no difference in abstract linguistic categories and rules between children and adults. There are two versions of this assumption found in acquisition studies. One is the strong continuity hypothesis which assumes that functional categories, as part of universal grammar, are fully developed at the outset of language acquisition, but may not be active in the beginning (e.g. Lust, 1999). The weak continuity hypothesis, however, though also under the assumption that principles are available from the beginning of development, suggests that structures need to develop over time and require lexical exposure (Clahsen, Eisenbeiss & Vainikka, 1994).

T E L I C S A N D I N G R E S S I V E S I N R U S S I A N

No matter which of these hypotheses is assumed correct, it is unclear what is meant by the claim that a functional category is active and developed, and what counts as proof that it is fully developed or mature. We know that far beyond age z ;o , when most UG theories suppose that parameters are set, children restrict their aspectual use to specific Aktionsarten and do not treat all aspectual forms alike in grammar. I t has been argued by Olsen & Weinberg (1999) that there is something like an 'optimal learning strategy' (also cf. the subset principle Wexler & Manzini, 1987) which claims that children restrict grammatical aspect first to specific lexical classes, i.e. they apply a restrictive rule first and then modify the rule when they find positive evidence that their target language allows aspectual usage which is independent of Aktionsart.

The question of how such processes of rule expansion and abstraction are structured is at the heart of functional approaches to language acquisition. One type of functionalist theory has come to be called the 'usage-based approach'. In this approach, children are taken to construct general abstract linguistic categories out of item-specific constructions (Tomasello, 1992; Lieven, Pine & Baldwin, 1997; Tomasello & Brooks, 1999). Applied to aspect, a usage-based theory begins with the assumption that lexical and grammatical aspect interact systematically, since grammatical constructions are first learned in a lexically specific way. The explanations for the generalization of the initial lexical restrictions are based on general cognitive and learning principles, such as analogy and general pattern finding principles, schematization, and abstraction. Another example of a functional approach when applied to the acquisition of aspect, is the prototype account proposed b y Li & Shirai (2000). They claim that children start out with the semantic prototype of each aspectual category and then move on to Iess prototypical exemplars. The central issue then is to make explicit how such an extension of a restriction in usage takes place, i.e. which factors are relevant 'to get from here to there'.

What emerges from this brief survey of theories is that the major challenge of all theories, whether formal or functional, is to explain what factors are relevant in order for the child to give up the lexical restrictions s/he is operating with in the beginning. This is the primary question this paper addresses, and therefore its results are relevant for both types of theories.

The contribution of the present paper is to investigate how a complex category such as perfective aspect in Russian can be acquired. Russian has a perfective and an imperfective aspect. The morphology is very complex and there is no single, dedicated marker for either aspect. Definitions of aspect vary widely with theories. Forsyth has presented an especially clear definition: 'A ~erfective verb expresses the action as a total event summed up with reference to a single specific juncture.' (Forsyth, 1970: 8). Aspect

strongly intertwines with Aktionsart and except for telic Aktionsart, all other Aktionsarten come only in one aspect. Usually, languages have more than the two prototypical Aktionsarten of telics and duratives and this includes Russian. T o claim that the category of aspect has been acquired, these non-prototypical Aktionsarten need to be acquired as well. Thus, if we want to learn about the extension of the initial lexical restrictions, we must look at the acquisition of these other Aktionsarten. The Aktionsarten investigated here are telic and ingressive verbs.' Tefics are highly prototypical for perfective aspect, ingressive verbs, which are a infrequent category are relevant only in specific communicative contexts. The two Aktionsarten are complementary in the sense that telics designate the result or goal of an event (e.g. proc'itat' 'read through.PVF') and ingressives, the beginning of an event (e.g. zaplakat' 'start crying.PVF').

I suggest that perfective aspect is learned in a piecemeal fashion by unifying several Aktionsarten into the abstract category of perfective aspect. At first there is no semantic representation of grammatical aspect, but the child deals with individual Aktionsarten. Furthermore, I claim that the different Aktionsarten are learned independently from each other in a context-specific way. Context here refers to structure of the discourse in which a form occurs. The telic Aktionsart is the most frequent and context- independent Aktionsart, i.e. it occurs in all communicative contexts and thus should be an earlier development than the more context sensitive ingressive Aktionsart. Since verbal forms are always embedded in discourse and the complexity of discourse is an important feature of communicative behaviour, I investigate the use of these forms in communicative situations that vary in discourse complexity. Such an approach is necessary because the choice of Aktionsarten and aspect is largely dependent on the immediate discourse context. Thus, we cannot expect a priom' that children use specific forms independent of the level of discourse complexity. T o find out about the role of discourse complexity for the acquisition of telics and ingressives, I look at two levels of discourse campIexity assessed by two studies. Study r focuses on the production of narratives, and Study t aims at the production of isolated utterances.

The paper is structured in the following way: I first briefly summarize the most important features of Russian Aktionsarten with special emphasis on telics and ingressives. Then, I summarize the results of a previous comprehension experiment on isolated utterances (Stoll, 1998), which lead to the main hypothesis of this paper that the learning of aspect as a category

[ I ] In CGermanic and Romance languages telicity is determined at the VP level, and not as in the Slavic languages by the verb on its own. In the Germanic and Romance languages researchers usually use the term telir VP or telic predicate (e.g. Dowty, 1979; Krifka, 1992).

T E 1 , I C S A N D I N G R E Y S I V E S I N R U S S I A N

is context-driven. T o test this, I introduce a more complex level of discourse, assessed by a narrative production experiment (Study I). The results of this narrative production experiment (Study I) are then compared to the earlier results of the comprehension experiment (Stoll, 1998). Next, the results of an isolated sentences production experiment (Study 2) are compared with the results of the comprehension experiment (Stoll, 1998). In the conclusions, I discuss the results of this paper and embed them in current research on the acquisition of aspect and Aktionsart.

A K T I O N S A R T E N I N R U S S I A N

Language acquisition research on aspect usually relies on the Vendlerian classification (Vendler, 1967) of temporal classes of verbs or an expansion of this classification (Van Valin & La Polla, 1997; Weist et al., 2004). However, this classification is not universal (Breu, 1994; Bickel, zooo) and actually never was intended to be (Vendler, 1967: 98). For Russian, this classification cannot be carried over (Flier, 1985; Stoll, 1998). As shown in Stoll (1998) the Russian Aktionsart system consists of at least five Aktionsarten, which are supported by the evidence collected in several linguistic tests. These Aktionsarten are: telics (focusing on the endpoint/ result of an event, sl'est' 'eat up.PVF'), duratives (the event itself without boundaries is focused, igrat' 'play.IPFV'), semelfactives (punctual verbs, e.g. pvygnut' 'jump up once.PVF'), delimitatives (verbs with a delimited time span the event is taking place, e.g. podurnat' 'think for a while.PVFY) and ingressives (verbs focusing on the beginning of an event, e.g. zaplakat' 'start to cry.PVF'). I concentrate here on telics and ingressives.

The telic Aktionsart Telics include the result or the goal of the event they are denoting, e.g. umeret' 'die. PFV'. Whether this result or goal is actually reached, is a matter of aspect, i.e. when the person was in the process of dying, but did not actually die, one would use imperfective aspect 'urnirat'die.IPFV'. If the person died one would use the perfective aspect. So, in Russian it is possible to say:

(1) O n umival, no ne umev. He.NOM die.IPFV.PAST.3SG.M but not die.PFV.PAST.3SG.M 'He was dying, but he didn't die.'

Thus, the imperfective version of this verb makes no statement of whether this goal that is part of the lexical verb semantics, has been achieved. In this study, however, I concentrate on telics in perfective aspect, which is also the most frequent aspect in which this Aktionsart

STOLL

occurs. In this aspect, the verb form entails that the goal or end is achieved. Morphologically, perfective telics are either marked by one of several prefixes or stem suppletion e.g. raz-dot' 'distribute.PFV', vzjat' ' take.PFV'.

Telics have a special status in the Russian aspect system. They are the only Aktionsart that come with strict pairing of an imperfective and a perfective verb form. All other Aktionsarten occur only in the perfective or imperfective form.

The ingressive Aktionsart

Semantically, ingressives denote the beginning of an action or an event. There are two major morphological markings of ingressives. Traditionally, the term Aktionsart subsumes only synthetic verbal categories. However, since in this study I am interested in the semantics of ingressives and not in its morphological expression, I include the analytic verbal construction in the analysis.'

Synthetic ingressives in Russian are commonly associated with the prefix za-, e.g. zaplakat' 'start crying.PFV', which is attached to a simplex durative verb. There are some other prefixes like $0- and u- that indicate ingressivity, but in this paper I will. concentrate on za- verbs, which are by far the most common ingressives. The prefix za- is polysemous and is not productive in its ingressive meaning, i.e. it attaches only to a limited set of durative simplex verbs.

Analytic ingressives consist of the perfective verbs statl.PFV 'become/ start' or nalinat'lnac'at' 'start.IPFV/PFV' plus the infinitive of an imperfective verb as in (2)

(2) On stal/nata'ol titat' pis'mo ' He started.3.PAST.PFV read.IPFV.INF letter'. He started to read the letter.

The main formal restriction of analytic ingressives is that the complement verb in the infinitive has to be imperfective. In contrast to synthetic ingressives analytic ingressives are productive and unmarked with respect to punctuality.

B A C K G R O U M D : TELICS A N D I N G R E S S I V E S I N C O M P R E H E N S I O N

A previous comprehension experiment (Stoll, xqg8) with preschool children (twenty two-year-olds (mean age = 2 ; 08, s.n. = o-og), twenty

[a] There is also an analytic telic construction with k o n h f ' stop.IPFV/konci't' stop.PFV+ verb.IPFV. This construction, however, is extremely rare and therefore was not considered in this study.

T A B L E I . Mean percentage of correct answers in the cornprehension experiment (StoEl, 1998)

Age Telics (n = I 4) lngressives (n = 2 )

four-year-olds (mean age = 3 ; 07, S.D. = o-oq), twenty five-year-olds (mean age = 5 ; 05, S.D. = o-of), and nineteen six-year-olds (mean age = 6 ; 06, s .~.=o.oq) showed that not all Aktionsarten occurring in perfective aspect are understood equally by children of all age groups. All children understood the teIic Aktionsart most easily and the ingressive Aktionsart proved to be most difficult, i.e. children of all ages performed poorly with ingressives, as shown in Table I .

A two-way ( 5 ) age groups x (a) Aktionsart ANOVA: showed that there was no interaction between Aktionsart group and age (F(4,94)=1.477, p = 0.2 I 5). Testing the within-subject variable showed that independent of age, telics are always better understood than ingressives (F(1,94) = I 8 I -67 I,

p < o . o o ~ ) . The mean average rate for telics is higher in some age groups than in others (F(4;gq)=s-gzz, p < o . o o ~ ) . A Tukey H S D post hoc test, however, showed that if one looks at individual groups in a pair wise comparison there is only a significant difference between the two-year-olds and all other age groups.

Furthermore, it is important to note that while all of the children, independent of age, understood telics to some extent, the same is not true for ingressives. The number of ingressives in this experiment was very low (two items). I thus chose a strong criterion of understanding here, i.e. a child was counted as understanding ingressives only if s/he understood both items.' According to this criterion, ingressives were understood by only a very small number of children in each age group (below 20% in each age group) (cf. Table 2). There is no statistically significant difference between age groups kZ= 1.10, df = 3, p =oa776).

These results demonstrated that perfective aspect is not learned as a unitary category and that not all perfective verbs are treated alike.

The main question that needs answering now is this: why are telics so much easier to understand than ingressives in this experiment? There is

[3] From now on, I ignore the results of the two-year-olds in the comprehension experiment because in the production experiments that form the main studies of this paper only children from three years onwards were tested.

81 I

S T O L L

T A B L E 2. Percentage of participants understanding ingressi~les i~ the isolated sentences comprehension experiment (Stoll, 1998)

Children understanding ingressives

a straightforward pragmatic reason. In isolation, a qualitative change of situation seems to be more communicatively relevant than a temporal change, where only the beginning of an event is expressed. Whereas a qualitative change of situation can be reported and focused on as a single event On ubil ego 'He killed him', a temporal change of situation, such as the beginning of an event On zasrnejalsja 'He started laughing', is typically seen in relation to some other action or event and makes most sense in such a linguistic context. Thus the semantics of the verb influences in which linguistic context a form typically occurs in (whether in isolation, concatenation of several utterances, long narratives). Another way of putting it is that ingressives are a dependent Aktionsart in the sense that they occur predominantly in concatenation with an antecedent. The comprehension experiment (Stoll, 1998) assessed only events in isolation. Therefore the difficulty children had with ingressives might be due to this specific context or indeed rather the lack of a necessary context for ingressives. If this is true, ingressives should be more salient in contexts in which such forms are more prototypical, such as the concatenation of events in a narrative. Bamberg & Marchman remark for a similar construction in German: '... in the adult German narratives, inceptive aspect [a term that corresponds to my ingressive Aktionsart, SS] -whatever surface form it takes - functions to signal the 'opening' of a narrative sequence of events which requires a subsequent 'closure' (Bamberg & Marchman, 1994: 564).' Accordingly, Aktionsarten with a temporal change of situation should be easier for children in a task that embeds forms in such a context. This reasoning leads to the main hypothesis of this paper, claiming that children learn Aktionsarten by their prototvpical contexts of use (HYPOTHESIS OF

COS'L'EX'~-DRIVEN LEARNLNG). Thus, children are sensitive to the context in which an Aktionsart

prototypically occurs and they learn them in these specific contexts. This would suggest that children learn ingressive verbs in the narration of consecutive events. By contrast telics are ubiquitous in all contexts, whether in isolated utterances or in the description of a complex event and are

TE1, ICS A N D I N G R E S S I V E S IN R U S S I A N

therefore an earlier development than ingressives. The prediction of this hypothesis thus is that children will use telics independent of context, but ingressives will only be used in sequenced events. It is only later that the meaning of an ingressive verb or a construction becomes detached from this context and becomes generalized into an abstract meaning.

S T U D Y I : N A R R A T I V E P R O D U C T I O N T A S K The goal of this experiment is to discover whether children indeed use ingressives in a context that is typical for ingressives. T o do this, I first look at the use of ingressives in their prototypical context, i.e. in the narration of sequenced events. We expect that if the hypothesis that children learn Aktionsarten in a context-driven way holds true, ingressives should be more frequent in a narrative than in isolated utterances describing a single event, i.e. we expect that children will produce ingressives in embedded contexts, involving sequences of events, but not in isolated utterances. However, to understand and linguistically express a sequence of events presupposes an advanced level of narrative competence. Thus, only children with sufficient narrative competence are expected to use ingressives.

Fifty-two preschool children (twelve three-year-olds, mean age=3;06, S.D. = 0-03 ; thirteen four-years-olds, mean age=4;07, S.D. =om03 ; fourteen five-years-olds mean age=S;o6, S.D. =0+03; thirteen six-years-olds, mean age=6;06, s.o. =0.03) of kindergartens in the center of St. Pctersburg took part in the experiment.

Materials

r h e stimulus was a picture book without words called 'Picnic' (McCully, 1984). The story proceeds as follows: a family of mice packs their things on a truck in order to go for a picnic. They start driving. The road is very bumpy and at some point, one little mouse falls off the truck, together with its toy mouse. From this point onwards, the adventures of the two parties, i.e. the little mouse with its toy mouse on the one hand, and the rest of the family on the other hand, develop independently. There are several switches between the two stories, which allow the reader to keep the important plot, i.e. the loss of the mouse, in mind. No one in the family realizes the loss of the little mouse and they drive further. Finally, they arrive at a picnic spot, and they start playing, bathing and having fun. Scene switch: meanwhile, the little mouse is alone in the woods looking

STOLI ,

for food; it discovers wild berries. Scene switch: the family continues enjoying their excursion, and several different activities are going on at the same time, e.g. playing, swimming, and ballgames. Scene switch: the little mouse eats many berries and lies down. Scene switch: everybody in the family sits down to eat. At this point they realize that one mouse is missing and they begin to search (with several simultaneous actions of different members of the family, shouting, crying, searching, etc.), but without success. Finally, they drive back, shouting and looking for the little mouse. Scene switch: the mouse hears the car corning and runs out on the street, which leads to a happy reunion. Then, there is a repetition of the theme, as the little mouse has forgotten its toy mouse in the woods. The mouse remembers it and runs back to get her toy mouse. Then the whole family starts the picnic all over.

Procedure

Two experimenters conducted the task in a separate room in the kindergarten. One experimenter operated the equipment and took notes, while the other experimenter, a native speaker of Russian, interacted with the child. In a arm-up period immediately before the experiment, the child was made familiar with the experimenter and a hand puppet (lion), which served as a listener to the story of the child. The lion was introduced as having bad sight but he was very eager to hear the story. As assessed by a pilot-study the puppet proved to be less intimidating than an adult listener, and the children were highly motivated to tell the story to the lion. After the warm-up the experimenter asked the child to look through the book, page by page. Then, the child was asked to tell the story depicted in the book to the lion, while going through the book for a second time. The child was reminded that s/he should tell the story of the book and not just describe the pictures. When necessary, the experimenter emphasized this several times. This was especially important in Russian kindergartens, where picture books are used frequently by teachers and speech pathologists (present in every Russian kindergarten) to assess the vocabulary of children. The stories of the children were audio taped and subsequently transcribed by a different native speaker of Russian.

R E S l i I - T S

Again, as in the earlier comprehension experiment (Stoll, 19981, all children used at least some telics. There is no difference in the use of telics across age groups. By contrast, not all children used ingressives. T o analyse the effect of age and experiment on the dependent binary

814

T E L I C S A N D KNCjRESSIlrES I N RUSSI.4N

T A R L E 3. Percentage of participants understanding ingressives in Stoll (1998) and using ingressives in Study I

Comprehension experiment Narrative production Age (Stoll, 1998) experiment (Study I )

variable, i.e. uselunderstanding or non-usejnon-understanding of ingressives, I used a logistic regression analysis. A child was counted as using ingressives when s/he used at least one ingressive. As hypothesized, there is a difference in the performance between ingressives in the narrative production experiment (Study I ) and the comprehension experiment (Stoll, 1998) on isolated utterances as shown by the overall model of a logistic regression Ole = 53.07, df = 3 , p <o-001). The percentages are illustrated in Table 3 .

The logistic regression further illustrated that there is a significant interaction between experiment and age group, i.e. the differences between experiments are due to age (Wald x2 = 4.49, df = I , p = 0.039, eP= 2.7). W e thus look at the influence of age in each experiment and age group separately. The general model of a logistic regression shows that in the comprehension experiment, there is no significant difference between age groups ( ~ " 0 . r 5, df = I , p =o-67) . In the narrative production experiment by contrast, we do find a significant difference between age groups (Wald x2= 10'77, df = I , p = o . o o ~ , eP= 3.06). The likelihood that children use ingressives in the narrative production experiment increases with age. Further logistic regressions tested whether there is a difference between experiments within age groups. For all age groups except three-year-olds, we find a significant difference between tests (three-year-olds: ~ ' = o . 3 0 , df= I , p =o.586, four-year-olds: Wald x2=7*71, df= I , p =oaoog, eP=26, five-year-olds: Wald x2= I I '27, df = I , p =o-001 , ea= 2 d 3 , six-year-olds: Wald xa= 12-86, df= I , p < o . o o ~ , efi=47.6).

These results are especially interesting because in the production experiment children had a choice of whether to use ingressives or not. Although they could have chosen to not use ingressives in their narrations, we observe that except for the three-year-olds, a higher percentage of

1 children used ingressives in the narrative production experiment than children understood ingressives in the comprehension experiment (Stoll, 1998). There is a clear increase over age in the number of children who use ingressives in the narrative production experiment. While children

STOLL

T A B L E 4. Proportion of children who use ingressives and mention at least two core story elements

Age Three-year-olds Four-year-olds Five-year-olds Six-year-olds

r .on (a/z) 0.28 ( 4 7 ) 0.82 (9/1 I ) 1.00 ( I [ /I 11

of all age groups have great difficulties understanding ingressives in isolated utterances, children from at least 4;o onward master them much better in a more appropriate communicative context, i.e. in a narration. These results seem to strengthen the hypothesis that the acquisition of Aktionsarten/aspect is context-driven. However, in the following we need to check whether they indeed embed ingressives in a sequence of events. They still might pick out isolated events as young children often do and use ingressives to describe them.

I N C R E S S I V E S .4ND N A R R A T I V E C O M P E T E N C E

I have argued above that ingressives need a sequence of events to be meaningful. T o construct a narrative or at least part of a narrative, a child must be able to understand events as sequenced in time, their possible causal relation, and the role of protagonists, etc. Thus, we can expect a correlation between narrative competence, which is one important milestone in the child's general cognitive development, and the use of ingressives in a story. In order to measure narrative competence relevant for the use of ingressives, I determined three critical core components of the story (following Berrnan & Slobin, 1994). Two of them needed to be mentioned by the child in order for a text to be counted as a narrative. The three components are: one, the loss of the mouse, two, the realization of the loss and the search and three, the family reunion.

Table 4 illustrates the correlation between children using ingressives and their mentioning at least two of the core components of the story.

Both of the three-year-olds who used ingressives mentioned at least two core components.4 Within the group of four-year-olds only two out of the seven children who used ingressives mentioned at least two core components. The five and six-year old children who used ingressives showed narrative competence. A chi-square showed that there is a significant difference between age groups (xi= 13'47, df = 3, p =o.ooq). The

[4] One of these children mentioned Core Component 3 (the reunion) indirectly, which makes clear that hc understood the story even without mentioning the other core components explicitly. Hn\verer, in Table 4 1 did not count such implicitly mentioned components.

T E L I C S A N D I N G R E S S I V E S I N R U S S I A N

three-year-olds, five-year-olds and six-year-olds, who used ingressives do not differ from each other, but the four-year-olds behave significantly different (x2 = I 2-34? df = I , p =o*oo2) from the rest, i.e. they preferred to use ingressives without embedding them in a sequence of events.

The behaviour of the four-year-olds suggests that telling a story on a larger scale is not a necessary criterion for the use of ingressives. However, the criterion of using core components, which actually is a measure of overall narrative cohesion, might be too strict and it might well be that children who tell parts of the story without explicitly mentioning all part of the plot use ingressives. Thus, I introduced a second, measure of narrative competence, which assessed whether ingressives were embedded in a sequence of events involving the same protagonist, i.e. independently of overall narrative cohesion. It was then tested whether there is a significant difference between the number of children in the age groups who always embedded ingressives in a sequence of events. Again, as for the former measure of narrative competence, there was a significant differ- ence between age groups @=13'33, d f = j , pto-001). This difference was again due to the four-year-olds (x2= 6.53, df= I , p = o . o ~ I ) . Thus, the four-year-olds did significantly more often use ingressives in non-embedded utterances than the children of the other age groups.

Thus, the four-year-olds behave differently from the other age groups according to both measures of narrative competence. This behaviour will be discussed later in connection with rhe results of Study 2 in which the four-year-olds also behaved differently.

Narrative competence measured here by the ability to sequence events for the use of ingressives then might partly explain the increase of the number of children using ingressives over age. The older the children get, the better their na.rrative competence becomes. However, there is wide individual variation within and across age groups in this task (cf. Stoll, 2001, 2003). This is shown by the fact that two three-year-olds could already use ingressives and mention the core components of the story, whereas two six-year-olds were unable to construct a coherent story and did not use ingressives at all.

S T U D Y 2 : I S O L A T E D S E N T E N C E S P R O D U C T I O N E X P E R I M E N T

In the previous two experiments we found that children of all age groups have difficulties understanding ingressives in a non-prototypical but grammatical context. However, in a prototypical context (in a description of a sequence of events) the overall percentage of children using ingressives is much higher. The context-driven learning of aspect would then predict that children do not use ingressives in a description of isolated scenes,

S T O L L

i.e. in describing an event with an isolated utterance. This prediction is tested in the following isolated production experiment.

M E T H O D

Participants

In all 39 children took part in the experiment (nine three-year-olds (mean age = 3 ; 07, S . D . =o.og) ten four-year-olds (mean age =4;og, S.D. =oaoq, eleven five-year-olds (mean age = 5 ;o5, S.D. =o.oq), and nine six-year- olds (mean age= 6;04, S.D. =o.oq). No child took part in both the comprehension and the production experiment in order to ensure that the two experiments did not influence each other.

I

Mate rials

I used the same stimuli used in the comprehension experiment (Stoll, 1998). The experiment took the following form. Every child was shown approximately thirty short films. Some children described only between 27 and 29 films because they refused to watch the whole set of films due to fatigue, others simply chose not to describe some pictures. Unlike the experiment in Stoll (1998), every scene was shown separately. The coding was done on the basis of the verb forms children used. The verb forms were classified according to the Aktionsart classification in Stoll (1998).

Procedzdre

Each child was tested individually in a separate room in a nursery school in St. Petersburg. The same two experimenters that carried out Study I

conducted the experiment. Again, the native speaker of Russian interacted with the child, while the other experimenter took notes and took care of the recording. The experiment for each child was divided into at least two sessions, each taking place on a difierent day. In the instructions given to the child the two protagonists of the films were first introduced. Then the child was told that short films, in which these two protagonists appeared, would be shown and that after each film a question would be asked about what had happened in the film. Every child was given three warm-up films to help them become more familiar with the procedure. T o exclude order effects, the order of films was randomized for each child. The children's answers were noted down during the experiment by the other experimenter. Furthermore, the sessions were audiotaped and later transcribed by a Russian native speaker who was not familiar with the goals of the experiment. These transcripts provide a check of the results noted during the experiment.

818

T E L I C S AND 1NC;RESSIVES I N R U S S I A N

TAB I. E 5 . Percentage of children understanding in StolE (I 998) and producing ingressive$ in Study z

Comprehension experiment Study z: isolated sentences (Stoll, 1998) production experiment

R E S U L T S

As in the other two experiments, children mastered telics best. All children used at least some telics to describe the scenes. This was not the case with ingressives.

The expectation according to the hypothesis that children learn Aktionsarten by context is that children will use fewer ingressives in the production task with isolated scenes, because this is a non-typical context for ingressives. As a second measure we then analyse the linguistic context in which ingressives occur.

First, we compare the number of children that understand ingressives in isolated utterances (Stoll, 1998) with the number of children that use ingressives in the isolated production experiment of Study 2 (Table 5 ) . A logistic regression showed that the best fitted model only included the type of experiment, i.e. only the type of experiment made a significant difference (Wald x2 = 15-23, df = I, p <O.OOI, eo= 6-85). Proportionally more children use ingressives in isolated production experiment of Study 2 than understood them in Stoll (1998) with the same stimuli.

Thus, so far, it seems that in both Study I and Study 2, irrespective of the targeted communicative context, children use ingressives.

Second, to judge whether children indeed use ingressives in isolation, we need to assess the immediate context of the ingressives used in the iso- lated sentence production experiment. As already pointed out, the possible contexts in which ingressives would appear in the present production experiment differed substantially from the comprehension experiment in Stoll (1998). In the production experiment children had a choice as to how to describe the event. They could either choose to describe it briefly with a single utterance, such as in (3), or they could give more context in which they would embed the ingressive, as in (4).

(3) Toh zaplakal. Tosha.NOM start.crying.PFV.PAST.3SG.M. 'Tosha started crying.' (M 6;o)

S T O L L

TABLE 6. Percentage of ingressives in isolated and concatenated utterances in Study 2

Ingressives used in Ingressives used in Age isolated utterances concatenated utterances

(4) MaSa podnjala golovu naverx, Masha.NOM raise.PFV.PAST.3SG.F head.ACC.SG up potom zasmejalas'. then start.crying.PVF.PAST.3SG.F 'Masha raised her head and then started laughing/burst out laughing.'

(F 4; 1 1 )

The relevant data are summarized in Table 6. It is important to note that, as expected, the overall percentage of

ingressives used in this task is very small and partly is due to the same participants. It is thus difficult to interpret the results because the power especially for the three- and four-year-olds is very small. The conclusions must, therefore, be tentative. The small number of ingressives itself, however, strengthens the hypothesis that the acquisition of Aktionsarten is context driven because the stimuli invited isolated utterances-a context militating against ingressives.

A binomial test (p=o-5) showed that only for the six-year-olds was there a statistically significant preference to use ingressives in embedded utterances (p=o-013). For the five-year-olds there was a trend, but no statistical significance at the 5% level (p=o.o92). For the three- and four- year-olds, the power was very small due to the low number of ingressives used. Looking at the percentages the youngest children (three-year-olds) behave like the older children. However, the three-year oIds used only four ingressives and the four-year-olds seven ingressives. Three of the four ingressives used by the three-year olds were in fact embedded but no significant preference was found for this age-groups (three-year-olds, p=o.6z5). The four-year-olds used seven ingressives overall of which only two were embedded (four-year-olds, $=oq453). In this age group the preferred context are thus isolated utterances. Again, as in Study 1, the four-year-olds seem to have a clearly different strategy than the children of the other age groups, mainly using ingressives in isolation. However, all the verb forms used by the three-and four-year-olds in isolation are of one and the same verb, namely zaplakat' 'start crying', which was depicted

TEI . ICS AN13 I N G R E S S I V E S I K R U S S I A N

in one of the stimuli. By contrast, in the embedded context zaplakat' 'start crying' was used only once by a four-year-old (in one out of the two occurrences) and not at all by the three-year-olds. Thus the use of ingressives in isolated contexts goes back to one verb in the two age groups. In the five and six-year olds the use of raplakat' in isolated contexts goes down to 33 U/o respectively. I t seems that the three-and four-year-olds used this one ingressive verb in an item-specific way. However, more research is needed to make conclusions about the behaviour of the four- year-olds.

I C O N C L U S I O N S

The results of these studies show that Aktionsarten, narrative ability of the child, age, and the communicative context of the forms in question are relevant for the acquisition for the grammatical category aspect.

The comparison of Aktionsarten in different contexts presented in this paper suggests that the level of contextual complexity of the linguistic context i s an important factor for the acquisition aspect for children under age 6;o . This has been demonstrated by the different pace in the acquisition of two Aktionsarten of perfective aspect. The two studies were based on the observation that children perform very poorly with ingressives in a comprehension task of isolated utterances (Stoll, 1998), which are a non-prototypical context for ingressives. I suggested that children would perform better in a task that aims at the prototypical context of ingressives, i.e. in the narration of sequenced events. This expectation was confirmed for the five and six-year-olds in Study I , the narrative production experiment. A second study tested the use of ingressives in a non-prototypical context, i.e. in isolated utterances. As expected, in the isolated sentences production experiment (Study z ) , the number of ingressives used was very small. The older children preferred to add the prototypical context for ingressives, i.e. they embedded them in a sequence of events and they did not do this with telics. This is also true for the three-year-olds if we look at the raw numbers which are, however, very small. Again as in Study r , the four-year-olds behaved differently and showed a preference to use ingressives in isolation. In both stuciies the four-year-old preferred to use ingressives in isolation. In Study 2 all ingressives occurring in isolation used by the three-and four-year olds were of the same verb, this, however, was not true for Study I .

A possible hypothesis to entertain would be that ingressives are acquired in a U-shaped way and four-year-olds might go through a reorganizational phase in which they use ingressives in non-typical contexts, i.e. in contexts in which they usually would not occur. In fact, one might even hypothesize that such a mechanism of generalization is necessary to allow the language

S T O L L

learner to get from pure imitation to a stage in which he can use language as a productive system. This, however, as already stated, needs further research.

A further relevant factor for the acquisition of the perfective aspect might be that there is a difference in cognitive complexity in the semantics between ingressives and telics. The function of ingressives is cognitively more complex, because not only does a single isolated event need to be registered in order to use ingressives appropriately, but also additionally they must register a sequence of events. In order to be able to present a sequence of events, certain cognitive prerequisites have to be met, such as the recognition of causal and temporal relationships, and the recognition of a protagonist in a story or sequence of events (cf. Berman & Slobin, 1994). Slobin discusses a similar case, the converb -ErEk in Turkish, for which 'proper L I S ~ requires an ability to manage attention flow in narrative' (Slobin, 1995: 366); hence this form is a relatively late development, and not acquired before age 7;O. Interestingly, Slobin (1995: 351) finds that even though the form -ErEk is frequent in speech directed to children, it is nevertheless virtually lacking in their own speech throughout the preschool period. The findings of this paper thus lend further evidence to Slobin's findings about the close interaction of linguistic and conceptual develop- ment and the complexity of a linguistic sign. However, they add one crucial component that interrelates with his findings, namely the importance of the discourse context in which a linguistic feature is embedded. For successful statements about the acquisition of a linguistic feature these factors need to be considered.

The results of this paper are compatible with a usage-based approach, which predicts that Russian children learn the grammatical category aspect in a piecemeal fashion. First, the child seems to have no general semantic representation of perfective aspect, but rather relies on the lexical class of Aktionsarten. In the comprehension experiment (Stoll, 1998). children start out with a much higher correct rate of telics (two-year-olds) than of ingressives. All of the children at least understood some telics. The three-year-olds already show a correct rate of 75% of telics in the comprehension experiment. Telics in the comprehension experiment show a clear development, reaching 95% correct answers for the six-year-olds. Ingressives, in contrast, are a later development. I argued that ingressives typically occur in a different context than telic verbs. While telics can occur in all contexts, hence also in isolation, ingressives need a minimum context of two concatenated utterances. Thus, the semantic representation of ingressives is context-dependent for quite a long time (at least up to age 6 ; o with a possible reorganizational phase around age 4;o) and only later this category gets abstracted from specific contexts and becomes an abstract semantic category.

TELICS A N D I N G R E S S I V E S 1N R U S S I A N

While the empirical results of this study are relevant for any theory, a usage-based approach seems to give most explicit recognition to the issues covered in this paper. The focus here is on the function of verbal forms in their specific linguistic contexts. Apart from being a grammatical

I

category, aspect is a pragmatic category defined by linguistic contexts (such as foreground and background) and extralinguistic contexts (such as politeness). Talking about the acquisition of aspect only makes sense when studying the pragmatic competence of children, i.e. their use of specific forms in specific contexts (for a study on the role of pragmatics in the acquisition of Russian aspect see Vinnitskaya & Wexler, 2001). Another reason why I favor a usage-based approach is that usage-based approaches do not assume innate functional categories. The mechanisms assumed for generalizing from early stages of categories to fully developed categories might be similar in nativist and usage-based accounts (cf. Olsen & Weinberg's rggg optimal learning strategy or Wexler & Manzini's 1987 subset principle and the prototype approach chosen by Shirai & Anderson, 1995) and both approaches to some extent may argue that a category builds up. However, the major distinction between these two types of theories is whether they assume aspect to be innate or not. Nativist approaches make thus 8 stranger claim. I therefore see a usage-based account as more parsimonious.

T o sum up, the comparison of the comprehension experiment (Stoll, 1998) and the two production experiment (Study I and Study a) shows in some detail how three major factors interact in the acquisition of Russian perfective aspect, namely: Aktionsart, complexity of context, narrative competence of the child, (which are measured here by the ability to sequence events). This supports Shirai's suggestion for the acquisition of Japanese tense and aspect morphology, namely, that we need to take a 'multiple factor perspective' (Shirai, 1998: 303). Of course, there are other factors-specifically, individual variation and caretaker input (Shirai & Andersen, 1995; Shirai, 1998; Aksu-Kog, 1998) -that need to be considered as well, as is true for language acquisition research in general but these other factors are outside the scope of the present study.

On a more general level, the results of this comparison call for detailed attention to different communicative contexts, with different discourse

i complexity in language acquisition, e.g. isolated utterances, short stories, free conversation data with diverse topics, etc. One context or one experiment can never be sufficient to judge whether a linguistic feature has been acquired; rather, we need a systematic comparison of a given form or category in different contexts. Only when we find that the form is equally well understood and produced in a range of diverse contexts with different levels of discourse complexity, can we confidently speak about the acquisition of a linguistic category.

I 823

STOLL

R E F E R E N C E S

Aksu-Koq. A. (1988). The acquisition of aspect and modality: the case of past referepice in Turkish. Cambridge : CUP.

Aksu-Kov, ,4. (1998). The role of input vs. universal predispositions in the emergence of tense-aspect morphology: evidence from Turkish. First Language 18, 255-80.

Antinucci. F. & Miller. R. (1976). How children talk about what happened. Journal of Child Laitguage 3. 167-89.

Bamberg, M. & Marchman. V. (1994). Foreshadoaing and wrapping up in narrative. In K. A. Berman br D. I. Slobin (eds). Rulatittg m!ents in narratiwe. Hillsdale. NJ: Erlbaum.

Berman. R. dr Slobin, D. I. (1994). Relating ewents in narrative: a crosslinguistic derelop- menial study. Hillsdale, NJ: Erlbaum.

Rickel, R. (2000). Unlogischer Aspekt: Zur Bedeutungsstruktur ron Aspekt und Aktionsart, besonders im Belharischen. In W. Breu (ed.), Probbme der Interaktion con Lesik iiizd Aspekt (ILA). Tiibingen: Niemeyur.

Bloom. L., Lifter, K. & Hafitz, J. (1980). Semantics of verbs and the development of verb inflection in child language. I~ngztage 56, 386-412.

Borer, H. & Wcxler, K. (19x7). The maturation of syntax. In T. Rocper & E. Williams (eds), i Parameter setting. Dordrecht: D. Reidel.

Breu, W. (1994). Interactions between lexical, temporal and aspectual meanings. Studies in Latrguage 18, 23-44

Bronckart, J. P. & Sinclair, H. (1973). Time, tense and aspect. Cognition z, 107-30. I Clahsen, H., Eisenbeiss, S. & Vainikka, V. (1994). The seeds of structure: a syntactic

analysis of case marking. In T. Hoekstra & B. Schwarte (eds). Language acquisition studies in generative grammar. Amsterdam: John Benjamins.

Dowty, D. (1979). TVurd, meaning and Montogue grammar .. the semantics of the verbs and timer in ge'erreratk:e se~nantics in Montague's PTQ. Dordrecht: D. Reidel.

Flier, M. S. (1985). The scope of prefixal delimitation in Russian. In M. S. Flier & A. Timberlake (eds). The scope of Sla7:ic aspect. Columbus, Ohio: Slavica Publishers.

Forsyth, J. (1970). A grammar uf aspect: usage artd mean in^ uf the Russian zwb. Cambridge: CUP.

Gagarina, N. (2000). The acquisition of aspectuality by Russian children: the early stages. ZAS Papers in Linguistics 15, 232-46.

Kritka, kt. (~ygz) . Thematic relations as links between nonlinal reference and teniporal constitution. In [. Sag & A. Szahlocsi (eds), Lexical matters 20-53. CSI,I.

14i, P. (1989). Aspect and Aktionsnri in Child Mandarin. Doctoral dissertation, University of Leiden.

Li, P. & Shirai, P. (2000). The acquisition of Iexicol and gmmmatical aspect. Berlin: Mouton dc Gruytcr.

Iieven, E. V. hI., Pine, J. 31. & Baldwin, G. (1997). Lexically-based learning and early grammatical development. Journal of Child Language 24, 187-zrg.

Lust. B. (1949). Universal grammar: the strong continuity hypothesis in first language acquisition. In W. C. Ritchie & T. K. Bhatia (eds), Hnrrdbook of CIzild Lariguagt Acquisition. San Dicgo: Academic Press.

Ulsen, M. B. & Weinberg, A. (1999). It~nateness and the acquisition of grammatical aspect ria lexical aspecl. Boston.

Pinker, S. (I q84). Langrcage Ienrnctbili~ nrld la~rgungc d~relopment. Cambridge, 31.4: Harvard University Press.

Sano, T. & Hyams, N. (1994). Axreenrent. fir~ite~ress, and the development i?f rzull subjects. Amherst, GLSA.

Shirai, Y. (1998). The emergcncc of tense-Aspect morphology in Japanese: universal predispositiun ? Firs! Lat~giragr IS, 281-3 10.

Shirai, Y. & Andersen, R. W. (rgy5). The acquisition of tense-aspect morphology: a prototype accuunt. La~zguagrgr 71. 743-62.

T E L I C S A N D I N C i R E S S I Y E S I N R U S S I A N

Stobin, D. I. (1995). Converbs in Turkish child language: the grammaticalization of event coherence. In M. Haspelmath & E. Konig (eds), Conwerbs in cross-linguistic perspective. Berlin : Mouton de Gruyter.

Stephany, U. (1985). Aspekt, Tempus und Modalitat. Zur Entwicklung der Verbalgrammatik in der Neugriechzschen Kindersprathe. Tubingen: Narr.

Stoll, S. (1998). The role of Aktionsart in the acquisition of Russian aspect. First Language 18,35178.

Stoll, S. (2001). The acquisition of Rzrssian aspect. Doctoral Dissertation, University of California. Berkeley.

Stoll, S. (2003). Kognitive Entwicklung und Aspekterwerb. In S. Haberzettl & H. Wegener (eds), Spracherwerb und Kunzeptualisiemg. Frankfurt am Main: Peter Lang.

Tomasello, M. (1992). First Verbs. Cambridge: CUP. Tomasello, M. & Brooks. P. L. (1999). Early syntactic development: a construction

grammar approach. In M. Barrett (ed.), The development of language. Hove: Psychology Press.

Van Valin, R. D. & La Polla, R. J. (1997). Syntax: stnrcture meaning and junction. Cambridge : CUP.

Vendler. 2. (1967). Linguistics in philosuphy. Ithaca, NY.: Cornell University. Vinnitskaya, I. & Wexler, K. (2001). The role of pragmatics in the acquisition of Russian

aspect. First Language zr, 143-86. Wagner, L. (zoox). Aspectual influences on early tense comprehension. Journal of Child

Lnnjpage 28, 661-81. Weist, R. M., Pawlak, A. & Carapella. J. (2004). Syntactic-semantic interface in the acqui-

sition of verb morphology, Journal of Child Language 31, 31-60. Weist, R. M., Wysocka, H., Witkowska-Stadnik, K. 8: Buczkowska, K. (1984). The defec-

tive tense hypothesis: on the emergence of tense and aspect in child Polish. Journal of Child Lat~guage 11. 347-74.

Weist, R. M. (2002). The first language acquisition of tense and aspect. In R. Salaberry & Y. Shirai (eds), The Lz acquisition of ter~se-aspect morphology. Amsterdam: Benjamins.

Wexler, K. (1999). Maturation and growrh of grammar. In W. C. Ritchie & T. K. Bhatia (eds), Handbook of child language acq~cisifion. San Diego: Academic Press.

Wexler, K. & Manzini, R. (1987). Parameters and learnability in binding theory. In T. Roeper & E. Williams (eds). Parameter setting. Dordrecht: Reidel.