clitic omission in european portuguese

CLITIC OMISSION IN EUROPEAN PORTUGUESE: * CORRELATION WITH NULL OBJECTS? JOÃO COSTA, MARIA LOBO, JAQUELINE CARMONA & CAROLINA SILVA 1. Introduction. There is a wide consensus in the literature on language acquisition that functional elements are acquired late, although there is crosslinguistic variation with respect to those elements that are produced later and to the rates of omission for different languages (see Radford 1994, Guasti 2002 for relevant data and for a review of the literature). Since pronominal clitics are elements whose syntax is dependent on the functional structure of the clause, it is expected that clitics are potential candidates for late acquisition. In this paper, we present a summary of recent discoveries on the acquisition of clitics, showing that clitics may be omitted or not depending on the language being acquired. Afterwards, we present the summary of the results of an experiment run with European Portuguese children in which accusative 3rd person clitics were elicited (Costa & Lobo 2006). The European Portuguese data show that, for this language, there is no positive correlation between clitic omission and the availability of past participle agreement in the adult language, contrary to the predictions made in Wexler et al. (2003). Bearing these results in mind, the aim of the paper is to compare two hypotheses that intend to explain the pattern of omission found in European Portuguese: feature checking vs. postsyntactic complexity. For a comparative evaluation of these two hypotheses, it becomes relevant to study the behaviour of children with non-accusative and with reflexive clitics. The goal of the paper is to present the results of two experiments: one eliciting reflexive clitics, and another experiment eliciting 1st, 2nd and 3rd person dative clitics. It will be shown that the different omission rates found allow for comparing the two hypotheses under evaluation, and favour the hypothesis in terms of post-syntactic complexity. *

The research developed for this paper was partly funded by the project “Técnicas Experimentais na Compreensão da Aquisição do Português Europeu”, sponsored by Fundação para a Ciência e Tecnologia (POCI/LIN/57377/2004).

Clitic omission in European Portuguese: correlation with null objects?

2

2.

Clitic omission crosslinguistically and analyses.

As referred to in the introduction, it is known that, in some languages, some functional elements may be omitted in the early stages of syntactic development. In Wexler et al. (2003), evidence is presented suggesting that clitic omission is not a universal phenomenon in language acquisition. These authors claim that there is clitic omission in French and Catalan, but not in languages like Spanish or Greek. It is further sustained that there is a double correlation: clitic omission is expected in languages with past participle agreement, and at the same ages in which root infinitives are found in child language. For example, in French, since there is past participle agreement in compound tenses, as shown in (1), clitic omission is expected: (1) Je les avais faites. I cl-acc-fem-pl had done-fem-pl This hypothesis makes clear predictions for European Portuguese. Since there is no past participle agreement in this language, clitic omission is not expected. This prediction was evaluated in Costa & Lobo (2005, 2006) for 3rd person accusative clitics. The elicitation experiment conducted in Costa & Lobo (2005, 2006) was modelled after Schaeffer (1997) (the same methodology was used in Wexler et al. (2003)). It was, however, adapted in view of the fact that European Portuguese is a null object language. Note that this specificity of European Portuguese prevents a clear evaluation of utterances in which a complement is not produced, since these do not constitute a clear case of clitic omission, but may, instead, be interpreted as instances of target-like null object. For this reason, the experimental protocol included a condition with contexts of strong islands in which null objects are not allowed. The inclusion of this context makes it possible to control for a context in which the omission of a complement is unambiguously taken as clitic omission. The results obtained in this experiment are summarized below: a) Portuguese children produce null forms in all contexts (simple clauses and strong island contexts), which reveals that there is clitic omission in European Portuguese; b) The prediction that European Portuguese should not have clitic omission because it is not a past participle agreement language was not confirmed;

João Costa, Maria Lobo, Jaqueline Carmona, Carolina Silva

3

c) Two major differences with respect to the languages studied in Wexler et al. (2003) were found. First, the omission rate found in European Portuguese is higher than in other languages (globally around 70%). Second, the age at which omission is found is higher, since 4 year olds still omit clitics. This age difference shows that there is no correlation with the ages at which children acquiring other languages produce root infinitives.

3.

Competing analyses for clitic omission.

In Wexler et al. (2003), in accordance with other studies conducted by Wexler, it is proposed that clitic omission is a consequence of a constraint subject to maturation, the Unique Checking Constraint. According to this constraint, a category may only enter a single feature checking relation. As such, in languages with past participle agreement, the clitic must check a D feature related to its own properties (against ClP), and enter a subsequent checking relation in AgrOP, which is reflected in the overt agreement markers. The impossibility for a category to enter more than one checking relation predicts that clitics will be problematic only in languages with past participle agreement. It is further predicted that past participle agreement, in the relevant languages, is in complementary distribution with clitics in the early stages of syntactic development. Since Wexler (1998) proposes that the Unique Checking Constraint explains the root infinitive phenomenon, it is further expected that there is an age correlation between the existence of root infinitives and clitic omission. At first sight, this analysis is problematic for European Portuguese, since, as we mentioned above, this language has clitic omission in acquisition, in the absence of a correlation with past participle agreement in the target grammar. However, as it was shown, the data are not entirely comparable, since the omission rates found are much higher in European Portuguese and since omission lasts until later in the process of acquisition. Taking these differences into consideration, Costa & Lobo (2006) propose that the omission found in European Portuguese is to be explained in different terms: as a consequence of post-syntactic complexity. Assuming that the availability of the null object construction in the adult grammar implies a specialization of the contexts in which clitic omission is not legitimate, and considering that, in European Portuguese, there is alternation between enclisis and proclisis (which, as shown in Duarte and Matos (2000), is problematic for children), the authors suggest that the clitic system is complex, since it implies making choices between different convergent syntactic outputs. Following ideas by Reinhart (1999), according to which, in comprehension, the choice between convergent derivations may create

4


processing problems, Costa & Lobo (2006) suggest that the choice between multiple convergent derivations may yield problems in production. In view of these two hypotheses, based on the effects of the Unique Checking Constraint (UCC) or in terms of complexity, it becomes important to find data enabling an explicit comparison between both hypotheses, since it would be legitimate to suppose that some variant of the UCC might be able to account for the European Portuguese data.1

4.

Comparing UCC and complexity.

In this paper, we present data from the production of reflexive clitics, dative clitics and 1st and 2nd person clitics. The rationale behind the research conducted is the following: if we find cases in which clitics do not freely vary with null objects, it is expected that complexity is reduced and production rates are higher. Crucially, this type of clitics enables an explicit comparison between the two hypotheses. The different types of clitics are similar in terms of feature specification, and it is defendable that they all enter more than one feature checking relation. According to the UCC, similar omission rates are predicted independently of the subtype of clitic. According to the complexity hypothesis, it is expected that clitics that do not alternate with null objects are less problematic, since these clitics do not imply a choice between multiple convergent derivations. Let us check the relevance of the different clitic types separately. Reflexive clitics differ from non-reflexive clitics, since the former do not alternate with the null object construction, as shown in the contrast between (2) and (3): (2)

A: B:

(3)

1

A:

E a Maria? and the Maria “What about Maria?” Não (a) vejo há que tempos. Not (her) see for a while “I haven’t seen her for a while” E o João? and the João

In Costa & Lobo (2005), for example, a hypothesis was advanced according to which the UCC was able to account for the European Portuguese data, assuming that clitics enter other types of multiple checking relations. Such a hypothesis was abandoned later since it implies questioning the maturational nature of the UCC.


B:

5

“What about João?” Não *(se) lava há que tempos. Not (himself) washes for a while “He hasn’t washed himself for a while.”

Dative clitics allow for comparing two competing hypotheses concerning the properties of the null object construction in adult grammar. According to Raposo (1986), this construction is available in 3rd person accusative contexts. A less restrictive view is presented in Costa & Duarte (2001), who contend that null objects occur in contexts of non-maximal VP. As such, they are legitimate substitutes for dative constituents. If it is shown that omission rates correlate with the availability of null objects, the data to be obtained may contribute to this debate, since Raposo’s view predicts that there is no similarity between accusative and dative contexts, while Costa & Duarte’s view predicts that there should be no major differences between the two types of clitics. Independently of this issue, there is consensus that null objects are ruled out in 1st and 2nd person contexts, as shown in (4): (4)

a. b.

Não *(me) vias há muito tempo. Not (me) saw for a while “You hadn’t seen me for a while.” Não *(te) via há muito tempo. Not (you) saw for a while “I hadn’t seen you for a while.”

4.1.

Elicitation of reflexive clitics.

In Costa & Lobo (2007), an experiment was conducted for the elicitation of reflexive and non-reflexive clitics. The methodology of elicitation was based upon Schaeffer (1997), and the following conditions were tested: a) Non-reflexive 3rd person singular accusative clitics; b) Reflexive 1st person singular accusative clitics; c) Reflexive 2nd person singular accusative clitics; d) Reflexive 3rd person singular accusative clitics. Unlike what was reported for the elicitation of non-reflexive accusative clitics, the strong island contexts were not tested, since reflexive clitics do not freely vary with null objects in any context. 24 children aged between 3 and 4 (average age: 3 years and 7 months) participated in the experiment. All children are from two kindergartens in the Metropolitan area of Lisbon. The protocol included 17 items (4 per each condition including reflexive clitics and 5 items for non-reflexive

6


clitics). The proclisis-enclisis alternation was controlled for, since the experiment included contexts favouring both clitic positions, which turned out to be irrelevant for matters of production. In the following table, we present the global results for the comparison between reflexive and non-reflexive clitics:2 Clitic

Null

DP

Reflexive clitics

47,4% (104/219)

41,5% (91/219)

11,1% (24/219)

Non-reflexive clitics

13% (13/100)

67% (67/100)

___

Table 1: elicitation of reflexive and non-reflexive accusative clitics

4.2.

Elicitation of dative clitics.

In Silva (in preparation), an experiment for the elicitation of dative clitics was conducted following a pilot study developed in Carmona & Silva (2007). The methodology of elicitation was similar to the one used in the previous experiment, and the following conditions were tested: a) Nonreflexive 1st person singular dative clitics in simple clauses; b) Nonreflexive 1st person singular dative clitics in strong islands; c) Nonreflexive 2nd person singular dative clitics in simple clauses; d) Nonreflexive 2nd person singular dative clitics in strong islands; e) Nonreflexive 3rd person singular dative clitics in simple clauses; f) Nonreflexive 3rd person singular dative clitics in strong islands. In simple clauses, the variation between proclisis and enclisis was controlled for. For each condition, two sentences were tested, and another two for controlling the proclisis and enclisis variation when this was relevant, which resulted in a total of 36 test items. We present the results for 11 children, aged between 3 and 4 (average age: 3 years and 7

2

In this presentation of the data, we do not present the results for the 1st and 2nd person conditions. This is due to the fact that the children revealed some problems in interacting with the toys, because of a problem in the experimental design. This resulted in several cases of transfer from 1st and 2nd person to 3rd person. For details, see Costa & Lobo (2007). These effects of the design were corrected and avoided in the experiment with datives.


7

months).3 All children are from a kindergarten in the Metropolitan area of Lisbon. In the following table, we present the global results for the production of dative clitics: Clitic 35/396

Null

8,8 %

205/396

DP

51,8%

20/396

Strong pronoun 5,1 %

136/396

34,3%

Table 2: elicitation of dative clitics In table 3, the rate of clitic production is presented according to context of elicitation: proclitic, enclitic or strong island: Enclisis Proclisis Strong islands Clitic 14/132 10,6 % 9/132 6,8 % 12/132 9,1% Null 78/132 59,1 % 70/132 53 % 57/132 43,2 % DP 5/132 3,8 % 5/132 3,8 % 10/132 7,6 % Strong 35/132 26,5 % 48/132 36,4 % 53/132 40,1% pronoun Table 3: results according to condition In tables 4-6, we present the results obtained according to the specification for person, in simple clauses (enclitic and proclitic position), and in strong islands: Enclisis 1st singular Clitic Null

2nd singular

3rd

7/22

31,8%

7/22

31,8%

0/22

0%

7/22

31,8%

11/22

50 %

13/22

59,1%

DP

0/22

0%

0/22

0%

3/22

13,6%

Strong pronoun

8/22

36,4%

4/22

18,2%

6/22

27,3%

3

The study in Silva (in preparation) tested all the different clitic types with 70 children aged between 3 and 6. For this paper, we selected the sample which enables a direct comparison with the results obtained for reflexive clitics.

8


Table 4: results for dative clitics according to person in enclitic contexts Proclisis 1st singular Clitic

2nd singular

3rd singular

2/22

9,1 %

7/22

31,8%

0/22

0%

8/22

36,4%

7/22

31,8%

15/22

68,2%

DP

0/22

0%

0/22

0%

3/22

15,6%

Strong pronoun

12/22

54,5%

8/22

36,4%

4/22

18,2%

Null

Table 5: results for dative clitics according to person in proclitic contexts

Strong islands st

1 singular Clitic

2nd singular

3rd singular

5/22

22,7%

5/22

22,7%

1/22

4,6%

8/22

36,4%

7/22

31,8%

11/22

50%

DP

0/22

0%

0/22

0%

7/22

31,8%

Strong pronoun

9/22

40,9%

10/22

45,5%

3/22

13,6%

Null

Table 6: results for dative clitics according to person in strong island contexts 4.3.

Synthesis and results.

The results of the two studies may be summarized along the following major aspects: a) The difference between the production of reflexive and non-reflexive clitics is significant, since there is a much higher rate of omission for reflexive clitics. b) The rate of omission of non-reflexive dative and accusative clitics is comparable. The only difference worthy to point out is that, in the dative condition only, and unlike what happens in accusative contexts, strong pronouns are produced.


9

c) A higher omission rate for 3rd person clitics was found than the one attested for 1st and 2nd person clitics. d) Similarly to what was found in Costa & Lobo (2005, 2006) for accusative clitics, the proclisis-enclisis variation did not play an important role in determining the production or omission of dative clitics. The results obtained appear to clearly favour the complexity hypothesis against a UCC based approach, since we found that, in cases in which there is no variation with the null object construction, the omission rates are considerably lower. This reduction in the omission rates was attested both for reflexive clitics, and for 1st and 2nd person clitics. Recall that a UCC based hypothesis would predict that one should not expect differences between clitic types, since for all of them one can argue that more than one feature checking relation is involved (at least, Case and Infl-attraction, as defended in Duarte and Matos 2000). As such, uniform omission rates were expected. According to the complexity hypothesis, it is expected that only clitics that freely vary with null objects be omitted at higher rates, since only these clitic subtypes impose post-syntactic choices between convergent derivations. The dative condition requires a further comment. In line with what was mentioned above, the results obtained may be interpreted as indirect evidence in favour of the less restrictive view on the null object construction defended in Costa & Duarte (2001), because children behave in 3rd person dative contexts just like in accusative contexts, displaying similar omission rates. Unlike what happens in accusative contexts, however, it is observed that, in the dative condition, strong pronouns occasionally emerge. These strong pronouns are present in the input (although they are considered substandard), which allegedly adds complexity to the system, since it makes it necessary to choose between a construction with a clitic, a null object and a strong pronoun. Finally, it is important to discuss the differentiated rates of omission. Since reflexive clitics and 1st and 2nd person clitics do not freely vary with null objects, one might expect that they would not be omitted at all. However, although the omission rates are different, there still is a quite high number of omissions, around 50% for reflexive clitics. It is, therefore, important to understand what the source of this omission rate is. In order to clarify this issue, it is relevant to compare these data with those obtained in Costa & Lobo (2005, 2006) for strong islands. In this study, it was shown that, in strong island contexts, there is also clitic omission, but at a lower rate than what was found in simple clause contexts. In these contexts, there was, instead, a higher rate of DP production. This result may be interpreted as revealing that children already display some

10


sensitivity to the special characteristics of this context, although they have not reached a stable knowledge of it. Likewise, it is possible to claim that children clearly distinguish reflexive from non-reflexive contexts. As such, we advance the hypothesis that the pattern of clitic omission found in European Portuguese is the reflex of an overgeneralization of the null object construction. In order for clitic production to become stable, and adult-like, children must have a steady knowledge of the contexts in which the null object construction is ruled out. Our data indicate that, in the stages under test, children already display some sensitivity to those contexts, although they have not yet reached a stable knowledge about the illegitimacy of null objects. Curiously, it is worth noting that the rates of omission in island contexts and in reflexive contexts are very similar, around 50%.

5.

Conclusion.

The studies reported in this paper on production of reflexive, dative and 1st-2nd person clitics aimed at obtaining data enabling a comparison between two alternative hypothesis for explaining the phenomenon of clitic omission in European Portuguese: in terms of feature checking and in terms of post-syntactic choices. It was shown that children omit less reflexive clitics than non-reflexive clitics, and less 1st and 2nd person clitics than 3rd person clitics. No difference was found between accusative and dative clitics. It was argued that the asymmetries detected between the different clitic types favour a hypothesis in terms of post-syntactic choices, since this hypothesis predicts a higher rate of omission only in contexts in which the clitic alternates with null object. A comparative analysis of the omission rates also allowed us to raise the hypothesis that clitic omission in contexts in which there is no alternation with null object is due to an overgeneralization of the null object construction. In general terms, the results obtained enable us to show that clitic omission is not a uniform phenomenon crosslinguistically and that some of the variation detected may be a consequence of the specific properties of the target language. If this conclusion is on the right track, it opens up the perspective that, in other languages, clitic omission may be a consequence of the existence of different subtypes of null object constructions.


11

References Carmona, J. & C. Silva (2007). A aquisição de clíticos acusativos e dativos em PE. In A. Coutinho & M. Lobo (eds.) XXII Encontro Nacional da APL. APL/Colibri, Lisbon, 199-210 Costa, J. & I. Duarte (2001) Objectos nulos em debate. In Razões e emoção. Miscelânea de estudos em homenagem a Maria Helena Mateus. Vol 1; INCM, Lisbon, 249-260 Costa, J. & M. Lobo (2005) A aquisição de clíticos em PE: omissão de clíticos ou objecto nulo?. In XXI Encontro Nacional da Associação Portuguesa de Linguística. Textos seleccionados. Lisboa: APL, pp. 285-293. Costa, J. & M. Lobo (2006) Clitic omission, null objects or both in the acquisition of European Portuguese?. In Selected Proceeding from Going Romance 2005. Costa, J. & M. Lobo (2007). Complexidade e omissão de clíticos: o caso dos reflexos. In A. Coutinho & M. Lobo (eds.) XXII Encontro Nacional da APL. APL/Colibri, Lisbon, 303-313 Duarte, I. & G. Matos (2000) Romance Clitics and the Minimalist Program. In Costa, J. (org.) Portuguese Syntax. New Comparative Studies. Oxford: Oxford University Press, pp. 116-142. Radford, A. (1994) The syntax of questions in child English. Journal of Child Language 21, pp. 211-236. Raposo, E. P. (1986) On the Null Object Construction in European Portuguese. In Jaeggli e Silva-Corvalán (orgs.) Studies in Romance Linguistics. Dordrecht: Foris, pp. 373-390. Reinhart, T. (1999) The Processing Cost of Reference-Set Computation: Guess Patterns in Acquisition. Schaeffer, J. (1997) Direct object scrambling in Dutch and Italian child language, UCLA Dissertations in Linguistics, 17. Silva, C. (in preparation). MA dissertation, Universidade Nova de Lisboa Wexler, K. (1998) Very early parameter setting and the unique checking constraint: a new explanation of the optional infinitive stage. Lingua 106, pp. 23-79. Wexler, K., A. Gavarró and V. Torrens (2003) Feature checking and object clitic omission in child Catalan and Spanish. In R. BokBennema, B. Hollebrandse and B. Kampers-Manhe (eds.) Selected Papers from Going Romance 2002.