Why Do Children Say Did You Went The Role of Do-Support

Why Do Children Say Did You Went?: The Role of Do-Support1 1. Introduction One central area of language acquisition research involves functional categories such as tense and agreement. Some researchers have argued that functional categories are completely lacking in child grammar (Lebeaux 1988, Radford 1990). Others have argued that they are partially represented (Clahsen 1990, Ingham 1998). Still others have argued that they are fully represented (Poeppel and Wexler 1993). All of these researchers have characterized children’s grammar primarily in terms of the omission of functional categories. The overproduction of functional categories has received relatively little attention. One such phenomenon, which has not yet been well accounted for, is tense-doubling errors. This study elucidates the contexts and reasons for the occurrence of doubling errors, a type of speech error observed in young English-speaking children. While tense and agreement are expressed only once in a grammatical sentence, in a doubling error, “tense and/or agreement is incorrectly expressed twice–once on the ‘fronted’ auxiliary and once on the main verb” (O’Grady 1997:166), as illustrated in (1) and (2) below (Children aged 1;10–2;6, in Hurford 1975): (1) *What’s that is? (Is is expressed twice.) (2) *What did you bought? (Both did and bought carry past tense.) Because doubling errors involve tense/agreement on both the auxiliary and the main verb, doubling errors can only be observed in environments that require both an auxiliary verb and a main verb. Therefore, yes/no questions, non-subject wh-questions, and negative sentences in current Standard English, which contain both an auxiliary verb and a main verb, are three possible environments in which to look for doubling errors. A common characteristic shared by yes/no questions and non-subject wh-questions is Subject Auxiliary Inversion (SAI), which is employed in current Standard English as a way to distinguish between statements and questions.2 SAI involves movement: moving the auxiliary into a pre-subject position. Chomsky (1965, 1993) proposed that there are two operations in movement. The first involves copying the target element (the auxiliary verb in the case of inversion) into a surface structure position. The second operation then deletes it from the original deep structure position. This mechanism is shown in (3) (O’Grady 1997:164).3 (3) Inversion via copying and deletion: Deep structure: they will go After copying: will they will go After deletion: will they ∅ go Because SAI involves the fronting of an auxiliary, the construction has to have an auxiliary and a main verb. Thus, constructions involving SAI are good candidates for doubling errors. In fact, doubling errors in constructions involving SAI in young English-speaking children were examined by Stromswold (1990). She reported a frequency of doubling errors of 0.4 % in question structures involving an SAI.4 1 This work was supported by Daiko Foundation research grant 5384. SAI was not a settled rule for creating interrogatives in Early Modern English (Radford 1997). Some varieties of English, such as Jamaican Creole (Radford 1997) and Hawai‘i Creole English (Carr 1972; Kent Sakoda, personal communication), do not have SAI. 3 Recent work in transformational grammar (e.g., Chomsky 1993, 1995) does indeed adopt this “copying and deletion” view of movement rules. 4 The 0.4% which Stromswold (1990) provided is calculated from a conflation of all SAI environments without classifying types of the environments. By focusing on different types of SAI, we will get a clearer view of doubling errors. See § 3 for the three SAI environments. 2 2 2. Previous analyses of doubling errors Previous studies have looked at two possible explanations for doubling errors: SAI and movement. Hurford (1975) attributed doubling errors to the child’s version of the SAI rule. This rule says that tense is copied to the target position without deleting the original tense, as described in (4) (this is different from the adult version of the SAI rule (cf. 3)). (4) SAI via copying without deletion: Deep structure: they After copying: will they  no deletion  Surface structure: will they will will go go will go Hurford considered doubling errors not as errors but as patterns that are licensed by the child’s grammar, and predicted doubling errors to occur in any sort of SAI environment, but not in other environments where SAI is not involved. I will refer to this as “the SAI hypothesis.” In contrast, Mayer et al. (1978) claimed that children incorrectly formulate some movement transformations as copying without deletion, as described in (5), and explained doubling errors as the result of this un-adult-like transformation. While considering doubling errors as licensed by the child’s grammar, Mayer et al. predicted doubling errors to occur not only in environments involving SAI but also in those involving any sort of movement formulated as copying without deletion. I will refer to this as “the movement hypothesis.” (5) Movement via copying without deletion: Deep structure: they not After copying: they are not  no deletion  Surface structure: they are not are are going going are going While both of these hypotheses helped identify general environments where doubling errors occur, this paper will show that both the SAI hypothesis and the movement hypothesis are insufficient to accurately account for the distribution of doubling errors. As an alternate solution, I propose that the main factor underlying doubling errors is the need for do-insertion. I will refer to this as the “do-insertion hypothesis.” Section 3 provides the details of the do-insertion hypothesis, the SAI hypothesis, and the movement hypothesis. Section 4 describes the methodology employed to evaluate the three hypotheses. Sections 5 and 6 present and discuss the results of this corpus study. Section 7 gives the conclusion. 3. The do-insertion hypothesis 3.1 Maratsos and Kuczaj 1978 Maratsos and Kuczaj 1978 is the first preliminary study to indicate that doubling errors occur more frequently in contexts that cannot be simply characterized by the involvement of SAI nor movement of a verbal element. Examining two English-speaking children’s naturalistic data, Maratsos and Kuczaj reported that the frequency of occurrence of doubling errors is about 15% in questions that begin with does (Does she eats it?) and about 10% in questions that begin with did (Did she went there?). However, doubling errors are rarely found in questions beginning with is (@ Is she is eating?5), are (@ Are you are eating?), or with a modal (@ Can you can finish?). Maratsos and Kuczaj interpreted these facts as children using the incorrect verb form or applying past morpheme affixes redundantly for did-questions and does-questions, inserting a redundant auxiliary for repeated modal questions, and using a common grammatical sequence (eg. NP-is) for is-questions and are-questions. Table 1 summarizes the results described by Maratsos and Kuczaj. While all of the five types of yes/no questions involve SAI and movement of a verbal element, doubling errors did not occur in all five types. (This fact motivated me to seek the characteristics of the environments of doubling errors.) 5 @ means unattested or very rare. Why Do Children Say Did You Went?: The Role of Do-support 3 Table 1. The doubling errors in various types of yes/no questions in two English-speaking children (D.C. and K.R., ages not specified in Maratsos and Kuczaj 1978). Yes/no questions Does NP V+3sg (does Q) Did NP Irr. past (did Q) Repeated is Q Repeated are Q Repeated Modal Q D.C (%) 8/60 (13.3) 7/40 (17.5) 0/56 (0) 0/15 (0) 0/0 (0) K.R (%) 1/17 (5.9) 5/32 (15.6) 1/38 (2.6) 0/19 (0) 0/0 (0) 3.2 Re-interpreting Maratsos and Kuczaj 1978 Note from Table 1 that doubling errors occur more frequently in yes/no questions that begin with an inflected form of do, but rarely in yes/no questions that begin with an inflected form of be or a modal; therefore, I reclassified these five categories of yes/no questions into three categories: do-yes/no questions (which begin with do), be-yes/no questions (which begin with be), and modal-yes/no questions (which begin with a modal). The reinterpreted results are shown in Table 2. 3.3 Three yes/no question types Table 2 shows that doubling errors occur far more frequent-ly in do-yes/no questions than in other yes/no questions. Under the assumption of Chomsky 1991, the crucial defining characteristic of do-yes/no questions is that they involve do- insertion, unlike other yes/no questions (see Table 3). The detailed derivations of the three types of yes/no questions are described in the following subsections to clarify how each theory applies to the three types of yes/no questions. Table 2. The interpretation of the doubling errors reported by Maratsos and Kuczaj (1978) in my classification of yes/no questions. Yes/no questions Do-yes/no question Be-yes/no question Modal-yes/no question D.C (%) 15/100 (15.0) 0/71 (0) 0/0 (0) K.R (%) 6/49 (12.2) 1/57 (1.8) 0/0 (0) Table 3. The differences in the formation of yes/no questions. Yes/no question types Examples Formation of yes/no questions Do-yes/no question Do you eat it? Be-yes/no question Are you eating? Modal-yes/no question Will you eat it? Agr to I, do-insertion into I, I to C, shown in Figure 1. V to Agr, Agr to I, I to C, shown in Figure 2. Agr to I, I to C, shown in Figure 3. 3.3.1 Do-yes/no questions To create do-yes/no questions, abstract Agr is moved to the I position, and according to Chomsky 1991, do is inserted in the I position. The inserted do, instead of the main verb have, is then moved to the C position (in the process of SAI). This process is shown in Figure 1. 4 Figure 1. The derivation of do-yes/no questions in the framework of Chomsky 1991. CP C’ C | Do I NP | you I’ I AgrP I to C movement Agr (SAI) Agr to I movement VP V | eat do-insertion NP it Thus do-yes/no questions involve do-insertion, SAI, and movement of the auxiliary verb. Therefore, all three hypotheses (the do-insertion hypothesis, the SAI hypothesis, and the movement hypothesis) predict doubling errors in do-yes/no questions. 3.3.2 Be-yes/no questions On the other hand, the processes for creating be-yes/no questions and modal-yes/no questions do not involve any operation that inserts additional elements, because the be-verb and the modal can themselves to be inverted in current Standard English. The be-verb and the modal are moved to the left of the subject. The process for creating be-yes/no questions is shown in Figure 2. The be-verb is moved from the V position to the C position (through the Agr position and the I position). Figure 2. The derivation of be-yes/no questions in the framework of Chomsky 1991. CP C’ C | Are I NP | you I’ I AgrP I to C movement Agr (SAI) Agr to I movement VP V V to Agr movement VP eating Be-yes/no questions involve SAI and movement of the auxiliary verb but no do-insertion. Therefore, the SAI hypothesis and the movement hypothesis predict doubling errors in be-yes/no questions, but the do-insertion hypothesis does not. Why Do Children Say Did You Went?: The Role of Do-support 5 3.3.3 Modal-yes/no questions The processes for creating modal-yes/no questions are shown in Figure 3. Abstract Agr is moved from the Agr position to the I position where, according to Chomsky 1991, the modal is base-generated. The Agr and modal are together moved to the C position in the process of SAI. Figure 3. The derivation of modal-yes/no questions in the framework of Chomsky 1991. CP C’ C | Will I to C movement (SAI) I NP | you I’ I AgrP Agr Agr to I movement VP V | eat NP it Modal-yes/no questions involve SAI and movement of the auxiliary verb but no do-insertion. Therefore, the SAI hypothesis and the movement hypothesis predict doubling errors in modal-yes/no questions, but the do-insertion hypothesis does not. In summary, there is a crucial difference among the three types of questions in their derivations: while be-yes/no questions and modal-yes/no questions do not involve any operation that inserts an element, do-yes/no questions involve the operation of do-insertion. Based on this defining characteristic and the results shown in Table 2, it is natural to assume that the occurrence of doubling errors is related to do-insertion. 3.3.4 Summary of predictions The contrast among the predictions of the three hypotheses is summarized in Tables 4, 5, and 6, where shadowed boxes indicate cases in which high percentages of doubling errors are predicted by each hypothesis. Table 4. The doubling error environment in yes/no questions predicted by the do-insertion hypothesis. Yes/no question types Do Be Modal Examples Do you eat it? Are you eating? Will you eat it? the do-insertion hypothesis. Table 5. The doubling error environment in yes/no questions predicted by the SAI hypothesis. Yes/no question types Do Be Modal Examples Do you eat it? Are you eating? Will you eat it? the SAI hypothesis Table 6. The doubling error environment in yes/no questions predicted by the movement hypothesis. Yes/no question types Do Be Modal Examples Do you eat it? Are you eating? Will you eat it? the movement hypothesis 6 Among the three hypotheses, only the prediction of the do-insertion hypothesis is consistent with Maratsos and Kuczaj’s (1978) data shown in Table 2: doubling errors occur frequently only in do-yes/no questions. This indicates that the cause is not just SAI or movement in general (which are involved in creating all three types of questions). Rather, it is do-insertion that is peculiar to the process of creating do-yes/no questions among the three types of yes/no questions. However, using aggregated data from Maratsos and Kuczaj 1978 without access to the original data is not very convincing. Furthermore, the authors did not provide the background information about the informants, such as ages or the dialect(s) of English involved. For the do-insertion hypothesis to be considered convincing, the results from Maratsos and Kuczaj 1978 need to be confirmed with a well-recognized large pool of data such as the CHILDES (Child Language Data Exchange System) data (MacWhinney 2000). Thus, one of the goals of this study is to examine a large pool of data, for which background information is available, to confirm this asymmetry found in Maratsos and Kuczaj’s data. The second goal is to study a second environment (negative declaratives) not studied by Maratsos and Kuczaj, which will help us to differentiate among the three hypotheses. I turn to a description of negative declaratives in the following subsections. 3.4 Three negative declarative types The examination of doubling errors in yes/no questions as discussed above will not allow us to isolate the effect of do-insertion from SAI and movement. Do-yes/no questions involve not only do-insertion but also SAI and movement. However, the examination of doubling errors in negative declaratives will allow us to isolate the effect of do-insertion from SAI, since do-negative declaratives involve only do-insertion and neither SAI nor movement. Therefore, examining doubling errors in negative declaratives will allow us to factor out the effect of SAI and movement. Examining negative declaratives will also allow us to differentiate between the SAI hypothesis and the movement hypothesis. While the two hypotheses make the same prediction in yes/no questions, they predict differently for negative declaratives. For this study, negative declaratives are classified into three types: do-negative declaratives (which have do to the left of the negative), be-negative declaratives (which have be to the left of the negative), and modal-negative declaratives (which have a modal to the left of the negative). Table 7 shows the derivation of the three types of negative declaratives. None of the three types of negative declaratives involve SAI, but all involve movement of an auxiliary to the I position. Only do-negative declaratives involve do-insertion.6 Table 7. The difference in the formation of negative declaratives. Negative declarative types Examples Do-negative declarative You do not eat it. Be-negative declarative You are not eating. Modal-negative declarative You will not eat it. Formation of negative declaratives Agr to I, do-insertion into I shown in Figure 4. V to Agr, Agr to I shown in Figure 5. Agr to I shown in Figure 6. The detailed derivations of the three types of negative declaratives are described in the following subsections to clarify how each theory applies to the three types of negative declaratives. 3.4.1 Do-negative declaratives To create do-negative declaratives, Agr is moved to the I position, and do is inserted into the I position. This process is shown in Figure 4. 6 Be-negative declaratives involve V movement, whereas the other two involve Agr movement. This difference is not important to the purpose of this paper, and the results in §5 will show that there is no effect of verb type. Thus, although be and modal verbs belong to different verb types, they do not behave differently with respect to the occurrence of doubling errors. Why Do Children Say Did You Went?: The Role of Do-support 7 Do-negative declaratives involve do-insertion but neither SAI nor movement of an auxiliary verb. Therefore, only the do-insertion hypothesis predicts doubling errors in this context. 3.4.2 Be-negative declaratives In contrast, the formation of be-negative declaratives and modal negative declaratives does not involve any insertion operation. To create be- negative declaratives, the be-verb is moved from the V position to the I position via the Agr position. This is shown in Figure 5. Be-negative declaratives involve movement of be but neither do-insertion nor SAI. Therefore, only the movement hypothesis predicts doubling errors in this construction. 3.4.3 Modal-negative declaratives To create modal-negative declaratives, abstract Agr is moved from the Agr position to the I position. However, the modal is base-generated in the I position. This is shown in Figure 6. Modal-negative declaratives involve neither do-insertion, SAI, nor movement of an overt verbal element. Therefore, none of the three hypotheses predicts doubling errors in this construction. Figure 4. The derivation of do-negative declaratives in the framework of Chomsky 1991. CP C’ C IP NP | You I’ I | do NegP Neg | not AgrP Agr Agr to I movement do-insertion VP V | eat NP it Figure 5. The derivation of be-negative declaratives in the framework of Chomsky 1991. CP C’ C IP NP | You I’ I | are NegP Neg | not AgrP Agr Agr to I movement VP V V to Agr movement VP eating 8 Figure 6. The derivation of modal-negative declaratives in the framework of Chomsky 1991. CP C’ C IP NP | You I’ I | will NegP Neg | not AgrP Agr Agr to I movement VP V | eat NP it 3.4.4 Predictions As shown above, none of the three types of negative declaratives involve SAI. Among the three types of negative declaratives, only be-negative declaratives involve movement of a verbal element (be is moved from the V position to the I position), and only do-negative declaratives involve the insertion of an element (do-insertion). This allows for clear distinctions in the predictions for each hypothesis: the do-insertion hypothesis predicts doubling errors only in do-negative declaratives, the SAI hypothesis predicts doubling errors in none of the three types of negative declaratives, and the movement hypothesis predicts doubling errors only in be-negative declaratives. The contrast among the predictions of the three hypotheses with respect to negative declaratives is shown in Tables 8, 9, and 10, where shadowed boxes indicate cases in which doubling errors are predicted by each hypothesis. Table 8. The doubling error environment in negative declaratives predicted by the do-insertion hypothesis. Negative declarative types Do Be Modal Examples You do not eat it. You are not eating. You will not eat it. the do-insertion hypothesis Table 9. The doubling error environment in negative declaratives predicted by the SAI hypothesis. Negative declarative types Do Be Modal Examples You do not eat it. You are not eating. You will not eat it. the SAI hypothesis Table 10. The doubling error environment in negative declaratives predicted by the movement hypothesis. Negative declarative types Do Be Modal Examples You do not eat it. You are not eating. You will not eat it. the movement hypothesis 3.5 Summary of predictions In sum, SAI is involved in all yes/no questions, but not in any negative declaratives. Movement of the verbal element is involved in all yes/no questions and be-negative declaratives. The involvement of do-insertion is restricted to do-yes/no questions and do-negative declaratives. That is, the SAI hypothesis predicts doubling errors in all yes/no questions, but not in any negative declaratives. The movement hypothesis predicts doubling errors in Why Do Children Say Did You Went?: The Role of Do-support 9 all yes/no questions and be-negative declaratives. The do-insertion hypothesis predicts doubling errors only in doyes/no questions and do-negative declaratives. These predictions are summarized in Tables 11, 12, and 13. Shadowed boxes indicate cases in which high percentages of doubling errors are predicted by each hypothesis. Table 11. The doubling error environment predicted by the do-insertion hypothesis. Do Be Modal Yes/no questions Do you eat it? Are you eating? Will you eat it? Negative declaratives You do not eat it. You are not eating. You will not eat it. Table 12. The doubling error environment predicted by the SAI hypothesis. Do Be Modal Yes/no questions Do you eat it? Are you eating? Will you eat it? Negative declaratives You do not eat it. You are not eating. You will not eat it. Table 13. The doubling error environment predicted by the movement hypothesis. Do Be Modal Yes/no questions Do you eat it? Are you eating? Will you eat it? Negative declaratives You do not eat it. You are not eating. You will not eat it. Maratsos and Kuczaj (1978) already reported that doubling errors were observed in do-negative declaratives but not in be-negative declaratives or modal negative declaratives, although no examples or quantitative results were provided. No study on doubling errors has been conducted related to the do-insertion hypothesis, including corpus studies or experimental studies. One of the reasons for this might be that researchers have looked at SAI environments without classifying the question type in calculating the frequency of doubling errors. For example, Stromswold (1990) reported that young English-speaking children made doubling errors with a frequency of 0.4 % in the entire SAI environment. However, if the types of yes/no questions are classified and do-insertion contexts are distinguished from non-do-insertion contexts, the asymmetry that I showed above might have been found. Thus, the corpus study reported in this paper will fill this gap. 4. Methodology The present study consists of corpus analyses of the distribution of doubling errors in yes/no questions and in negative declaratives. The analysis of distribution of doubling errors in yes/no questions aims to demonstrate the frequent occurrence of doubling errors in an environment involving movement of the verbal element, SAI, and doinsertion over an environment involving movement of the verbal element and SAI, but not do-insertion. On the other hand, the analysis of the distribution of doubling errors in negative declaratives serves to demonstrate the frequent occurrence of doubling errors in an environment involving movement of the verbal element and doinsertion over an environment involving movement of the verbal element but not do-insertion. Note that SAI is factored out in the second analysis. The contrast among the predictions of the three hypotheses has already been shown in Tables 11, 12, and 13. The subjects were selected from CHILDES as representative of current Standard English-speaking children. The sample includes one male and two females, from ages 1;6 to 5;1: Adam (2;3–4;10), Eve (1;6–2;3), and Sarah (2;3– 5;1) from Brown 1973. MLU is 1.829–4.973 for Adam, 1.524–3.437 for Eve, and 1.505–4.857 for Sarah. The MOR and COMBO programs were used for introducing codes into the transcripts and for selecting potential cases of doubling errors. Using CLAN (Child Language Analysis, MacWhinney 2000), all utterances in “V…?” and “…V… neg… ” were selected and examined by hand. In this corpus study, doubling of the same tense and number was counted as a doubling error only when the doubling was exact, i.e., mismatched tense was not considered. This was done because none of the three hypotheses predicts mismatched tense errors. (Additionally, this type of error was generally rarely observed.) I restricted myself to 3rd person singular for present tense for non-be-contexts, since only Uninflected do and uninflected main verbs are excluded from the counting of doubling errors, since it is impossible 10 to distinguish whether these are finite or non-finite based on their forms in English. This form allows us to see tense and number-person agreement for non-be-contexts in English. However, I did not restrict the analysis to any person and number for past tense, since past tense can appear with any person and number. For be-contexts, I did not restrict the analysis to any person, number, and tense, since English be has enough rich inflectional morphology to show person, number, and tense information. Both regular and irregular verbs were included in the examination. Echo yes/no questions were excluded from the examination. Ambiguous utterances and unintelligible speech were also excluded. 5. Results Table 14 shows that doubling errors occur in about 15% of yes/no questions that begin with do for all three children I investigated. However, they rarely occur in other types of yes/no questions, while all the three types of yes/no questions were produced by all the three children. The occurrence rate of doubling errors in Eve’s do-yes/no questions is higher than other the two children (33.3% in contrast to 14.6% in Adam and 18.2% in Sarah). However, Eve’s data are not very significant, since the total number of utterances is significantly smaller than that of the other two children. I also found that one of the three children (Sarah) made doubling errors at a frequency of 10.9% in donegative declaratives, but not at all in other types of negative declaratives,7 as shown in Table 15. The other two children did not produce doubling errors in any types of negative declaratives. Table 14. Result from yes/no questions. Yes/no questions Do Be Modal Adam 38/266 (14.6%) 1/382 (0.0 %) 0/391 (0.0 %) Eve 1/3 (33.3%) 0/23 (0.0 %) 0/8 (0.0 %) Sarah 20/110 (18.2%) 6/225 (2.7 %) 0/203 (0.0 %) Table 15. Result from Sarah’s negative declaratives. Negative declaratives Do Be Modal Sarah 19/174 (10.9 %) 0/10 (0.0 %) 0/321 (0.0 %) In sum, the asymmetry between do-contexts and non-do-contexts was confirmed in yes/no questions for all three children investigated in this study and in negative declaratives for one of the three children.8 6. General Discussion 6.1 Cross-linguistic markedness of do-insertion Through the examination of yes/no questions, doubling errors were observed in about 15% of yes/no questions that begin with do, while they were rarely observed in other types of yes/no questions. The results support the doinsertion hypothesis over the SAI hypothesis and the movement hypothesis. Through the examination of negative declaratives, it was found that one of the three children (Sarah) made doubling errors at a frequency of 10.9% in donegative declaratives, but not at all in other types of negative declaratives. This is strong evidence for the dosupport hypothesis over the SAI hypothesis, since do-negative declaratives involve do-support but not SAI. This also undermines the movement hypothesis because there is no movement of a verbal element involved in donegative declaratives. This fine-grained analysis of a well-known phenomenon reveals that doubling errors occur almost exclusively in do-contexts. I argue that this general asymmetry between the frequency of doubling errors in do-contexts and nondo-contexts is due to the demands of do-support itself, not to an incorrectly internalized SAI rule or to an incorrectly 7 Through the counting of doubling errors in be-contexts, it was observed that doubling errors are rare for both auxiliary be and main verb be. 8 A complication here, noted by William O’Grady, is that doubling in do patterns involves repetition of just a morphological feature (e.g., past tense agreement), whereas doubling in other patterns involves repetition of an entire word. This introduces another factor into the comparison that will ultimately have to be taken into account. Why Do Children Say Did You Went?: The Role of Do-support 11 formulated movement rule. This could be because do-support is a cross-linguistically marked process and a costly language-specific process. Do-insertion is peculiar to current Standard English,9 the only known SAI language among all the world’s languages. Tsunoda (1991) reported that out of 130 languages investigated, only 12 languages (German, French, Swedish, Norwegian, Danish, Dutch, Spanish, Czech, Hungarian, Serbo-Croatian, Italian, and current Standard English) are Inversion languages. Inversion languages move the verbal element occurring to the right of the subject in a declarative sentence to the left of the subject to create an interrogative sentence. Inversion is geographically particular to Europe and genetically specific to the Germanic, Romance, and Slavic branches of the Indo-European language Family and the Ugric branch of the Uralic language family. Among the 12 Inversion languages, only current Standard English requires that the inverted element be an auxiliary. The other 11 languages can create interrogatives by inverting the main verb. Tsunoda’s observation is summarized in Table 16. TABLE 16. Summary of Tsunoda’s observations (1991). Languages Current Standard English German, French, Swedish, Norwegian, Danish, Dutch, Spanish, Czech, Hungarian, Serbo-Croatian, Italian Other 118 languages Inversion in creating interrogative Yes Yes SAI in creating interrogative do-insertion Yes No Yes No No No No That is, as a strategy for creating interrogatives, the process of inversion is cross-linguistically marked, and the constraint that the inverted element has to be an auxiliary is highly marked. 6.2 Cost of do-insertion There is clearly a correlation between do-insertion and the constraint that the element to be inverted must be an auxiliary. Chomsky (1991) argues that this highly marked constraint triggers do-insertion. When there is no auxiliary to the right of the subject, do is inserted into this position as a “last resort” to accomplish inversion for interrogative formation. Consequently, the operation of inserting do, a by-product of this very marked constraint that the inverted element has to be an auxiliary, is also highly marked. Assuming that a marked operation bears a greater cost,10 do-insertion might be a type of operation that is costly enough to cause processing errors. The cross-linguistic evidence provided in §6.1 supports the notion that do is a marked language-specific operation. In the framework of the Minimalist Program, do-insertion is considered to be not a part of UG but a language-specific rule (Chomsky 1991). Chomsky also proposes that UG principles are “less costly” than language-specific principles. Thus, do-insertion is a language-specific process that bears a greater cost. I claim that the cost of do-insertion is the main factor underlying doubling errors,11 while the operation involved in question formation (e.g. SAI) is a secondary factor. 7. Concluding remarks This study evaluated the SAI hypothesis, the movement hypo-thesis, and the do-insertion hypothesis of the L1 acquisition phenomenon of doubling errors. The examination of yes/no questions and negative declaratives confirmed the predictions of the do-insertion hypothesis, but not those of the SAI hypothesis or the movement hypothesis. 9 Bonnie Schwartz pointed out that Bernese German has optional do-insertion (personal communication). The investigation of acquisition of optional do-support would be a useful topic for further study. 10 Here, “cost” is in terms of principles of economy, following Chomsky 1993. 11 Do is semantically null and not interpretable at LF. Thus, the high cost related to do-insertion may come from the semantic nullness of do, not the insertion operation per se. However, in English, inserted elements are always semantically null, such as the expletives it and there. It is impossible to distinguish a “semantically null” hypothesis from an “insertion hypothesis.” 12 An explanation for the non-occurrence of doubling errors in do-negative declaratives in Adam’s speech could shed further light on this phenomenon, but this is beyond the scope of this paper.12 The investigation of doubling errors in other do-contexts (wh-questions and emphatic do) is ongoing. Difficulty in acquiring do is also expected in second language acquisition. This is a fruitful topic for future research. Difficulty in acquiring expletives is also expected, since an expletive is considered to be an inserted element and to be semantically null, the same as do is in the late insertion model. 12 The non-occurrence of doubling errors in do-negative declaratives of Eve is not significant, since the total token of her do-negative declaratives is very small. This might be because Eve is very young.