The Cartography of Ibero-Romance Agrammatic Deficits

This paper aims at examining whether grammatical errors produced by Broca’s aphasics are a consequence of a selective impairment of functional categories in three closely related Ibero-Romance languages — Catalan, Galician, and Spanish — for which almost no work had hitherto been done. In addition, a reinterpretation will be proposed under cartographical terms (Cinque 1999, 2002, 2006, Belletti 2004, Rizzi 2004) of previous structural neurolinguistic models of agrammatic production, more specifically the Tree-Pruning Hypothesis (Friedmann 1994, Friedmann & Grodzinsky 1997, and subsequent work). Cartography has been applied to the field of language variation. However, the present article constitute a completely new use. Since the Tree-Pruning Hypothesis was based on a model of monolithic nodes, the application of the cartographic tree structure provides us with further insights about the degree of structural preservation or damage of functional categories.


The Departure Point
The departure point is the Tree-Pruning Hypothesis (TPH) of Friedmann (1994Friedmann ( , 1998)), Friedmann & Grodzinsky (1997, 2000), and much subsequent work.This hypothesis was first proposed after the observation of a clear dissociation between tense and agreement production in Hebrew and Palestinian Arabic agrammatic speakers.While tense was found to be impaired, agreement turned out to be spared in agrammatic subjects.This dissociation was accounted for in structural terms.According to the TPH, the functional heads C, T, or Agr may be underspecified in agrammatism.According to the authors, an underspecified node cannot project any higher (Friedmann & Grodzinsky 1997: 420).The possible sites of the deficit are represented in (1).
I am grateful to Cedric Boeckx, Anna Gavarró, Na'ama Friedmann, Stavroula Stavrakaki, and two anonymous reviewers for their comments and suggestions.I also acknowledge support from the projects HUM2006-13295-C02-01 (Ministerio de Educación y Ciencia) and 2009SGR1079 (Generalitat de Catalunya).
focus on production, some comprehension tasks have also been run to assess participants' performance aiming at a unified account across modalities.

Negation
Contrary to what is observed for multiple functional elements, where evidence shows a very robust pattern cross-linguistically, negation in agrammatism has given rise to a lot of debate due to conflicting evidence documented in the literature.While it has been found to be spared in Japanese (Hagiwara 1995), Hebrew and Palestinian Arabic (Friedmann & Grodzinsky 2000), and French (Lonzi & Luzzatti 1993), some studies report that it is damaged in English (Rispens et al. 2001).
To check the degree of preservation of negation in agrammatic Ibero-Romance, we ran a production task including 25 items containing simple tenses (simple present, imperfect, simple future) plus 25 items with complex forms and verbal periphrases.Such a design allowed not only to test production skills with respect to negation but also to check repetition skills for tense and agreement as well as complex verbal clusters.
As illustrated in examples ( 5) and ( 6) from Galician, subjects were asked to negate the declarative sentence produced by the experimenter.To do so, a negative marked had to be added to the given structure with no other modification required.

Target answer:
Os nenos non actuaban o martes.the boys not perform.IMP.3.PL the Tuesday 'The boys were not performing on Tuesday.' (6) Vós destes en frega-los pratos.Galician you take.PRET.2.PL in wash.INF-the dishes 'You took to washing the dishes.'

Target answer:
Vós non destes en frega-los pratos.you not take.PRET.2ND.PL in wash.INF-the dishes 'You didn't take to washing the dishes.' The results of these tasks, summarized in Table 1, show that, despite its functional nature, negation is mostly spared for mild agrammatics.Though crosslinguistic differences with the control group proved to be significant, correct responses in the experimental group were as high as 97.5%.The main error type we documented was the omission of the negative marker.To account for our results under the terms given by the TPH, we have to adopt an account based on the low generation position of negation, thus justifying its preservation.To do so, we will assume the complex system of positions postulated by Zanuttini (2001) and claim that agrammatic subjects can construct their negative structures relying on NegP4, situated in the lower portions of the IP-field. (7) NegP1 3 NegP3 3

Tense and Agreement
After testing negation, we proceeded with immediately close portions of the IParea, those for tense.In order to assess the performance of the agrammatic participants on tense and agreement, we analyzed the results of the negation task as for participants' accuracy in repeating the finite tense (simple present, imperfect, and simple future with first and third person singular and plural agreement) in sentences containing four to five words.The production results, summarized in Table 2, show that, while tense is damaged (with an error rate of 14.6% across languages), agreement is spared (2.5%).This dissociation turned out to be significant in a Wilcoxon signed rank test (Z = -3.318,p < 0.01).No differences were found across languages.As for agreement, statistically significant differences with the control group were not found.Despite being extracted from a repetition task, the results show a clear parallel with those documented in previous studies (Martínez-Ferreiro 2003, Gavarró & Martínez-Ferreiro 2007).The TPH provides us with exact predictions regarding the behavior of agreement, which is expected to be relatively spared.The fact that agreement is cross-linguistically less impaired than tense indicates that the two functional categories behave differently in agrammatism.In this work, along with Chomsky (1995 et seq.), we assume that AgrP is no longer an independent functional node but rather an operation taking place in a designated position lower than TP (past) (Gavarró & Martínez-Ferreiro 2007).
Regarding the nature of the errors, most were substitution errors being omissions restricted to the drop of the complete verbal form. 2 While tense errors revealed a consistent tendency towards the forms of the simple present (see Graph 1 below), the number of agreement errors was so scarce, that no default form was documented (as in Graph 2).
In addition to these results, substitutions of a finite verb by a non-finite verb form and main verb omissions were strikingly scarce.No non-finite forms were detected in Galician or Spanish and, in the case of Catalan, only two examples were registered.Participants also produced two omissions of lexical main verbs: one in the Catalan sample and the other in the Galician sample (representing 0.5% of the total number of errors).
Graph 1: Tense substitutions in agrammatic Ibero-Romance 2 Since the three languages under investigation display a stem-based morphology, omission of tense or agreement markers is banned (Grodzinsky 1990).

Graph 2: Agreement substitutions in agrammatic Ibero-Romance
Going back to tense errors, the present was not only the better preserved form (6% errors in Catalan and 2% in Galician and Spanish) but also the most productive form for substitution (62.8% of errors), followed by the future (27.9% of errors) and the past tense (9.3% of errors). 3Structurally, although Cinque's hierarchy does not provide us with a specific functional head for present tense, the results seem to point to the ordering in ( 8), where the present would occupy a low position in the TP-field.The assumption of the present as the default form in a framework just sensitive to the [±past] distinction would not predict the preference for future forms vs. past forms. 48) TP (past) 3 In addition to the production results, we also tested participants' abilities for tense recognition by means of a comprehension task which included 25 items with simple tenses (simple present, imperfect, simple future).Items paralleled the structure of those included in the repetition task with a length of four to five words.After hearing a sentence read by the experimenter, participants were asked to select the right match among the stimulus visually presented; both example sentence and stimulus picture are provided in (9).As illustrated in Table 3, out of 125 responses per language, there were 17.6% errors in Catalan, 13.6% in Galician, and 16.8% in Spanish.Misidentifications, as in the case of production, mainly leant towards the present according to the ordering TP (past) > TP (future) > TP (present) .This is illustrated in Graph 3:

Graph 3: Tense misidentifications in Ibero-Romance agrammatic comprehension
The statistical analysis revealed no significant differences between production and comprehension (Wilcoxon Signed Rank test, p > 0.05) even though the chance of providing the correct answer varied across tasks.While in the repetition tasks, subjects could opt for up to 6 simple tenses in indicative -simple present, simple past (except for Catalan in which this form is analytic), imperfect, pluperfect (only in Galician), simple future, and simple conditional -in the comprehension task, the chance probability was one in three (i.e.simple present, simple future, simple past).

Modals, Aspectuals, and Temporal Auxiliaries
The next step towards the characterization of the agrammatic Ibero-Romance IPfield was to observe the repetition skills displayed by both our pathological and our control sample with respect to complex verbal forms.To that aim, we took our evidence from the negation task.Since Galician lacks compound tenses, the experimental design had to be adapted. 5In the case of Catalan and Spanish, the test contained 12 periphrastic forms plus 13 compound tenses, while in the case of Galician, all 25 items were periphrastic forms.All six agreeing forms were included in the design.
A summary of the results has been plotted in Tables 4 and 5.The former indicates the degree of preservation of temporal auxiliaries.Insofar as these forms crucially depend on high parts of the IP-area, they were susceptible to impairment as we have already discussed for tense in main verbs.58.3% of the erroneous utterances lacked an auxiliary form.Differences were significant with control subjects at a level of p < 0.01 (Mann-Whitney U test, Z = -3.106).The type of errors documented in our sample has been listed in order of decreasing frequency (10).Table 5 shows the correct responses for periphrastic constructions.

5
The notation compound tenses vs. periphrastic forms will be used throughout this section, the former referring to clusters of temporal auxiliaries + past participle and the latter to clusters including modals or aspectuals (+ preposition) + non-finite verb forms.Notice that the traditional nomenclature 'temporal auxiliaries' has been preserved to trace a clear cut in between those forms exhibited by Catalan and Spanish but absent in the case of Galician.
According to Veiga (1991), the Galician system derives from protoromanic varieties previous to the temporalization of compound forms.But the ban against perfect auxiliaries has an exception, as in the case of Portuguese, with the verb ter (from Latin tenere 'to have') which may be used as an auxiliary verb: The use of the auxiliary ter 'to have' in Galician substitutes the absence of the auxiliary haber 'to have' thus constituting a true morphosyntactic/functional form (see Giorgi & Pianesi 1997   The results indicate a clear increase in the number of errors with modal and aspectual auxiliaries (57% preserved across languages, 138 out of 245 responses) with respect to tense (15% errors) and with respect to temporal auxiliaries (18.5% errors), even though statistical differences (at a 5% level) were only found for Galician.
Regarding the nature of errors, 79.2% had to do with a simplification of the verbal cluster.A detailed analysis according to frequency is given in Graph 4:

ModP ability 3 […]
Errors were broken down by category and summarized in  These results do not seem to support Cinque's hierarchy with respect to modal and aspectual heads.However, it might also be the case that portions of the structure including immediately consecutive functional categories (as in the case of modals and aspectuals) may be damaged 'as a block' without reflecting an increasing level of difficulty -in line with Chinellato's (2002) Field Damage Hypothesis.Further testing may provide a better understanding on this specific issue.

Clitics
The last category observed in the IP-area was that constituted by clitics, including both object clitics and reflexive forms.In order to check how agrammatic deficits affected the highest portions of the IP, we tested both our experimental and our control populations for their production and comprehension skills.
The languages under investigation show considerable formal variation at this respect.While in Catalan and Spanish, reflexive and non-reflexive clitics appear in pre-verbal position with finite and post-verbally with non-finite forms and imperatives (12), in Galician, they tend to appear in post-verbal position (13).( 12 Exceptions to the Galician enclitic pattern can be found with negation, with most subordinate constructions, with quantifiers, and with focalizations.In addition, in infinitival clauses, there are contexts in which both patterns appear in free variation (Uriagereka 1995, Raposo & Uriagereka 2005).
In order to be properly uttered, clitics, with independence of their position in relation to the verb, crucially depend on the projection of a functional projection residing between the CP-and the IP-areas -namely F (Ledgeway & Lombardi 2005, Raposo & Uriagereka 2005).To obtain production data that allow us to test the behavior of these forms in agrammatic speech, we ran an elicited production task with the support of pictures that included 13 items aimed at eliciting object clitics (14) plus 12 items for reflexive forms (15).The comprehension task was a sentence-picture matching task again including 13 items for object clitics and 12 items for reflexive forms supported by pictures.These items were parallel in structure to those included in the production task as for number of words and clitic position.An example is given in ( 16): (16) La hermana mayor la suena.Spanish the sister older her blow-PRES.3RD.SG 'The older sister is blowing her (younger) sister's nose.' The general results have been summarized in Tables 7 and 8   Clitics were found to be compromised, albeit to a varying degree.A significant dissociation was found between object clitics and reflexives in production: While reflexives reached a maximum correctness rate of 85% for all three languages, object clitics were more severely impaired.Significant differences between object clitics and reflexive pronouns (Z = -3.409,p < 0.01) were found in a Wilcoxon Signed Rank test.
Comprehension results show that, despite impairment with respect to controls, the percentages of correct responses are higher than those for production.But, as happened with the results for tense, the fact that the pattern of errors is shared -with 3 rd person object clitics more severely affected than reflexive forms -has implications for the picture of agrammatism offered in studies like Grodzinsky (1990Grodzinsky ( , 2000) ) or Friedmann & Grodzinsky (1997), where production is characterized in a way completely different from comprehension.In agreement with Luzzatti & Guasti (2000) and Friedmann (2006Friedmann ( , 2008)), the proper description of the deficit should hold across modalities.
However, a possible strategy allowed by the design can have favored the interpretation of object clitics.As shown in ( 16), the materials included a picture with one participant and a picture with two participants.If it was the case that participants suffer from a general deficit across reflexive and non-reflexive forms, they would not get any information regarding the clitic and be left with a partial utterance (e.g., the girl wash -neither her nor herself).Such a move would have favored the selection of the option with the two participants and, as a conesquence, fewer errors in object pronouns and a higher number of errors in the interpretation of the reflexive forms would have been detected.Nevertheless, our production task indicates that reflexive forms are preserved cross-linguistically up to 85%, meaning that participants would have more chances to correctly identify the reflexive form than the object clitic (e.g., the girl wash -either her or Ø).This would explain the 92.2% correct comprehension of reflexive forms and, since the absence of clitic would still lead to the selection of the picture with two participants (in this case the correct answer), the dissociation in between object clitics in production and comprehension. 6 As for the nature of errors, out of the 141 errors, 82% of the agrammatic responses lacked a clitic.A list of errors in order of decreasing frequency is provided in ( 17 Additionally, despite the fact that these forms are attached to the verb enclitically in the case of Galician and proclitically in Catalan or Spanish, neither differences among languages nor ordering errors were attested. The high structural portions of the syntactic tree involved in the production of clitics (18) -relatively higher than verb morphology -make them susceptible to impairment, as predicted by the TPH (Friedmann 1994 et seq.).However, structural considerations would not explain the dissociation observed between clitic forms.Since the structural position would be the same for reflexives and object clitics (Uriagereka 1995) -be they enclitic or proclitic -, the dissociation may be attributed to: (i) absorption of one argument in the case of reflexive forms (Belletti 1982, Grimshaw 1982, among others), (ii) the licensing of a pro-object in the case of 3rd person object clitics (e.g.Jaeggli & Safir 1989), or (iii) the existence of inherently reflexive lexical entries for some verbs (Reinhart & Reuland 1993) or the cluster reflexive + verb as a result of a lexical operation (not a syntactic one, hence not susceptible to impairment in agrammatism).

Summary of Findings -IP-Field
At this point, before abandoning the IP-domain, we can draw some interim conclusions.According to the summary of the mild agrammatic data represented in Graph 5, the IP-field of Ibero-Romance agrammatics is selectively impaired.The pattern of damage, attributed to syntactic factors, can be accounted for in structural terms.

Graph 5: Ibero-Romance mild agrammatic errors in the IP-field
The results also show a striking similarity among Ibero-Romance varieties, confirmed by statistical tests (Graph 6).This is attributed to the similarity in the grammatical system of the three languages under investigation.
Graph 6: Ibero-Romance mild agrammatic production Both Graph 5 and Graph 6 confirm that there is an increasing percentage of errors as a function of the structural position an element occupies in the syntactic tree, with relatively low functional categories better preserved than higher ones.For mild subjects, almost completely spared abilities were detected for sentential negation and agreement, but high error rates were documented for modals, for aspectuals, and for object clitics.
The data are consistent with a structural account along the following lines -with agreement as an operation taking place in positions lower than TP (past) : Agrammatism offers an argument for the low position of both T (pres) and Neg in a cartographically developed tree structure.
Our findings were also replicated with a moderate agrammatic subject.As expected by hypothesis, the pattern of deficit turned out to be the same as that for his mild agrammatic counterparts but with increased percentages of errors.The results are summarized in Table 9 and Graph  Graph 8, finally, summarizes the IP-field of the three populations under investigation.
Graph 8: Degrees of severity in agrammatic Ibero-Romance

The CP-Field
So far, we have assumed the TPH and sentential structure proposed by cartographic approaches to the syntactic representation (Cinque 1999 and further collected in Cinque 2002, 2006, Belletti 2004, and Rizzi 2004, for example) and provided evidence for the inner structure of the IP-field based on agrammatic Ibero-Romance.In this section, we will focus on the structures crucially depending on the left peripheral area.Two constructions will be investigated: interrogatives and subject embeddings.

Interrogative Sentences
In order to obtain data on question production, we ran an elicitation task which included 25 items: 13 aimed at eliciting a wh-question and 12 designed to force yes/no-questions.In addition, wh-questions were controlled for their distribution in seven adjunct and six argument questions as well as for the methodology used to elicit them.Two types of tokens were introduced in the design.Based on Friedmann & Grodzinsky (2000), type I tokens did not include the wh-word in the instructions given by the experimenter (20), while in type II tokens, inspired by Crain & Thornton (1998), participants were presented with the wh-word (21).An example of the elicitation method for yes/no-questions is reproduced in ( 22). ( 20 In addition to the production task, a comprehension task aimed at checking participants' command of wh-words was run.The 25 tokens of the picturesentence matching task included 13 pictures for (subject and object) question comprehension (23) and 12 tokens for wh-word comprehension (24).Our results show that production and comprehension skills in both wh-and yes/no-questions differ between agrammatics and control speakers.The general production results for the agrammatic participants have been plotted in  Table 11 shows a dissociation between question types which turned out to be statistically significant at 5% (p < 0.05, Z = -1.993).Damage was seen to be more severe in the case of wh-questions than in yes/no-interrogatives.Moreover, the most frequent strategy used by subjects to overcome problems in wh-question formation was replacing them with a yes/no-question (see Graph 9).Neither methodology nor the argument/adjunct distinction affected the outcome significantly.
Friedmann & Grodzinsky account for this dissociation in terms of presence vs. absence of CP, with yes/no-questions claimed to be rooted in TP and hence more accessible for agrammatic participants.However, according to Suñer (1994), a non-overt element heads every yes/no-question in Spanish (correctly produced at 67.8% cross-linguistically in our mild agrammatic sample).The phonetically unrealized operator is claimed to be rooted in the CP-area.In addition, for the population under investigation, wh-questions were correctly produced at 52.3%, indicating that access to CP is not completely banned.Rizzi (2002), for example, provides us with arguments for a dissociated position for the two question types included in the CP-area, with Int(errogative)P and Foc(us)P as the crucial nodes (25).

FinP
According to Cruschina (2007), the interrogative operator would occupy the position Int (higher than Foc -landing site for wh-elements) where it is basegenerated together with other elements such as why.Nevertheless, the claim that null interrogative operators in polar questions are base-generated in Int is problematic for a truncation account, since it would attribute a higher position to yes/ no-questions than to wh-questions (Int > Foc).
Regarding the errors in yes/no-questions (illustrated in Graph 10), the agrammatic results showed a revealing pattern indicating a clear trade-off in between yes/no-and why-questions.
Graph 10: Distribution of errors in yes/no-questions The cartographic approach predicts the use of why questions in substitution for yes/no-questions in Catalan (5/58 errors in our data) and Galician (13/58) straightforwardly.The fact that only why, but no other wh-operator, enters into competition with the production of yes/no-questions may indicate that the null operator in yes/no-questions and why compete for the same structural position, and that this position is different from the one for ordinary wh-operators.In our Ibero-Romance sample, the overt operator why seems preferable for some agrammatic participants with respect to the null operator of a yes/no-question, raising the question of the role of overt vs. null elements.This also applies to the uses of Com és que 'how is it that' found in the Catalan data.
In addition to the wrong responses, an analysis of correct answers including an overt subject revealed that there is a clear dissociation in between experimental and control subjects with respect to the strategy used to formulate yes/no-questions.Agrammatic subjects favor SV over VS, contrary to the pattern displayed by control subjects.Tables 11 and 12 summarize, respectively, the experimental and control data, illustrated in Graphs 11 and 12.The inversion of subject and verb seems to be systematically avoided either by using the non-inverted option (in the case of yes/no-questions) or through the insertion of some element compatible with the order SV (such as Com és que 'how is it that' to substitute for wh-questions).Evidence for difficulties with VS structures in agrammatism can also be found in the studies by Garraffa (2008) or Beretta et al. (1996).
According to Belletti (2001Belletti ( , 2004)), there are two requirements for inversion to take place: (i) a pro-element and (ii) a right-peripheral focus position located in the higher portions of the VP-field.Since the verb would occupy the same tense position in the TP domain, avoidance of the order VS can be attributed to a deficit in licensing a pro-element in pre-verbal position.Agrammatics may be opting for the alternative with an overt element in pre-verbal position in order to avoid the use of an expletive element.Avoidance of VS accounts for 94.5% of the correct answers with overt subjects in the data from our experimental group.Expletives (overt or null) do not contribute to the meaning of a sentence being susceptible to impairment in agrammatic speech.
Regarding comprehension results, even though comprehension of whquestions might have benefited from a non-syntactic structure based on the selection of the figure not addressed in the question (such as "Look for a figure other than the policeman" in the context of a question like Who does the policeman help?), 8 the results, summarized in

Embedded Sentences
The last piece of evidence included in our testing battery comes form the production of subject embeddings, another construction that crucially depends on the left periphery, tested by means of an elicitation task with pictures.The 25 tokens specifically designed for that aim included 24 subject relatives (26) plus one object relative ( 27).Our results indicate that the production of subject relatives was impaired.Significant differences were found in the contrast between experimental and control subjects (Mann-Whitney U test: p < 0.01, Z = -4.904).Even though the mean rate of errors was 36.39%,inter-subject variation was prominent (it ranged from 20% to 96% correct responses).Regarding error distribution, the classification of errors according to frequency has been illustrated in Graph 13 below.
With the scant data we have to date, we can only conclude that our Ibero-Romance agrammatic subjects either avoid or fail to produce well-formed embedded structures (at the level of non-pathological subjects) when these require the involvement of the CP-layer.In the case of embedded structures such as those under investigation, since the projection of ForceP -according to Rizzi (2002), the highest functional projection of the left-peripheral area -is required, structural accounts would straightforwardly predict the observed deficit.As with the IP-field, cross-linguistic differences were not found.
Graph 15: Ibero-Romance mild agrammatic production Even though access to the left-peripheral area is restricted, mild participants may present a certain degree of preserved abilities.This is observable through the contrast with the results for our moderate agrammatic who failed, for example, to produce a single subject embedding.However, in contrast to the mild agrammatic sample, the Catalan moderate agrammatic does what we would expect under the TPH ( Finally, Graph 17 summarizes the IP-field of the three populations under analysis.

Graph 17: Degrees of severity in agrammatic Ibero-Romance
With the limited data available, we can only conclude that our Ibero-Romance agrammatic subjects either avoided or failed to produce well-formed structures (at the level of non-pathological subjects) when such structures required the participation of the CP layer.However, a proper account for our results must involve some additional factors.

Conclusion
Our results indicate that most functional categories under investigation were not completely damaged in our sample of agrammatic participants.We account for both across-subject variation and within-subject variation in the following terms: (A) Non-pathological adult subjects are endowed with the resources to complete structures up to the left end of the left periphery.
(B) In the case of deficit, the ultimate height they reach decreases; however, this does not necessarily entail that it is constant to the same exact extent

( 9 )
El chico abrió el bote.Spanish the boy open.PAST.3.SG the jar 'The boy opened the jar.' 3 This error pattern (present > future > past) has also been documented for other languages such as Greek (Koukoulioti & Bastiaanse 2010).

(
23) A qui ajuda el policia?Catalan To who help.PRES.3.SG the policeman 'Who is the policeman helping?' Target response: Subject points to the tourist.(24) Què va menjar en Joan?Catalan what AUX.PAST.3.SG eat the John 'What did John eat?' Target response: Subject points to the plate of food.
these be.PRES.3.PL the bananas that cost.PRES.3.PL three euros 'These are the bananas that cost three euros.'Targetutterance: CatalanÉstes son os plátanos que custan dous euros.these be.PRES.3.PL the bananas that cost.PRES.3.PL two euros 'These are the bananas that cost two euros.' 8 I am grateful to one of the reviewers for the comments on this specific issue.
Graph 14: Ibero-Romance mild agrammatic errors in the CP-field

Table 1 :
Negation of simple and compound tenses

Table 2 :
Repetition of tense and agreement The results indicate that comprehension is also affected in mild agrammatics.

Table 4 :
Auxiliary verb repetition for a complete discussion of Portuguese).

Table 9 :
Notice that, as in sections 6 and 7 below, the performance of the moderate population is only based on the data from one Catalan moderate agrammatic speaker.
Moderate agrammatic IP-fieldGraph 7: Ibero-Romance moderate agrammatic errors in the IP-field SG go.INF to somewhere and you want.PRES.2.SG know.INF-the date 'I am going to go somewhere and you want to know the date.'

Table 12 :
Subject-verb inversion (control subjects) Table 13, show that agrammatic participants have preserved knowledge of wh-words.

Table 13 :
Comprehension of wh-words PRES.3.PL the flowers that see.PRES.3.SG the John des de la seva finestra.fromthehiswindow 'These are the flowers that John sees from his window.'SGthe.tree that see.PRES.3.SG the John des de la seva finestra.fromthehiswindow 'This is the tree that John sees from his window.'Thegeneralresultsare summarized in Table14.