Revision Strategy and Self-Diagnosis in Italian L1 and L2 Argumentative Essays

The present study investigates the effects of Revision Strategy (RS) on Argumentative Essay (AE) writing. Specifically, the study provides an analysis of 58 short AEs written by L1 and L2 university students. Two different levels of errors are examined: linguistic (Spelling, (Morpho)syntax, Lexicon, Syntax, Punctuation) and structure (Text structure and Coherence) errors, in order to measure how RS influences the texts. The comparison between RS and non-RS samples shows a significant difference in favour of texts with RS. In fact, we can surmise that RS positively influences text production both in L1 and L2 writing. In the third section we provide a quantitative analysis of the errors and we find some common issues within the specific rhetorical and linguistic


The Educational Context of Writing at University, Between First and Second Language
Taking into consideration the wide-ranging theories on academic writing (Tang 2012;Canagarajah 2013;Candlin, Crompton, Hatim 2015), in this paper we shall discuss short argumentative essays (AE) of university students. We will focus on one aspect of the writing process in particular, revision strategy (RS) i.e. the self-diagnosis of the text by the writer (Connor, Farmer 1990;Zito et al. 2007). Our purpose is to suggest guidelines to achieve a RS model for self-evaluation of the AE. This RS model is also oriented (but not only) to academic writing, based on a linguistic-educational theory and on textual quantitative and qualitative research. 1 For some time now, the debate on assessment has held a central role for scholars and linguists studying language learning: today evaluation and self-evaluation is a crucial point in SLA (Second Language Acquisition) theory (Aryadoust, Riazi 2017;Warchulski 2016;Oscarson 2014;Yang, Xu 2008;Harris 1997) and also in L1 and L2 writing theory (Casaneve 2013;Moeller, Creswell, Saville 2016). However, the general parameters of interest proposed by Hamp-Lyons (1990) are generally accepted: especially in regard to a) the attention to the structure of the test, b) the characteristics of the writer and c) the role of the person who reads (intended both as a reader and reviser).
Following the path marked by other linguistic works on academic writing in L2 (Hinkel 2002), we will focus on linguistic and rhetorical textual aspects as well on the general organization of information in the speech. We will not discuss para-textuality aspects (annotation, bibliographic indication etc.), nor, of course, specific content-related aspects, nor finally the socio-cognitive aspects (Barrot 2017).
The central question then becomes: "which linguistic and argumentative devices must writers consider during self-evaluation of an AE, in order to improve their writing skills?". The aspect of selfregulation (Zito et al. 2007) must, in fact, be considered as a central moment of the complex linguistic space in the learning of writing, which implies a virtuous circle from the writer, towards the writer (as a reader). And then, Corno (1999a, 197) also argues that autonomy and awareness in writing and reading are the most important features of developed writers, who must not only be able to produce a correct text but also reflect on the possibilities of reviewing and improving their writing.

Italian for Academic Purposes
Despite the needs of Italian L2 international university students (Fragai, Fratter, Jafrancesco 2017), there are still very few studies (see the decade-long research presented in Mezzadri 2017) on the "Italian L2 for academic purposes". Furthermore, the Italian language occupies an important place in European humanistic studies (Gay 2014;Hempel 2011). Conversely, many English-speaking countries works deal with academic writing (EAP) with particular regard to the knowledge required by a L2 writer (Silva, Yang, Wang 2017;Belcher, Braine 1995). These studies almost always have two objectives: they aim to provide some operational models and illustrate the results of experimental research (see Candlin, Crompton, Hatim 2015). Therefore, in the following analysis we propose a teaching model starting from some experimental data.
Let us turn to two very important and contrasting L2 SAE issues. These issues differentiate the AE from other communicative scenarios (Mazzotta 2015) oriented to the intercultural perspective (Lenci 2009), including university writing (Celentin 2013). On the one hand, this kind of text represents an area of linguistic expression in which no interference is allowed, unlike oral performance or less controlled writing contexts. That is, the linguistic output must be a manifestation of an excellent linguistic competence, and it must necessarily reflect the most advanced language knowledge; especially when the argumentative essay is the thesis: this necessity represents a great effort, as it equates the L1 and L2 writing tasks. On the other hand, the academic essay is a production that involves a large amount of time for planning and preparation; so every writing is not just a communicative moment, but also an instance for metalinguistic discussion and for RS application. In other words, the argumentative essay is the most precise L2 communicative act, but it is also the type of production that requires/allows for the greatest and most pondered planning. For these very reasons, self-evaluation allows some didactic achievements in common between L1 and L2 writers (Hamp-Lyons 1991).
The issue of university students' writing skills has been studied for many years in the context of Italian linguistic education (Corno 1999b;Lo Duca 2005). Piras and Picamus (2003, 223), e.g., fif-teen years ago argued that the Italian language of native university students is: acquisita per sentito dire, come certi testi di canzoni inglesi, in parte giusti in parte no, una lingua vista (non proprio letta) sui libri, ripetuta nei brevi episodi di verifiche scolastiche, ma mai assimilata veramente, fatta propria, usata come convinto strumento di comunicazione, veicolo di contenuti anche culturali.
A few years earlier, the research edited by Lavinio, Sobrero (1991, 27) suggested "a dramatic picture" of the Italian writers. Although more than twenty-five years have passed since the above-mentioned Italian studies, both the students' problem of thesis writing and the teachers' proposal of an effective writing model remain topical issues; in fact, our research presents a new scenario in which students seem to have the same problems with the academic AE and the thesis. The continued interest in this area is evidenced in prolific academic writing by Italian scholars (Balboni 2011;Boscolo 2001;Boscolo, De Marco 2008;Cerruti, Cini 2007;Dias et al. 1999;Zaccaro 2004;Andorno 2014;Gualdo, Raffaelli, Telve 2014;Piemontese, Sposetti 2015;Cacchione, Rossi 2016;Lubello 2017;Prada 2017;Alfieri 2017) and international scholars (besides the publications cited, Kutz 1998;Tang 2012; Journal of Second Language Writing, which offers various studies on L2 writing in L2, often academic writing oriented).

2
The Study

Participants
We use a 58 text corpus collected during an extensive six-year educational experience (2013)(2014)(2015)(2016)(2017)(2018). The quantity and the quality of the errors present in these texts will be the starting point of the following analysis. The 58 writers come from six different classes in six years (composed of a variable number between 7 and 12 students per year) of a workshop aimed at writing the master's thesis in Semiotics. We wish first to highlight that the six years period of the study, corresponding to six differentiated educational contexts, is an added value for the corpus representativeness. The classes were always composed of a native Italian speaker majority (36 native students, 62%) and a non-native minority (22 students, 38%, with different L1: Spanish, Portuguese, Chinese, Arabic, Russian). 2 All the students (31 males and 27 females) were between 23 and 25 years of age.
Overall, half of the students (29, including 18 natives and 11 nonnative students) did not receive any explicit instruction for the text editing, while the other half received an explicit self-assessment tool (RS) structured in seven levels of linguistic and rhetorical analysis.

RS Model and Methodology
Below [tab. 1] is a list of seven linguistic areas on which RS protocol is based. This list is built as a check list (see Appendix 1) that writers must use during self-diagnosis of their own texts. The RS consists of several questions aimed to focus the writers' attention on some aspects of their texts in 5 language areas (A) and 2 argumentative text structure areas (B). During the first 8 hours of the workshop, the contents of this list were explained in detail to all 58 students with various examples and exercises. That is, we listed and discussed the most common spelling errors, syntax errors etc. in university students AE; we also discussed some rules and strategies to avoid them. Hence, the only difference between the two groups is that only one group (29 students) was subsequently given an explicit list of questions for the self-correction of their text (RS protocol); that list was not given to the other 29 students group. 2 For L2 writers the level of Italian certification was advanced (QCER >/= B2), as required by the master's degree program. In addition, we point out that we will not specifically discuss in this article the errors due to interference from the mother tongue of non-Italian speakers, both because the interpretation is not always univocal, and because this is not the focus of this study. However, we will sometimes refer to some errors due to interference in the following paragraphs.
Obviously, these contents represent an approximation with respect to many possible errors of both L1 and L2 writer's interlanguage. However, it should be noted that the proposed questions try to tackle many of the most recurrent errors documented in scientific literature (cited in the previous paragraphs) that describes the problems of university students' writing.
The texts of the corpus have been elicited with the following writing task: starting from the reading of an argumentative / expositive text or from another textual content (photograph, video) -the students wrote a new argumentative text on the same subject, from 500 to 700 words long. The writers were asked to compose a thematic index, such as those in the examples in table 4 and table 5. 3 In summary, the students did the writing work in three stages.
1) The preparatory work of the text carried out in the classroom.
That included the analysis of the original text, the exposition and discussion of the individual writer's ideas, the individual structuring of an outline. 2) The first draft of the text, given as homework. Upon completion, the works were sent immediately to the teacher.
3) The student's revision of their own texts. The revision was based on the RS protocol (see Appendix 1) previously explained and commented on, or the writer were simply told to revise their text without the RS protocol.

Results. Errors
In graph 1 we can see the number of errors in the final text, considered in relation to the various areas and Italian L1/L2 language. As seen in the first and the second column, e.g., the final draft of the text without RS had 9 spelling errors for L1, and 32 spelling errors for L2 texts (in the first column); in the final draft of the text with RS there were 4 spelling errors for L1, and 31 spelling errors for L2 texts (in the second column). In regard to the data on graph 1, it should first be noted that, very predictably, there are many more errors in texts written by non-natives, although these texts represent only 38% of the corpus. Secondly, for almost all indicators, the number of errors in texts with RS is lower than those without RS, but this data must be better specified.
To begin with, in some areas there is not enough difference between the groups to be considered statistical proof of the success of RS: the first field is precisely Spelling, but also Syntax of Italian speakers and Text structure in the L2 texts. Secondly, the Lexicon of natives is a significant exception (one more error in the sample with RS). Moreover, it is difficult to interpret such small differences (+/-1) unequivocally because they can be reported as dissimilarities between two writers' samples (with and without RS). However, the overview of graph 1 shows an unequivocally positive difference in the overall quality of texts produced with RS. The students could not be the cause for the discrepancy between the two groups, especially considering that it is a sample obtained from classes distributed over six years. Finally, it is interesting to note that the best effects of RS are seen for Punctuation and Coherence, for both L1 and L2, and, contrary to L2 texts, a very positive result can then be seen in the Text structure for native texts.
In graph 2 we can observe the differences between the texts before (α) and after (β) revision.

Errors in L1 texts Errors in L2 texts
Graph 1 Total number of errors produced by L1 and L2 writers, and by group, without and with RS, in texts after revision First of all, it is very interesting to note that, even with some exceptions, the comparison between the texts before and after the revision shows improvement in the quality of the argumentative texts. More importantly, then, is that this improvement is more significant for texts with RS: in these texts there are fewer errors than in the text before revision and, above all, they show a more noteworthy difference. In fact, looking at the overview of the L1 and in L2 texts together, we can observe the following situation: -for Spelling: -8 errors (from 49 to 41) in texts without RS, compared to -8 errors (from 53 to 35) in texts with RS; -for the Lexicon: -2 errors (from 37 to 35) in texts without RS, compared to -9 errors (from 36 to 27) in texts with RS; -for Syntax: -12 errors (from 75 to 63) in texts without RS, compared to -30 errors (from 75 to 45) in texts with RS; -for Punctuation: -7 errors (from 57 to 50) in texts without RS, compared to -25 errors (from 47 to 22) in texts with RS; -for Text structure: -4 errors (from 19 to 15) in texts without RS, compared to -13 errors (from 21 to 8) in texts with RS; -for Coherence: -1 error (from 26 to 25) in texts without RS, compared to -12 errors (from 25 to 13) in texts with RS.
The only exception to this overview, although significant, is represented by the (Morpho)syntax, in which the difference between before (α) and after (β) the revision is higher for texts without RS: a total of -11 errors (from 75 to 64) in texts without RS, compared to -7 errors (from 67 to 60) in texts with RS.

Errors in L1 texts Errors in L2 texts
Another important aspect is that the revision process for Italian speakers sometimes results in less significant results. Above all, for the (Morpho)syntax and the Lexicon of texts reviewed with RS, the students have not actually reduced the number of errors. The same phenomenon is observed for Punctuation in L2 without RS. However, in this case, it is confirmation of the positive effect of an explicit self-correction tool (RS).
The difference in the number of errors between α and β often starts from a lower number of errors for texts with RS. This fact can be explained by the individual differences between the members of the two samples, but perhaps it can also be explained considering that the writers with RS were told they would receive the RS tool. This preliminary awareness may have positively influenced the texts written by the sample with RS.
At this point, we can also consider all the errors as a whole, observing graph 3. In the final editing the percentage distribution of texts with more or less than 5 errors shows a significant difference in favor of texts with RS.
Graph 3 Texts in the final version with less than 5 errors and with 5 or more errors According to the diagram [graph. 3] we can distinguish between L1 texts with and without RS, and L2 texts with and without RS. In absolute terms, these percentages correspond to the following numbers: -texts produced in L1 without RS: 3 texts with less than 5 errors in relation to all indicators compared to 15 with more errors; -texts produced in L1 with RS: 7 texts with less than 5 errors compared to 11 with multiple errors; -texts produced in L2 without RS: no text with less than 5 errors compared to 11 with more errors; -texts produced in L2 with RS: one text with less than 5 errors compared to 10 with more errors.
Considering the statistical data, and regarding the possibility of constructing well-written texts in relation to the seven indicators, we can surmise that RS positively influences text production both in the case of L1 and in the case of L2.
The evidence of only one L2 text compiled excellently (-5 errors) must be considered positively, especially given the difficulty of the writing task.
Finally, it seems to be useful, especially for the texts in L2, to provide additional data [tab. 2]. The average number of errors in all 22 texts in L2 is about 16 errors / text. But we find 17,8 errors / text in texts without RS, while the average is considerably lower in texts with RS: 14 errors / text. So, the difference between the average number of errors is 3,8/ L2 text (and the difference between the average number of errors is 3,3 / L1 text). 3 Linguistic Errors Analysis

Spelling
In this research corpus we observe that the most frequent errors in the L1 texts concern the accent -monosyllables and disyllables, accented or not ("in quell'epoca ci fù" / "non credevano piu" / "portare a se" / "cosi" instead of "così") -and the positioning of the apostrophe ("un evoluzione" / "qual'è il livello"). However, it should be noted that all these errors, except when they form existing words -the most frequent in all texts is the confusion between the accented adverbial "sì" and "si" pronominal (I) -may have been limited by the autocorrect feature used in the main word processors.
(I) Dobbiamo far si che gli altri sappiano quello che facciamo It is interesting to note that among the errors of the L1 texts with RS (only 4) none were underlined but not corrected automatically ("fù" / "piu"); the only errors present were those not underlined ("un evoluzione").
The same automatic spell checker also seems to greatly limit spelling errors in L2 texts, although such errors are more frequent and certainly much more differentiated than errors in native texts. So, first in the texts with RS there are the same error categories as the L1 texts, especially in accent placement, despite the autocorrect where the automatic data validation function underlines the error but does not provide the spelling correction ("perche in questo mondo" / "la situazione e piu complessa" etc.). Secondly, some typical L2 learning variety errors are predictably observed: for example, the confusion between the comma and decimal point in the Italian writing of numbers: "2,500". In the L2 texts, there are also errors typical of L1 substandard Italian, in the phrase segmentation, and in relation to phonetics ("Nel lambito"). It should be noted that all the selection errors between Capital/small letters have been eliminated after the revision with RS but not after the revision without RS.
Regarding spelling, it is also important to specify that the difference between error and mistake is difficult to define in the textual view we take here. Certainly, it is sometimes possible that some deviations that we have considered as errors (competence deviations) are actually mistakes (performance deviations). Moreover, this doubt is seriously reduced by the very nature of the RS and any processes of revision of the text. In fact, the error due to distraction or incorrect typing (typo) can be easily eliminated during the text revision process. In any case, however, the objective of this research is to consider the quality of the texts produced under different writing conditions, and not to evaluate the deep nature of individual deviations.

(Morpho)syntax: Morphology and Morphosyntax
Reading some data from the 140 total errors of our corpus, we find that most involve morphological agreement errors in L2 texts: e.g., "ci vuole un buono ragione" (with RS), "è stato un momento più complessa" (without RS). In L1 texts, we find this morphosyntactic error as well, but they can often be referred to an accidental performance deviation as in the example (II).
In L2 sample we find some consonant length errors ("la pennicilina", in a text with RS) that are at the same time spelling errors. And in both L1 and L2 samples we find derivation errors that result in an existing word ("letterario" in place of "letterale" in a text L1 with RS). They are, therefore, also lexical errors. Finally -especially in L2 texts -there are some errors related to verb conjugation or to the selection of the auxiliary ("ognuno era scritto la sua idea", in a text without RS), or to deviation in the personal pronoun paradigm ("Alcuni gli hanno chiamato 'Nazisti'") which are also syntactic errors. 4 However, we must also remember a type of specifically morphological error. It is not very frequent in our corpus but it is very typical of the learning varieties of both L1 children and L2 adults. We find the regularization of the form and the generalization with overextension of morphological rules: e.g., "gigantissimo" (L2 text without RS). In contrast, errors in the definition of gender and number that give rise to neologism ("molto usata e popolara", L2 text without RS) and errors in the choice of the article paradigm ("I effetti", L2 text with RS) are more recurrent in our corpus. They are typical of the learning variety in L2 or made by natives in substandard Italian. Finally we find, especially in L2 texts, some problems in the very irregular areas of the Italian language; e.g., in the presence or absence of the article before the possessive adjective (III) or in the phrasal expression (verb + proposition).
In the L1 texts, there is also a problematic and significant datum: there is no difference in the number of errors between the α version 4 We considered the wrong selection of the preposition as a morphosyntactic problem ("è rivolta dei [= agli] telespettatori"), but we considered the confusion between direct and indirect complement as a syntactic problem ("rinunciano la [= alla] possibilità"). and the β version of the texts with RS, while for the freely revised texts we can see a contained improvement (from 14 to 11): it is an event in contrast to the other data. It can be interpreted as a difficulty of many learners in the recognition of some morphosyntactic phenomena. However, there are only 8 morphosyntactic errors, distributed over three texts and not modified by the revision with RS; in almost all cases these are morphosyntactic agreement errors (IV) and in a single case it is a lexical derivation error ("nel panorama letterale").

Lexicon
It is very important to emphasize that lexical inadequacy includes many types of errors or inaccuracies in word selection. It concerns both Italian L1 and the L2 writers. Scientific language often needs univocal lexical choices, and this is an advantage especially for L2 writers, because it makes it possible to avoid having to choose between different meanings which are, in contrast, always present in the use of common or literary language. However, in the AE about non-technical topics, lexicon is much wider and embraces many variables of common language. So, in the L2 AE we can find some lexical transfers from L1 (V).
On the other hand, the international scientific literature has often pointed out that L2 writers use less varied sequences than native speakers in their texts (Salazar 2014;Chen, Baker 2010). A frequent type of lexical error -here analysed -is the paronymy (i.e., words used instead of others with a similar form but different meaning or synctactic use): this is common in the L1 texts (e.g.: "un altro evento indissolubile dai primi due", with RS) and in the L2 texts ("preferiva gli studi socievoli" without RS), with and without RS.
A second type of error, very common in all L1 and L2 texts, with and without RS, is to choose a word outside of its context of use: it is often a fixed-word combination or lexical collocation error ("danno [>fanno] schifo" in a L2 text with RS) as in the examples VI and VII: Mentre si considera dar un passo avanti…, si danno chi sa quanti in-dietro… (L2 text without RS). (VII) La scelta risiede principalmente nel forte interesse… (L1 text without RS).
However, these are also more general inaccuracies ("dure parole… sostenute", L1 text without RS); in the L1 texts, these are often errors of metaphor construction ("un secolo predominato da colpi di scena", L1 text with RS). The presence of this type of error, both with and without RS, allows us to suppose that they are the most difficult errors to detect, as well as to correct. It is perhaps because these are combinations of words infrequently used by the writers.
In the L2 texts, there are also noteworthy "false friends" -i.e. a negative L1 transfers (V) -very typical in interlanguages, even at an advanced level. Also, the invention of words that do not belong to the target language -an error between morphology and lexicon choices -are not comparable to L1 writers errors; they are often made through analogous affixation processes. In our corpus we find: "prepararsi diligenzamente al mondo del lavoro" (with RS), and "per quello che ho memoriato" (without RS). Incidentally, these are already highly studied phenomena, especially for suffixes -mente e -mento (Barni, Troncarelli, Bagna 2008). 5 In our sample, we can observe a significant difference in L1 texts, between with RS and without RS (from 21 to 12 errors); vice versa, the RS seems irrelevant to L2 texts (14 lexical errors with RS versus 15 without RS). It should be added that, as for Morphosyntax, in the L2 texts there is not a quantitative difference between α and β versions; that is, between before and after the revision. In regards to lexicon errors, this difference is neither in the L2 texts reviewed with RS nor in the L2 texts reviewed without RS. The only positive data for the lexicon, therefore, is the difference between +/-RS for L1 writers. This is a very problematic point, about which it is difficult to draw any conclusions since this ineffectiveness is incongruent with the revision effects on the other linguistic areas. We may perhaps interpret these data as follows: the category of choice of words is hardly questioned (between α and β) by the native writer or by the non-native writer, and only RS -that is, an explicit invitation to consider the lexicon -addresses the students' attention to this aspect. However, we need to find a way to address the attention of L2 students more precisely.
Finally, if we look closely at the typology of the most persistent L1 lexical deficits, only in one case do we note an erroneous lexical selection (VIII), while in the other cases these are choices related to inappropriate metaphorical language.

Syntax
A recent study by Lu and Ai (2015, 25) focuses attention on the different levels of syntax checking by the writers, concluding that learners with different L1s implies significant differences in syntax skills, even in cases where they have comparable levels of proficiency. Finally, Yang et al. (2015) consider syntax control a quality parameter of a well-formed L2 text. However, it is very difficult to define the set of related-syntax errors.
The first piece of RS advice (Appendix 1) is to simplify long sentences (above 35 words): the texts have also been evaluated according to this parameter (error =/> 45 words sentence). Only the L1 writers with RS did not write any sentence composed with more than 45 words. In L1 texts without RS we found 4 sentences composed of more than 45 words. It should be noted that very often the quantitative parameter is relevant since it could also be an indication of a qualitative text deficit. An example is the following (IX).
IX is a syntactically complex 57 words text: it has a hypotactic structure and we can observe an incidental clause. In contrast, the last part is juxtaposed without any connecting device but a comma. It is important to emphasize that it is, therefore, a not only quantitative, but also qualitative error. Both in the L1 and L2 texts we also considered as syntax errors some cases of erroneous expression of the determinateness (X), and some cases of confusion between direct and indirect complement (XI). Furthermore, we considered as a syntax error any wrong pronoun selection (XII).
There are also some very problematic aspects of textual cohesion in the field of both the syntax and the textual analysis. In the Italian language, sentence (frastic) and text (transfrastic) syntax are very close but not full matching concepts (De Beaugrande, Dressler 1994 2 ). The cohesion of a text occurs through a certain number of linguistic devices. The most problematic -for both Italian L1 and L2 writers -are the following: pro-forms, paraphrase, repetition and conjunctions.

Punctuation
We specify that we have considered only the logical-syntactic punctuation; that is the punctuation function that indicates the syntactical relationship between the parts of the speech and that disambiguates the text. Following this general rule, we have considered erroneous only the marks that have broken the main fixed rules of punctuation in Italian. These rules are: a. do not divide the text with punctuation marks between the verb and its subject and/or its direct object. The exception is a quantitatively and qualitatively very heavy subject (ex., a long phrase with a relative clause); b. put a full stop, a semicolon or a colon when the grammatical subject changes; c. open and close an incidental proposition, with commas, dashes or brackets; d. use the semicolon to divide heavy portions of text (ex. the points of a list); e. use the colon to introduce a list or a part of a text with the function of explanation.
The most frequent punctuation error in our corpus is the use of the comma. The comma is often used as a passe-partout syntactic mark to replace intermediate and strong pauses. In other cases, it is absent at the beginning or at the end of the incidental clauses or placed between the subject and verb.
In the L1 texts with RS, we don't find any of the first type (a) errors, and we find few of the second type (b). In L2 texts, intermediate pause marks (:;) are frequently replaced by the full stop and there is no significant difference between the texts produced with or without RS. Quotation marks ("") are always correctly used, but sometimes the author of the quotation is not correctly reported.
Regarding the spacing (whitespace), we have considered wrong spacing as an error only when it was a frequent feature in the text and therefore it was certainly not a typo. We have considered this case as a single error. We have considered all the apostrophe errors as spelling errors. Finally, in only one L1 text without RS can we find two cases of ellipsis (…) error ("un evento culturale bandito per sem-pre… la Love Parade. L'ultima").

4
Argumentative Text Structure Analysis

Text structure
Classical rhetoric, both in the Greek and Roman worlds (Barthes 2000 3 , 89 ff.; Mortara Garavelli 1999 3 , 59 ff.), described the argumentative text as an articulated structure: we provided a simplification of this structure to the students as an explicit RS tool (Appendix 1). We omitted the complex categorizations that distinguish different types of topics and the wide range of canonized τόποι ("topoi") within the discourse; in contrast, we focused on the most general level of the argumentative text development as follows [tab. 3]: Table 3 Argumentative text structure -introduction > topic and thesis presentation, in an introductory paragraph; -main part > ideas supporting the topic/thesis in logical order and examples, in some developmental paragraphs; -summary and conclusion > a short main summary of the ideas and one explicit sentence, in a concluding paragraph.
In this area, it is also necessary to consider the quantitative aspect of the text. We have given the students the task of writing a 500/700 words text, and negatively evaluated the text structure of less than 400 word writings. The extreme brevity was the main imperfection in the text structure, even if well structured. In the 58 corpus texts, we have recorded 23 errors of AE text structure, with a significant difference between L1 texts with and without RS. In L2 texts the difference is less significant because almost half of the writings did not reach 400 words, either with RS or without RS (in both cases 5/11). However, it should be noted that none of the texts with RS presents other text structure errors. In contrast, the lack of conclusion and/or explicit (3 cases in L2 samples and 3 cases in L1 samples) is present in texts without RS.
If we look at two examples of a thematic index with RS [tab. 4] and without RS [tab. 5], we can find a remarkable difference. (1) Introduzione. Il problema del lavoro nero in Italia (2) Il lavoro come indicatore dell'identità sociale (3) Conclusione. Lavoro nascosto, identità nascosta (1) Diffusione esplosiva della notizia online (2) il fatto in sé (3) intenzioni ed effetti di senso (4) Non ci sono fonti affidabili (5) Autorità della notizia data dalla condivisione It is important to emphasize that the two different indexes are the result of two different text structures. The one in table 4 is based on the table 3 model, and therefore it presents a linear argumentative structure. Vice versa, the one in table 5 corresponds to a free and less organized outline, written more like a collection of ideas than an argumentative structure.

Coherence
In our corpus we can find many errors in paragraph structure (9 errors in L1 texts and 16 errors in L2 texts). In many cases, the section beginning/end does not correctly correspond to the paragraph beginning/end. Let's take just one example (XXII): (XVIII) È noto che il discorso pubblicitario è caratterizzato dalla persuasione, ovvero dalle strategie messe in gioco per convincere il destinatario (quasi sempre un "consumatore") a fare qualcosa (nella maggior parte dei casi "consumare"). Questo ragionamento banale giustifica l'importanza dell'advertising nella logica capitalista in cui viviamo e il fatto che esso sia essenzialmente accusato di valorizzare e rinforzare il consumismo. Tuttavia, la pubblicità, quando è fatta bene, ha un altro tratto importante: la creatività. Sebbene sia un'esigenza di mercato, data la disperata necessità di richiamare in ogni modo l'attenzione, essa è chiaramente la ragione per la quale un annuncio ci può ancora sorprendere. (L2 text with RS) We can clearly see the use of a topic sentence at the beginning of the section. It has the function of guiding the reader in constructing meaning in the ensuing text and of indicating its content. But the text is actually formed by many sentences that correspond to many paragraphs instead of being included in a single paragraph. Information structure. We can analyse the information structure using the two pragmalinguistic categories topic/comment, given/new often used in pragmatic and communicative utterance analysis. The opposition topic/comment is not synonymous of given/new opposition. However, because of the ambiguity of these two oppositions, František Daneš (and then Nystrand 1986, 69 ff.), proposes classifying three patterns of topic development -also in this case explicitly mentioned in the RS: -linear progression: the new topic-information of a sentence refers to the given comment-information of the preceding sentence; -constant topic progression: the given topic-information of a sentence is the same as the given topic-information of the preceding sentence; -derived topic progression: the new topic-information of some sentences is different but derived from only one given overriding theme.
Here is one example of an error in composing a paragraph (XXIII) with a constant topic progression.
As we can see, in XXIII there are four sentences (a-d). The second one (b) is not consistent with the topic progression because it does not have the same topic as (a), (c) and (d) (i.e. "La legge 6.131"). Incidentally, the topic progression analysis was already proposed, among others, by Connor and Farmer (1990) as a RS useful for evaluating the L2 text structure.
The information structure also concerns both the logical relationship between paragraphs and the explanation of this relationship through a connective device: a conjunction ("dunque", "infatti" etc.) or a phrase with a connective function ("Alla luce di quanto detto finora" etc.). In our sample there are a total of 7 errors in L1 texts and 23 errors in L2 texts. The native Italian speakers with RS have made fewer mistakes. The difference between texts with and without RS, on the other hand, is not very significant for L2 texts.

Concluding Remarks
The foregoing discussion has attempted to suggest some guidelines in order to achieve a RS model for self-evaluation of the AE. By empirically examining the AE, we highlighted the writers (both in L1 and in L2) difficulty in controlling the language and the rhetorical/ informative structure of the texts. Moreover, predictably, we can see that the greatest number of errors has been produced by Italian L2 writers.
The differences between mother tongue acquisition and L2 acquisition are inherent to different areas: cognitive, affective and social. Many theoretical approaches emphasize the non-comparability of the two learning contexts, while others underline the importance of the linguistic universals as a common basis for the two learning situations. The rhetorical and textual context that we have considered, however, is the linguistic and argumentative revision strategy by adult learner; that is, students enrolled in a master's degree course in Italian language with very advanced linguistic skills. From our point of view, therefore, the drafting of an Italian L2 AE requires some specific L2 points of attention, but also some that can in contrast be considered in parallel with the L1 writing. In fact, these L2 writers already learned the grammatical and lexical structures, for such an advanced writing task, even if with the many exceptions mentioned above. This does not mean to equalize the linguistic skills of L1 and L2 writers, but only to emphasize some common issue within a specific rhetorical and linguistic-textual framework. This seems to also be confirmed by writing problems of the Italian writers' AE. As we have seen, they often have problems with all linguistic areas: spelling, morphology, vocabulary, syntax etc. As shown by contrastive rhetorical studies, there are some interferences between L1 and L2 at the informative and argumentative level as well. These transfers on the one hand are proof of the difference between the L1 and L2 writers, but on the other hand, they highlight the need to make explicit the revision process at all levels of text control for both L1 and L2 writers.
Regarding the results of this study, we must finally point out two limitations that, however, represent opportunities to study the topic of this research further. First, the qualitative difference between the texts before (α) and after (β) the revision should be better investigated. In this paper, however, we only considered the quantitative point of view. Secondly, we should consider the qualitative relationship between texts with and without RS in a more systematic way. However, we submit that the contribution of our paper rests, if only, on the importance of RS in the academic AE.