Nominal Agreement in L2 Speakers of Italian: Suggestions for a Teaching Plan

This paper addresses the topic of adult acquisition of nominal agreement in Italian, a crucial issue in teaching Italian as Second/Foreign Language. Building on a corpus containing spontaneous and semi-spontaneous production data from two advanced L2-speakers of Italian, I show that nominal agreement can be problematic even in the last stages of the acquisition process. The discussion of the instances of missing agreement in the corpus suggests that these are not due to a missing knowledge of the agreement rules in Italian, but instead on processing and production. In particular, some contexts prove to be more difficult than others: gender agreement (i.e., agreement with feminine nouns) is more difficult than number agreement, and the presence of two or more modifiers that refer to the same noun increases the error rate. Another difficulty is offered by the cases in which an item external to the determiner phrase (DP) has to agree with the subject of the clause. All these issues are tackled from a teacher‟s perspective: I highlight how the various contexts could be addressed and considered in the classroom, in order to help the learners to improve their production. Therefore, this paper argues that the best way to tackle this issue is by coupling knowledge of the formal rules of the language and of the acquisitional process with practical application in the context of second/foreign language teaching.

precisely, although the results of these studies are not completely overlapping, they all provide evidence of great difficulties, especially in the first stages of acquisition (Franceschina 2005).
In this paper, I illustrate the production of two advanced L2-speakers that have German as L1 and that have lived in Italy for about thirty years. Notwithstanding this long period and the rich and various inputs in Italian they have received, their production shows many examples of missing agreement. I discuss these cases from a didactic point of view, paying attention to what the most important sources of errors are, and how teachers could improve the production of correct gender agreement by their students (Note 1).
My research thus offers two new perspectives, which have been less developed in the literature to date. First, I do not focus on the first stages of acquisition, but on the final stage in these speakers" interlanguage. This is an important difference, because the previous studies on Italian focused on people that had recently started to learn Italian (e.g. Chini 1995); therefore, their results could be typical of a transitory stage, and the situation could change when the overall proficiency of the L2-learners improves. Second, this paper takes a teachers" perspective in the discussion of the "errors", focusing less on the psycholinguistic part of this investigation. In particular, I highlight the conditions that make the realization of agreement particularly difficult for the analyzed speakers, in order to help teachers to focus on them. One case in point is agreement with an item that is outside the Determiner Phrase ("DP"), the phrase that contains the noun and all adnominal modifiers (see also Alemá n Bañón et al. 2018).
The article is structured as follows: I first illustrate contrastively the gender agreement rules in Italian and German: German also has nominal agreement, but in a more restricted number of contexts than Italian ( § 2). Subsequently, I briefly resume the main observations made in the literature to date about gender agreement in Italian ( § 3.1) and in other languages ( § 3.2). Section 4 introduces my study and discusses the corpus on which the analysis is built, the biography of the two speakers and the videos from which the data are taken. Then, in Section 5 I discuss the data and show what the main sources of non-target-like production are. Finally, Section 6 is devoted to the conclusions.

Nominal Agreement Rules in Italian and German
I use the term "nominal agreement" (and, in short, "agreement") to refer to the morphosyntactic process whereby some items such as adjectives, quantifiers, and in some languages also participles, share the same gender, number and case (if the language has morphological case) with the nominal head (Note 2). Agreement should be kept separate from gender assignment, which refers to the individuation of the grammatical gender that is intrinsic of a noun in a given language. In this section I focus on agreement, and I briefly discuss gender assignment only to give a full picture of the situation that L2-learners have to deal with when they acquire gender agreement.

Nominal Agreement in Italian
Italian nouns have two genders, masculine and feminine. In most cases, gender assignment is predictable thanks to the nominal morphology. Most Italian nouns end with the vowels -o or -a; the former is usually associated with masculine, the second with feminine e.g. porto ("harbour.M"), vs. casa ("house.F"). Exceptions exist, but they are quite seldom, e.g. mano "hand.F", problema "problem.M". In addition, there are some less frequent endings that are often, but not always, predictable: nouns ending in -ù, for example, are feminine (virtù "virtue.F"), while nouns in -e are ambiguous. However, they often end with a suffix that carries a fixed gender feature, e.g. -trice is feminine (lavatrice "washing machine.F"), while -ore is masculine (direttore "director.M"), see Chini (1995).
All elements within the DP agree with the noun, regardless if they occur in prenominal or postnominal position (1). Adnominal elements also agree when they are used as nominal predicates (2). Finally, compound verb forms built with a past participle agree with the subject in the passive, and in compound tenses when the auxiliary verb is essere "be", but not when it is avere "have" (3) (Note 3): (

Gender Agreement in German
German has three genders: masculine, feminine and neuter. Gender assignment is less predictable than in Italian, although some suffixes carry a specific gender. The suffix -chen, for example, assigns neuter gender (Mädchen "girl.N"), while -in assigns feminine (Lehrerin "teacher.F"). In most cases, however, information about the gender of each lexical item has to be stored in the lexicon.
on the one hand, it is more complex because adnominal elements are inflected not only for number and gender, but also for case. On the other hand, only prenominal elements agree with the noun. Since the noun is generally the rightmost element in the DP, we can reformulate the rule saying that all elements in the DP agree with the noun. Elements external to the DP, such as predicative modifiers and verbal participles, always occur in a default, non-inflected form, cf. (4-6) with (1-3): Summing up the agreement patterns of the two languages, a first observation is that both languages do have agreement (unlike languages like English). However, its distribution is different, because in Italian adnominal and predicative items always agree, and so do verbal predicates when they occur with the auxiliary be or come. In German, on the other hand, only modifiers in the DP agree with the noun.

The Acquisition of Nominal Agreement in SLA
The literature to date has mainly focused on gender agreement, while number agreement has been paid much less attention to. The main studies on the acquisition of gender agreement in Italian as a L2 are discussed in Section 3.1, while Section 3.2 sums up the literature about gender agreement in other Romance languages.

The Acquisition of Gender Agreement in Italian as L2
Native speakers of Italian acquire gender agreement very early in their childhood (see e.g. Pizzuto & Caselli 1992, Chini 1995, Kupisch et al. 2002, and the same holds for bilingual children, although there might be a slight delay (Cantone 1999, Serratrice 2000, Kupisch et al. 2002. The study of the acquisition of gender agreement in L2 Italian has a longer tradition, and it mainly deals with learners that have a low or intermediate level of proficiency: this tradition started at least in 1990, when three studies on this topic were published: Berruto et al. (1990), which focuses on article selection by immigrants in the German part of Switzerland; Berretta (1990), which is devoted to the acquisition process of Italian morphology and therefore mainly investigates how nouns and adjectives are inflected. Finally, Valentini (1990) deals with the acquisition of both gender and number agreement by three Chinese immigrants that are in their first stages of acquisition.
Both Berretta and Valentini observe an overextension of -a in the first acquisition stages; Berretta also notes that her speakers produce target-like agreement as soon as they reach an intermediate stage. Moreover, Valentini shows that learners have less problems, even in the first stages of acquisition, when the noun triggering agreement has a [+human] feature: they thus seem to be sensitive to natural, more than to grammatical, gender (e.g., un maestro vecchio "a.M teacher.M old.M). This observation is confirmed by Berretta & Crotta (1991) in a longitudinal study of a learner of Italian with Cantonese as L1 and English as (advanced) L2.
A more comprehensive study of gender agreement is offered by Chini (1995). She studies longitudinally the production of eighteen adult acquirers of Italian that have different L1s: Farsi, English, German and French. Her data show that there is an implicational scale in the acquisition of gender agreement, going from the 3rd person pronoun to the past participle: (7) Anaphoric (pronoun) of 3rd person > definite article (>) indefinite article > attributive adjective > predicative adjective (>) past participle (Chini 1995: 285) As far as the L2-learners with L1 German are concerned, Chini shows that some of them first pass through a stage in which they tend to overextend -a. Then, they tend to use a default form which corresponds to the masculine singular. At the end of the observation period, there is a clear-cut difference between the four informants: two perform almost at ceiling with both adjectives and participles (in all functions). The other two perform at chance, with a percentage of target-like production around 50%. It should be noted that Chini"s longitudinal study follows the speakers in their first months of contact with Italian, stopping around the tenth month of the acquisition process.
A more recent study on the acquisition of gender agreement is Bianchi (2012). Bianchi analyses two populations, adult simultaneous bilinguals of Italian and German, and German L2 speakers of Italian with a high level of proficiency. In her study, she focuses on gender assignment and on agreement of the past participle with an object clitic pronoun. The two groups of informants performed in a very similar way, and this is taken by Bianchi to indicate that Age of Onset is not a crucial factor in the acquisition of agreement rules, at least when the L1 of the speaker also has some sort of agreement. What seems to play a more significant role in the target-like production of agreement is the amount of input: the speakers with more exposure to Italian performed better than the others. A further result is that the informants were more accurate in the acceptability judgement task than in the production task. According to Bianchi, this is in line with the Missing Surface Inflection Hypothesis ("MSIH"; Pré vost & White 2000), which claims that morphological deviances from the target should be attributed to limitations in performance and production, rather than to problems with the syntactic representation.

Studies on the Acquisition of Gender Agreement in Other Romance Languages
Studies on gender agreement in other Romance languages have mainly focused on speakers with English as L1; i.e., on learners that do not have any gender agreement rule in their L1 ISSN 1948-5425 2021 grammar. Therefore, one of the main issues discussed in the literature is whether there is a critical period for the acquisition of gender, i.e if adult L2-learners that do not have gender in their L1 can master this part of the L2 grammar. This debate opposes studies such as Carroll (1989) and Hawkins & Franceschina (2004), which support the hypothesis of a representational deficit, to White et al. (2004); the latter shows that gender can be acquired by L2-speakers in a native-like manner, in line with the MSIH.

International Journal of Linguistics
Another interesting study is Kupisch et al. (2013), which shows that French/German simultaneous bilinguals and advanced L2-speakers of French with German as L1 do not have any problem in French gender agreement. They thus also support the MSIH and, more specifically, they conclude that access to Universal Grammar (UG) does not deteriorate with age, at least when the L1 of the speakers also possesses a gender feature (for the Full Access Hypothesis, see White 1989, Schwartz & Sprouse 1996. Most studies on gender agreement in L2 Romance deal with learners of Spanish in the USA that have English as their L1. Montrul et al. (2008) compares heritage speakers of Spanish in the USA to L2-speakers, showing that both populations have some deficits with gender agreement; however, L2-speakers" deficits concern mostly the oral domain, unlike heritage speakers who have more problems with the written domain. Foote (2015) analyzes the production of intermediate and advanced learners, compared to native speakers of Spanish, testing the effect of gender attraction (which occurs when a complex DP contains two nouns that do not match in gender, e.g. La puerta del baño "the door.F of-the bathroom.M"). In her experiment, slightly more agreement errors occurred when the DP contained a feminine head noun; moreover, the opaqueness of the head noun"s gender correlated with a significantly higher amount of errors.
In the last decade, gender agreement in SLA has been the focus of neurolinguistic studies as well, which also target the language pair English (L1)-Spanish (L2). Alemá n Bañón et al. (2018) analyse the effects of structural distance between the head noun and the agreeing item. They detect that overall proficiency in the L2 is a strong predictor for the ability of the learner to individuate errors in gender agreement, while structural distance does not seem to play a major role. Grüter et al. (2012) investigate the relation between gender assignment and gender agreement. In their production experiment, participants make much more assignment than agreement errors: the proportion is more than 10:1 (114 vs. 10 errors in total). This result is interpreted as evidence for the fact that the difficulties of advanced L2 learners originate from the lexical properties of gender (which concern gender assignment), while the syntactic properties (responsible for agreement) cause minimal problems. Similar results are found by Holger Hopp in various studies: Hopp (2013) shows that the lexical aspects of gender are more problematic than the syntactic ones. Moreover, an eye-tracking experiment on gender agreement in L2 comprehension revealed that "variability in the use of gender agreement in real-time L2 comprehension reflects lower levels of activation of and access to gender nodes in the adult bilingual mental lexicon rather than representational deficits in L2 grammars" (Hopp 2013: 53). Moreover, Hopp (2016) shows that there is a causal relation between gender assignment and gender agreement: the ability of assigning the correct gender to a specific noun is a prerequisite for agreement processing. According to him, these facts can be International Journal of Linguistics ISSN 1948-5425 2021 explained through the Lexical Bottleneck Hypothesis ("LBH"; Hopp 2018): some non-nativelike results in syntactic processing, for instance in gender agreement, can be due to factors that are syntax-external, and in particular to lexical processing.

Presentation of the Study
In the previous section, I discussed the results of the investigations on L2 learners of Italian with a low proficiency level. In particular, we have seen Chini"s (1995) implicational scale in the early stages of the acquisition of agreement ( § 3.1); useful observations also come from White et al. (2004), Bianchi (2012) and Kupisch et al. (2012): learners with advanced proficiency may fail to produce agreement in some cases, but according to the MSIH this is due to processing and production errors ("mistakes", in Corder"s 1967 terminology) (Note 4). In Hopp (2013), problems with gender agreement are due to a lower activation of, and access to, gender nodes. Morever, Foote (2015) has shown that feminine nouns and nouns with opaque gender are more difficult to process. Finally, various neurolinguistic studies discussed in § 3.2 show that gender assignment is more difficult for L2-learners than gender agreement, and that the latter depends on the former (e.g., in the LBH).
In the present study, I chose to discuss gender agreement from a different angle: while the literature to date mainly focuses on the outcome of the acquisition process, here I would like to take a teacher"s perspective, discussing two concrete cases of advanced L2 speakers (with German as L1) that still show a high number of errors in gender agreement. In line with the didactic aim of this paper, the methodology chosen is a qualitative discussion of some typical errors with the aim to help teachers to address the topic of gender agreement in the classroom quantitative considerations, on the other hand, are beyond the scope of this paper. The data discussed in this study come from a corpus built of my own transcriptions of video interviews and video recipes that are available online; in these videos, the two analyzed speakers appear. Since both speakers have the same L1, it is possible that some error patterns may be due to negative transfer from the L1. However, I think that the most problematic areas, e.g. cases in which more than one modifier is present or in which agreement targets a DP-external modifier, should hold more generally, since they do not have any reflex in German; therefore, they should not be the result of a transfer from their L1.
Accordingly with the paper"s focus on the didactic level, the illustration and discussion of the errors found in the corpus is intended to suggest possible teaching interventions. On the one hand, teachers can learn what the most common error patterns and sources are, and where errors are more resistant even when the overall proficiency increases. Second, since all examples come from videos that are freely available on YouTube, teachers can use these videos to illustrate and discuss with their students the deviations from the target, in order to stimulate a metalinguistic reflection on nominal agreement.

The Speakers
As mentioned above, the corpus is based on a number of videos uploaded on YouTube, which show two persons which are well-known to the general public in Italy: the chef Ernst International Journal of Linguistics ISSN 1948-5425 2021 Knam and the former model Ela Weber. These two speakers have been chosen because they have the same L1, a similar biography and a comparable production as far as agreement is concerned.
The biographical notes available online tell us that Ernst Knam ("EK"; born 1963) is a chef and confectioner from Southern Germany. He moved to Italy in 1989, when he was 26 years old. He lives in Milan, is married to an Italian woman and has four children. He became famous in Italy because he took part to various TV-shows dedicated to bakery. Ela Weber ("EW"; born 1966) is three years younger than Knam. She is also Southern German, and she also had international experiences when she worked as a model; in 1988 she moved to Italy. She is married to an Italian man and has no children.
Although we do not have any specific information, both EK and EW allegedly acquired Italian spontaneously, in contact with native speakers: their professions require a strong commitment and a conspicuous time investment, while knowledge of Italian is not a fundamental skillsince they both work(ed) in international environments, they could probably use English to communicate at work.
Note that at the beginning of this investigation I also consulted videos of other L2-learners of Italian that have German as L1. However, in their production I could not find any error with respect to agreement, apart from isolated slips. This observation agrees with the observations of Chini (1995): when learners reach an intermediate-upper proficiency level, only some of them retain errors in agreement, while others produce only correct forms, cf. § 3.1.

The Videos
The present research is based on a number of videos containing spontaneous or semi-spontaneous speech. For EW, I considered an interview given in 2018 to a programme of the national broadcast company Rai1, lasting 14"55 (Note 5). In the interview, EW seems to answer spontaneously to the questions and starts a free dialogue with the interviewer (we may assume that she just had a general knowledge of the topics that would be touched upon in the interview). In the case of EK, on the other hand, I analysed seven shorter videos: three interviews and four video-recipes. The total duration of his videos is 53"15 (Note 6).
For each video, I transcribed all sentences in which there was at least one item requiring agreement. I did not consider the use of articles, following Bianchi (2012) and others that consider them an instance of gender assignment, rather than gender agreement. This interpretation is confirmed by the fact that the two speakers never have problems with the selection of the article, which always occurs in the expected form (with only three exceptions in total). All the other modifiers requiring agreement were tagged with respect to their word category and to the position they occupy in the clause: whether they are internal or external to the DP, and in the last case whether they are used as nominal predicates in a copular sentence or in compound verb forms.
Finally, some fixed expressions "noun + adjective" (e.g. panna montata "whipped cream") were not considered, because the speakers are likely to have them acquired as fixed forms. And indeed, they always occur target-like in the corpus. ISSN 1948-5425 2021

Unproblematic Contexts
As mentioned above, a first context that does not seem to constitute a problem for the analyzed speakers is gender assignment: the selection of the article or determiner is always target-like, even when there are other adnominal elements that do not show the expected agreement: In these examples the articles show the correct gender and number, while the following adjectives do not. There are only three errors of assignment, which are evidently production errors (slips).
A second part of the grammar where I did not find any error is the agreement with a masculine singular noun. There are only two exceptions, both by EK, for example: The striking difference between the performance at ceiling with masculine singular nouns, and the numerous errors with plural, and especially with feminine nouns, suggests that the masculine singular should be seen as a default form in the mental grammar of EW and EK. Indeed, the agreement errors usually consist of an extension of the masculine singular to a context in which another form, plural or feminine singular, is expected. This pattern thus diverges from what Berretta (1990), Berruto et al. (1990) and Valentini (1990) observed for speakers with low proficiency of Italian, which tend to overextend the -a ending (cf. § 3.1); on the other hand, it agrees with acquisitional studies on Spanish that consider the masculine as the default (cf. Carroll 1989, Foote 2015. The absence of errors in these two configurations gives us valuable information on how agreement should be addressed in the classroom. Considering first gender assignment, we can build on Bianchi"s (2012) observation that the speakers" performance depends on the amount of input received in the L2: in our case, since the two speakers have lived in Italy for more than 30 years and have an Italian-speaking family, we can assume that they have received frequent and various input in Italian. Correct gender assignment thus depends on explicit learning of the main rules, but also on constant use and practice of these forms in everyday International Journal of Linguistics ISSN 1948-5425 2021 life. Therefore, it should be discussed and exercised in class during the first stages of acquisition, but it affects less advanced learners that receive rich input from the context.
Concerning the role of the masculine singular inflection (-o) as default form, a consequence of this fact is that learners will correctly realize the expected form when the noun is masculine singular. The focus of the exercises and activities should therefore be on feminine, as well as on masculine plural, nouns. On the other hand, the fact that we do not find many errors in the students" production when the noun is in the masculine singular does not necessarily entail that they are agreeing the modifier correctly. It might also be the case that they are just using the default form, which happens to be homophonous with the masculine singular. Therefore, teachers should try to make sure that the correct production in the masculine singular is not an accident, but that the agreement system is mastered as a whole, with the masculine singular occupying one cell of the paradigm like the feminine singular and the plural forms.
In the next section, I focus on the contexts in which we do find a higher error rate: feminine singular, and masculine and feminine plural.

Agreement With Feminine and With Plural Nouns
Overall, modifiers occur with the expected ending in six out of ten cases. However, there is strong variation among the different gender/number combinations, ranging from 55% with feminine singular nouns (73/132 occurrences) to around 75% with masculine plural nouns (26/34 occurrences). I suggest that these percentages point to the fact that the speakers are to some extent aware that modifiers have to agree with the noun, and thus they have a mental representation of gender agreement; however, they are inconsistent due to production errors (cfr. the MSIH discussed in § 3). In the following subsections I discuss each gender/number combination separately; then, I illustrate two cases that are particularly problematic: when there is more than one modifier referring to the same noun, and when the modifier is DP-external.

Feminine Singular Nouns
The agreement with feminine singular nouns produces the highest error rate. Only around 55% of the modifiers is marked with the feminine morpheme, the remaining 45% are in the default form.  ISSN 1948-5425 2021 www.macrothink.org/ijl 66 In (11), the feminine noun torta "cake" is followed by the default form of the adjective famoso "famous", while we would expect the feminine form famosa here. A similar context is (12), from EW: the feminine noun cosa "thing" is followed by the superlative modifier bellissimo "very beautiful", instead of the correct bellissima.
In EW"s production we also find a very interesting and quite unexpected type of error: during the interview she produces a number of sentences in which she or her sister are the subject of the clause. However, although these subjects bear natural [+feminine] gender, EW often uses the default masculine singular form in compound verb forms, witness (13) EW is talking about her sister, who came to Italy while EW was ill. In the second part of the sentence the subject is io "I", which refers to EW herself. However, in both clauses the compound past tense is formed with the default participles rimasto "stayed" and rimesso "recovered", instead of the target-like rimasta and rimessa.
The data of the feminine singular forms show us two things: first, gender agreement errors are persistent, at least for some speakers; and they are more frequent than number agreement errors (cfr. below). Therefore, teachers should pay special attention to the realization of the feminine morpheme, with all types of modifier. Second, differently from what Valentini (1990) showed for speakers with low proficiency of Italian, in EW the presence of natural gender does not improve the production of the correct agreement: actually, it is at least as problematic as agreement with nouns that have grammatical gender. Therefore, teachers should not give for granted that L2-learners will have less problems when dealing with [+animate] or [+human] feminine nouns: in fact they should exercise these forms to the same extent as nouns with grammatical gender.

Plural Nouns
I discuss feminine and masculine plural agreement in the same section because they are much less frequent in the corpus. The expected forms occur in 75% of the cases when the noun is masculine plural, and in 70% of the cases when it is feminine plural. However, in this case there is some difference between the two speakers: EW produces just one error, in a complex construction in which there is ne-extraction. Most notably, in this occurrence she does not use the default form, but the feminine plural instead of the masculine plural: this might be due to the complexity of the agreement pattern:  ISSN 1948-5425 2021 "Friends, acquaintances, I had many of them." EK, on the other hand, performs better with the masculine plural (76%) than in the feminine plural (50%). An example of non-expected production is (15) "and the others arrived, one after the other" In this example the postverbal subject is in the masculine plural, but the past participle is in the default form.

Some Particularly Problematic Contexts
In this section, I discuss two contexts that have proven to be particularly difficult for both EW and EK. The first context is "multiple" agreement: the co-occurrence of various modifiers that refer to the same noun and must all agree with it according to the Italian grammar. The second context concerns agreement with DP-external items.

"Multiple" Agreement With the Same Noun
In our data, it is quite frequent that more modifiers refer to the same noun, especially in EK"s videos. The corpus contains various combinations: a prenominal and a postnominal modifier co-occur within the same DP (16), or two postnominal modifiers do (17); a third possibility is that two predicative modifiers are coordinated (18). Curiously enough, most of the times the speakers realize agreement on one of the modifiers, while the other(s) occurs in the default form: In (16), the prenominal possessive adjective nostri "our" is correctly inflected for masculine plural, but the participial adjective montato "whipped" is in the default form. On the other hand, the example (17) contains two postnominal adjectives: the first one (leggera "light") is in the feminine form, the second (estivo "summery") is not. Finally, in the last example (18) two participial adjectives are used as nominal predicates of a feminine singular noun (base International Journal of Linguistics ISSN 1948-5425 2021 "basis"): here the first term of the coordination is in the default form, while the second has feminine agreement.
These examples show a more general pattern, indicating that L2-learners find multiple agreement particularly difficult. I suggest that this shows that the cognitive load increases dramatically whenever agreement has to be established with more than one modifier, especially when the different modifiers are not adjacent (as expected by the MSIH). Note that in these cases there is not a clear pattern telling us which of the modifiers will occur with the expected inflection, and which not: it rather looks like a random choice. Therefore, multiple agreement is a topic that should be discussed in the classroom, possibly with the help of videos such as those that are analyzed here. In addition, the different combinations of two (or more) modifiers should be targeted by specific exercises that may help the learners to improve their production in this particularly problematic part of the Italian agreement rules.

Agreement of DP-External Items
A second configuration in which the error rate is particularly high concerns DP-external modifiers. Two cases are possible: in the first, the modifier is used as nominal predicate of a noun; in the second, a past participle is used in a compound verb form. In both cases, the agreeing item is part of the predicate and not of the determiner phrase. The percentages of expected answers are particularly low here: around 45% of the predicative modifiers is correctly inflected. In the case of participles in compound verb forms, the percentage decreases to less than 30%. Therefore, these modifiers occur in the incorrect default form in the majority of cases (55% and around 70%, respectively). This sharply contrasts with DP-internal modifiers, where the error rate is below 25%. Italian requires agreement of the past participle with the subject here, and therefore we would expect the feminine form pesata "weighed". In (22), on the other hand, EW uses a compound past form ("passato prossimo"): since the auxiliary is essere "be", agreement with the feminine subject (vacanza "holiday") is required. However, the participle occurs in the default masculine singular form (durato "lasted" instead of durata).
Agreement with elements that are part of the predicate therefore proves particularly difficult for L2-speakers, even for some of those that have a high overall proficiency and have lived in Italy for a long time. This may be due to two reasons: first, there is more distance between the noun and the modifier here, the modifier being outside the DP. However, this is unexpected if compared to the results of Alemá n Bañón et al. (2018). A second reason for the high error rate is that agreement with the subject is not required in all compound verb forms: participles agree with the subject only when the auxiliary is essere "be" (with passives and compound tenses) or venire "come" (with passives only), but not when it is avere "have", which is used as auxiliary in compound tenses when the verb is transitive or unergative.
Therefore, it is extremely important that teachers of Italian as L2 let the students practice these contexts with specific, dedicated exercises. Students should interiorize that agreement does not apply only within the DP, but it also targets external modifiershowever, not in all contexts. This part of the grammar should thus be treated with particular care, helping students to gain confidence with the specific configurations that trigger agreement.

Conclusions
In this paper, I have discussed a topic of the Italian grammar that offers great difficulties even to some advanced learners of Italian as L2. I have focused on two speakers that have German as L1 and an advanced overall proficiency in Italian. They have also benefited of a huge amount of input in their L2, since they have both lived in Italy for around 30 years and have an Italian-speaking family.
Notwithstanding this favourable context, they still have problems with nominal agreement, especially with the feminine singular nouns. These problems also arise when grammatical gender matches natural gender, i.e. when the referent is [+human]. In addition, I have highlighted two configurations that are the most problematic ones: first, when multiple agreement is required (i.e., when there is more than one modifier in the DP or in the clause), speakers tend to inflect only one of them. Second, when the modifier is DP-external, i.e. ISSN 1948-5425 2021 when it is used as nominal predicate or as part of a compound verbal form, speakers have a particularly high error rate, far above 50%.

International Journal of Linguistics
The literature on Italian has mainly focussed on agreement errors in speakers with a low proficiency (e.g. Berretta 1990, Berruto 1990, Valentini 1990, Chini 1995. On the other hand, literature on Spanish as L2 shows that gender agreement should be much less problematic than gender assignment (Grüter et al. 2012), or it should follow from it (Hopp 2018). Furthermore, structural distance has been shown not to correlate with a higher error rate (Alemá n Bañón et al. 2018). However, the results of my corpus study show that even speakers that are under the best conditionsdecades-long, rich and diversified input in the L2 and no problems at all with gender assignmentcan have great difficulties to master gender agreement. In particular, there are some complex contexts (e.g. cases of structural distance) that trigger high number of errors.
Albeit this study is not intended as a psycholinguistic study on the acquisition of gender agreement, it allows me to show an important fact that risks to be neglected in quantitative, experimental studies: how various a classroom can be. Since teachers are dealing with concrete cases, constituted by different individuals that have various linguistic and L2-acquisitional backgrounds, a qualitative discussion of single case can be exemplary of some commonalthough minoritarianpatterns found in real life. Therefore, teachers of Italian as L2 should not give for granted that intermediate and advanced speakers have the ability to produce agreement in all contexts; conversely, agreement should be focused on even in later stages of development, through specific exercises and dedicated discussions that target those contexts that are particularly problematic for L2-acquirers of Italian. In particular, videos like those discussed here can be useful to rise metalinguistic awareness in the learners. Table 1 resumes the suggestions for a teaching plan dedicated to acquiring nominal agreement. Table 1. Recap of exercises suggested on the basis of the corpus data

Syntactic context
Proposed teaching activity Gender assignment Focus on the rules that permit to predict the gender of a noun at the beginning, then improve the mental lexicon of the speakers with rich input. Gender agreement with masculine singular Try to check whether students are applying agreement or are just using the default form; choose activities that allow to consider the masculine singular within the broader picture of the different gender/number configurations. Gender agreement with feminine singular and with plural nouns Select diversified activities, especially those that allow to exercise the spoken language, and propose them frequently.
Gender agreement with two or more modifiers Think of all possible combinations and use all of them to make learners more aware of its pervasive presence. Gender agreement with DP-external modifiers Use exercises that involve agreement within full sentences (e.g., transform a verb in the present tense in a compound past, or an active in a passive form).
Future research should target the acquisitional parthow can we explain these patterns of missing agreement from an acquisitional point of view?and also focus on speakers with different L1s, in order to establish whether the difficulties illustrated here are typical for learners with German as L1, and whether learners with a different linguistic background have difficulties in contexts that are different from what I found in this study.