Abstract
Recent studies in the second language acquisition of negation have focused on polarity items and their licensing contexts. Although several studies show a correlation between higher degrees of second language (L2) proficiency and the acquisition of the target L2 structures, less attention has been given to the relation between the acquisition of polarity sensitivity and different types of lexical polarity items. This study addresses this gap in the literature by measuring the processing of polarity-sensitive items under the scope of negation in L2 Spanish. Our findings indicate that learners’ acquisition of polarity sensitivity differs across different types of lexical polarity items.
I Introduction
Research on negation in second language acquisition (SLA) has traditionally focused on the negation of the main verb of the phrase (see Dimroth, 2010). Recently, though, some studies in SLA have investigated more complex polarity relationships that involve the co-occurrence of negation and other lexical items within the phrase (Gil et al., 2019; Marsden et al., 2018; Mitsugi, 2022; Puig-Mayenco and Rothman, 2020; Yamada, 2022). The present work advances this research agenda by examining the extent to which different polarity items (PIs) affect the processing of polarity-sensitive phrase structures among second language (L2) learners of Spanish. Specifically, we focus on polarity sensitivity in terms of the grammatical licensing of negative concord items (NCIs) and the anti-licensing of positive polarity items (PPIs) within the scope of clausal negation in Spanish. Using a self-paced reading instrument and a vocabulary size measure, we argue that learners’ processing of these phrase structures varies according to the individual type of PI. The unique contribution of this work, therefore, is an empirical testimony to the significant role of individual lexical items in the second language learning of syntax–semantic constructions.
II Polarity sensitivity and PIs in Spanish and English
A key characteristic of Spanish syntax is negative concord (NC), a feature whereby two negative elements in the phrase only contribute one semantic negation. Spanish is a Non-Strict NC language (see Giannakidou, 1997, 1998, 2000), which is to say that lexical items such as nadie ‘nobody’ can appear postverbally within the scope of negation but can also express negation in the preverbal subject position without an additional preverbal negative marker. For example, in (1a) NC is realized between postverbal nadie and a preverbal negative marker no, which express one total negation in the phrase. The phrase in (1b) is ill formed because the postverbal nadie does not co-occur with any form of preverbal negation. In (1c) and (1d), we observe how Spanish NC is ‘non-strict’: nadie can appear preverbally, where it expresses negation without concordance with another preverbal negative marker.
(1) a. No vino nadie. ‘Nobody came.’ b. * Vino nadie. c. Nadie vino. ‘Nobody came.’ d. * Nadie no vino.
1
Due to the unique distribution of lexical items like Spanish nadie, Giannakidou (2020) calls these kinds of words ‘negative concord items’, whose definition is copied in (2).
(2) An expression α is an NCI (also known as ‘n-word’) iff [if and only if]: (i) α can be used in structures containing sentential negation or another α-expression, yielding a reading equivalent to one logical negation; and (ii) α can provide a negative fragment answer (i.e. without overt negation).
2
(Giannakidou, 2020: 459)
For several decades the literature has debated whether these words are inherently negative. While data such as (1a) seem to indicate that NCIs are not inherently negative, data such as (1c) seem to show that NCIs can express negation all on their own without other negative markers. Therefore, on the one hand, those who argue that NCIs are inherently negative have had to explain how they can appear in NC structures such as (1a). Haegeman and Zanuttini (1991), for instance, propose a process of negative absorption at LF which would cancel the negative meaning of such items under the scope of a negative operator. On the other hand, those who argue that NCIs are inherently non-negative (Bosque, 1980; Laka, 1990) must explain how they can appear preverbally to provide semantic negation as in (1c). Laka (1990), in particular, argues that these words are licensed in the scope of negation, and that in the preverbal position a phonologically null negative element licenses the appearance of the preverbal NCI.
Approaches within the Minimalist Program (Chomsky, 1995) have also provided accounts for these data (Tubau, 2008; Zeijlstra, 2004, 2008). Zeijlstra, in particular, argues that NCIs are non-negative indefinites that carry a feature [u
(3) a. No[ineg] vino nadie[uneg] b. Op¬[ineg] Nadie[uneg] vino
Thus, by Zeijlstra’s analysis, the role of postverbal nadie in a NC configuration in (3a) is accounted for, and in (3b) it maintains its status as inherently non-negative without additional negative markers.
Negation in English differs from Spanish in that negative phrases such as ‘nothing’ or ‘no one’ can appear postverbally without concordance with preverbal negation, as in (4a), where English negative phrases such as ‘no one’ seem to express negation alone. NC is indeed produced, though, in some varieties of English (see Labov, 1972), such that the phrase in (4b) expresses one instance of negation.
(4) a. I see no one. b. I didn’t see no one.
English also features negative polarity items (NPIs) such as the ‘any’-words, which are licensed in non-veridical environments including the scope of negation, as in (5a), and interrogatives, as in (5b) (see Giannakidou, 1998, 2011; Ladusaw, 1979). 3 Unlike NCIs, NPIs cannot express negation alone, as in (5c), where ‘anyone’ cannot be interpreted negatively. 4
(5) a. I didn’t see anyone. b. Did you see anyone? c. * Anyone came.
Finally, both English and Spanish feature positive polarity items (PPIs), or expressions which cannot receive a narrow-scope reading within the scope of clausal negation (following Nicolae, 2012a, 2012b). For English, PPIs include the ‘some’-words, and for Spanish, PPIs include the alg- series of lexical items, as shown in (6).
(6) a. I didn’t say something. * b. No dije algo. *
To account for the data in (6), Nicolae (2012a, 2012b) proposes that PPIs have contextually relevant alternatives that must be exhaustified, or factored into the utterance’s meaning, following Chierchia’s (2006, 2013) analysis of NPIs. They argue that for PPIs the exhaustification of alternatives takes place via an operator E which checks a feature [+DE] and requires an assertion to entail its alternatives. By this approach, the affirmative phrase in (7a) gives an assertion that entails its alternatives (John saw a member of sub-domain D entails that John saw a member of super-domain D′). The problem with negative phrases such as (7b) is that the alternatives are negated and are no longer entailed by the assertion.
(7) a. E[DE] John saw someone[+DE] b. E[DE] John didn’t see someone[+DE] *
In (7b), it is the alternatives (John didn’t see a member of super-domain D′) that now entail the assertion (John didn’t see a member of sub-domain D). This semantic violation of the E operator accounts for the anti-licensing of PPIs under the scope of negation.
At this juncture we may take account of the terminology to be used in this article. In Table 1 we outline the primary PIs to be discussed: NCIs, NPIs, and PPIs. Both Spanish and English feature PIs which are licensed in the scope of negation, and PPIs which are anti-licensed in that position. A key difference is that Spanish NCIs can express negation on their own, whereas English NPIs cannot.
Terminology followed in the present article.
Table 1 also divides these PIs into two syntactic types relevant to the present study, which we will call the thing-type and the ref-type. We will call Spanish algo ‘something’ and nada ‘nothing’ the thing-type PIs, while alguno ‘some’ and ninguno ‘none’ will be called the ref-type PIs due to their ability to make reference to a noun mentioned in the discourse.
III Polarity sensitivity in SLA
Several recent studies have begun to explore the L2 acquisition of polarity-sensitive items and syntax. In L2 Korean, Song and Schwartz (2009) analysed the acquisition of obligatory scrambling of Korean NPI amwuto in negative questions, where learners showed increases in target-like production and acceptance of these structures with increasing L2 proficiency. Marsden et al. (2018) and Gil et al. (2019) measured L2 learners’ knowledge of English NPI licensing contexts in relation to what is taught in typical classroom instruction, with a focus on any-type NPIs, and found that the NPI licensing contexts best acquired by learners were those that were taught in classroom instruction. Marsden et al. (2018) also found that learners’ knowledge of NPI licensing contexts increased with greater English proficiency. Both studies argued that the array of English NPI licensing contexts can be acquired in the L2, despite their varying degrees of coverage in pedagogical sources. Puig-Mayenco and Marsden (2018) also examined English NPI ‘any’ in L3 English acquisition, using an acceptability judgment task. They found evidence of transfer to L3 English from first language (L1) Catalan that caused an acceptance of NPI ‘anything’ in infelicitous positions (i.e. ‘*Anything was not written down’). Since similar phrase structures are also disallowed in these speakers’ L2 Spanish, but allowed in their native Catalan, this study defended a case of L1 to L3 transfer in the domain of NPI licensing.
Mitsugi’s (2022) recent study is one of the few to discuss polarity sensitivity in L2 sentence processing. They used a visual-world eye-tracking experiment to measure L2 Japanese learners’ processing of sentences with NPI adverbs, and found that the processing of negated main verbs was influenced by the appearance of NPI adverbs, which in Japanese can appear preverbally, as in their example test item copied in (8).
(8) zenzen heya-o soji-shi-masen-deshi-ta at.all-NPI room- ‘I did not clean the room at all.’ (Mitsugi, 2022: 7)
In particular, upon encountering preverbal NPI adverbs in the language stimulus, L2 learners predicted the appearance of sentence-final negated verbs better than when NPI adverbs were absent. Although L2 proficiency (as measured by self-reports) was not a significant modulator of their performance in the experiment, this study showed that learners are capable of acquiring the associated syntax and semantics of polarity-sensitive items in the L2.
While the above studies have widened our knowledge regarding the learning of NPIs, very few studies have discussed NCIs or NC in SLA. Austin et al. (2013) measured negation usage among kindergarten-age speakers of Spanish as a heritage language, who at times produced non-target-like uses of NCIs in an oral production task. The authors argue that these speakers’ knowledge of English may have influenced the production of phrases such as those in (9), which lacks grammatical NC.
(9) * Le gusta ningún pastel. ‘He likes no cake.’
These particular non-target-like uses of negative phrase structures, where NCIs are placed postverbally without preverbal licensing, were the focus of Yamada (2022), who measured NC among English-native L2 learners of Spanish in an oral elicitation task and found that target-like NC is produced more frequently as learners gain Spanish proficiency. Furthermore, the target-like usage of NC did not appear to vary by the NCIs tested, which were nada ‘nothing’ and nadie ‘nobody’. In other words, learners did not produce target-like NC better with one NCI or the other. The author did find, however, a certain learner preference for nada as a kind of blanket NCI answer to wh-questions asking both ‘what’ and ‘who’ at lower levels of L2 proficiency. That is, lower-proficiency L2 learners seem to overuse nada in contexts where it would be most appropriate to use nadie. Puig-Mayenco and Rothman’s (2020) study also stands out in that it is one of the few to account for different types of PIs, as it used a sentence–picture matching task to analyse L3 English learners’ interpretation of person-type PIs (‘nobody/anybody’) in subject position, and thing-type PIs (‘nothing/anything’) as objects in conditional phrases, in order to study the nature of transfer in language acquisition. While their study addresses the acquisition of different PI types, its objective was not to compare one PI type against another in the same phrasal context. Outside the domain of SLA, one study of interest that discussed the role of NC and individual PIs is that of Miller (2012), who investigated the use of NC in English first language acquisition. Miller used the Brown child-language corpus (Brown, 1973) to analyse the spontaneous speech of Sarah, who began to produce NC in English at 3;6 years of age. They noted that Sarah preferred to use NC structures with ‘nothing’ over NPI structures with ‘anything’ (‘I don’t get nothing’ vs. ‘I don’t get anything’) but preferred NPI structures with ‘any’ over NC structures with ‘none’ (‘I don’t have any’ vs. ‘I don’t have none’). In this sense, this study provides a glimpse at one particular language learner whose usage of polarity syntax was influenced by individual PIs.
To sum up, the studies reviewed in this section have investigated the L2 acquisition of polarity-sensitive items with a primary focus on their licensing contexts. A topic less explored by these studies, however, is the unique contribution of individual lexical PIs themselves in the acquisition of polarity syntax. That is, although Song and Schwartz (2009) analysed the syntax of negated questions with the NPI amwuto, other NPIs were not included in their test. Similarly, while the acceptability judgment instrument of Marsden et al. (2018) and Gil et al. (2019) included the NPIs ‘anyone’, ‘anything’, and determiner ‘any’, it was not determined if NPI licensing contexts were better ascertained with certain NPIs. While Puig-Mayenco and Rothman (2020) studied two PI types, the goal of their work was not to compare one against the other. Yamada (2022), more recently, tested different NCIs in the target-like L2 production of NC and found no difference between the two. However, Yamada’s study was the only one to specifically consider between-NCI differences thus far and, as such, further research on this topic is warranted. In summary, while researchers working in L2 negation have begun to focus more closely on constructions of greater relative complexity, an open discussion that merits further study is the degree to which the L2 acquisition of polarity sensitivity varies according to individual PIs.
IV Research question
With the literature reviewed above as our point of departure, we seek to answer the following research question:
Research question: Do L2 vocabulary size and individual PIs modulate the processing of sentences with NCIs and PPIs under the scope of clausal negation?
As discussed above, previous research has shown how L2 proficiency modulates learners’ knowledge regarding grammatical and ungrammatical uses of PIs. However, less is known about the role of individual PIs. Although Yamada (2022), among otherwise scarce studies, showed that NCIs nada ‘nothing’ and nadie ‘nobody’ did not modulate differences in the production of target-like NC in L2 Spanish, no other PIs were compared, leaving the topic still open to discussion. Thus, the present study investigates how L2 learners at different degrees of L2 vocabulary size process phrase structures featuring different classes of PIs, in the context of the scope of clausal negation.
V Methodology
The present research experiment consisted of a self-paced reading task, the 3K-LEx vocabulary size test (Robles-García, 2020), and a biodata questionnaire which included the bilingual language profile (BLP) (Birdsong et al., 2012), a self-report instrument designed to measure language dominance.
The self-paced reading task is an experimental paradigm that measures L2 sentence processing (see Jegerski, 2014, 2015, 2018a, 2018b; Keating and Jegerski, 2015; Lee et al., 2022; Roberts and Liszka, 2021). Participants completed the self-paced reading task using the PsychoPy software (Peirce et al., 2019) running on a 13" MacBook Pro laptop computer, and were given 5 practice items at the start of the task and a 5 minute break after completing half of the sentences. We employed a moving-window design with an uninformative mask, which is to say that participants read each sentence one word at a time, and each word in the sentence was masked as a continuous series of underscores. Response times (RTs) were collected upon pressing a button to advance from one word to the next.
In designing the task, we created 32 target items to populate a total of 4 counterbalanced presentation lists, and we pseudorandomized each list with filler sentences such that target sentences did not appear one after another. The 32 target items manipulated two variables: PI type and grammaticality. We manipulated the class of PI by testing the ref-type and the thing-type PIs. We then manipulated grammaticality by positioning these PIs in the scope of clausal negation, where the NCIs are grammatical (ninguno ‘none’ and nada ‘nothing’) and the PPIs are ungrammatical (alguno ‘some’ and algo ‘something’), as discussed in Section II. These two variables created a total of four versions of each of the 32 target items. We exemplify the schema for one of the target items in Table 2.
Target sentence schema.
Table 2 shows one total target item, divided into 4 possible sentences, each of which would appear in a different counterbalanced list. These counterbalanced lists were cycled for every participant: Participant 1 received list A, Participant 2 received list B, etc. Therefore, a given participant only read one out of the 4 versions of the target item shown in Table 2 (for a more detailed explanation of this approach, see Keating and Jegerski, 2015: 5–10).
The ref-type and the thing-type were chosen because they differ significantly in frequency and function, and thus serve well the overall objective of the study, which is to ascertain whether or not differences in PI modulate the processing of the associated phrase structures. To confirm their differences in frequency, we effectuated a simple search of these terms in the Web/Dialects corpus of Davies’ Corpus del español (Davies, 2016), reported in Table 3.
Polarity item (PI) frequency in Davies’ Corpus del español.
Table 3 shows the PIs in question, their total frequency counts, and the total frequency of strings matching any verb followed by the PI. These specific strings were searched to have a general idea of how often these PIs appeared in postverbal position. It is clear that the thing-type PIs largely outweigh the ref-type PIs in terms of usage frequency. Furthermore, as pointed out above, they differ notably in function as well, in that the thing-type PIs do not require the presence of some referent in the discourse in the same way that the ref-type PIs do. For example, as in the test item shown in Table 2, the ref-type PIs alguno and ninguno refer to the noun documentos ‘documents’ previously mentioned in the phrase. The thing-type PIs algo and nada, however, do not require reference to a particular noun in the discourse. Given that Yamada (2022) chose to measure differences between nada and nadie, our decision to measure the ref-type and the thing-type in the present study offers the advantage of comparing two PI types that are essentially more different than nada and nadie in function and in orthographic and phonetic makeup.
In creating the 32 target items we varied the proper names, the verb, and the object noun. As shown in Table 2, each target item contained two proper names, the first one containing 5 characters and the second one containing 6 characters. The verbs appeared in the present tense third-person singular and were among the 5,000 most frequent words of a corpus of 20 million words, found in A frequency dictionary of Spanish (Davies and Davies, 2018). The nouns used as direct objects appeared in the plural and were selected from cognates or vocabulary found in the first-year Spanish language textbook used at the university where data were collected. All sentences contained 13 words and ended with the words en este momento, to allow for an ample spillover region following the critical PI.
Each sentence in the self-paced reading task was followed by a poststimulus distractor task that asked participants to choose the image which best represented the sentence they had just read. An example of the poststimulus distractor task for the target item presented above in Table 2 is shown in Figure 1, where the correct answer would be A.
Example of poststimulus distractor task.
For the 32 poststimulus distractor tasks following target items, half of the answers were A, and half of the answers were B. In Figure 2 we summarize the overall flow of the self-paced reading task. As shown in Figure 2, in the course of one experimental trial, participants read each word of the sentence one at a time, and then completed the poststimulus decision immediately after. In total, participants read 128 sentences during this task. Thirty-two of these sentences were target sentences and 96 were filler sentences, which were unique to this study and were not being used for another experiment. Of the 96 filler sentences, 16 included the sentential negator no. This was done deliberately so that the target items would not be the only items to show negation. Filler sentences were grammatical or ungrammatical, so that the target items would not be the only items to show ungrammaticalities. None of the filler sentences included PIs. A list of all experimental sentences can be found on the IRIS database (www.iris-database.org).
Self-paced reading task flow summary.
Upon finishing the self-paced reading task, participants completed the biodata questionnaire and the 3K-LEx, a receptive yes/no vocabulary size test built to measure the knowledge of the most frequent 3,000 words in Spanish. In this test, participants see a word and must answer ‘Yes’ if they know the meaning of the word and ‘No’ if they do not know the meaning of the word. The test contains a total of 162 words, 108 of which are real Spanish words and 54 of which are pseudowords. Scoring favors ‘Yes’ decisions of real words but penalizes ‘Yes’ decisions of pseudowords: a participant can obtain a maximum score of 3,000 by deciding ‘Yes’ to all real words and ‘No’ to all pseudowords, and a minimum possible score of –3,000 by deciding ‘No’ to all real words and ‘Yes’ to all pseudowords. In the present experiment, participants completed the ‘Test 2’ version of 3K-LEx in Qualtrics, a copy of which can be found on the 3K-LEx Github page (https://github.com/problg00/3K-LEx). The 162 items were randomized and participants only saw one word at a time. ‘Yes’ or ‘No’ decisions were made by clicking on-screen buttons labeled ‘Yes’ and ‘No’. We chose a vocabulary size measure because the present study tests the acquisition of a grammatical feature in interaction with lexical items. Furthermore, in piloting we found that traditional multiple-choice exams were rather time intensive. The 3K-LEx has the advantage of being relatively quick to complete and, additionally, L2 vocabulary size has been shown to correlate with L2 oral lexical proficiency (Uchihara and Clenton, 2020), L2 vowel perception (Bundgaard-Nielsen et al., 2011), and L2 language dominance (Bonvin et al., 2021).
A total of 69 participants enrolled in Spanish courses ranging from second-semester Spanish language to upper-level content courses at a liberal arts college in the United States completed the present study. According to the biodata questionnaire, a total of 6 participants reported a language other than English as their native language or language spoken at home. In order to specifically analyse the acquisition of an L2 Spanish grammatical feature by L1 English speakers, these participants were subsequently removed from the dataset. For the remaining 63 participants, the scores on the poststimulus distractor test ranged from 46.9% to 100% (M = 93.3%, SD = 8.3%). We removed 14 participants who scored less than 90% (less than 29 out of 32) on the poststimulus distractor task, leaving a total final dataset of 49 participants (with collective poststimulus task scores of M = 96.5%, SD = 3.5%). Of these 49 participants, 34 reported being female. Their ages ranged from 18 to 22 years (M = 19.6, SD = 1.1). Forty-six participants reported having studied Spanish for at least 1 year in high school (M = 3.5 years of study, SD = 0.7), and 6 participants reported having studied abroad in a Spanish-speaking country for at least one month. Their scores on the 3K-LEx ranged from 222.24 to 2,583.54 (M = 1299.99, SD = 606.49).
VI Results
We shall first present a descriptive overview of our results. For the RT data collected on each word of the target sentences, we removed raw RTs less than 100 ms. We then calculated residual reading times (RRTs) by taking a regression equation for each word position in the sentence, for all trials of the experiment, and calculating expected reading times from the number of characters of the given word. These expected reading times were subtracted from the participants’ actual reading times, giving the RRT for each word position. RRTs that fell outside of 3 standard deviations from the mean were excluded from the dataset. Figure 3 gives a global view of mean RRTs for the last 7 words of the target sentences.
Global view of residual reading times (RRTs) by grammaticality and polarity item (PI).
In order to analyse the effect of the experimental manipulations, we will take a closer look at the RRTs produced at the critical PI region and the spillover region following the PI. Thus, as shown in the example target phrase in (10), we will analyse RRTs on the word in the PI position, the en word, and the este word.
(10) Carla tiene los dos documentos pero Andrea no tiene nada1 en2 este3 momento.
To this end, we performed three separate mixed-effects linear regressions using the lmer function of the lme4 package in R (Bates et al., 2015). The first regression analysed RRTs produced for the PI, the second regression analysed RRTs for the word en directly after the PI, and the third regression analysed RRTs for the word este. With the removal of outliers from each dataset, the ‘PI’ regression analysis contained a total of 1,540 observations (after 1.8% data loss from the aforementioned outlier trimming), the en regression analysis contained a total of 1,547 observations (after a 1.3% loss), and the este regression analysis contained a total of 1,549 observations (after a 1.2% loss). For each regression, the dependent variable was the RRT in milliseconds. The independent variables were the grammaticality of the sentence (coded as ‘Yes’ for grammatical or ‘No’ for ungrammatical), the PI type used (coded as ref for the ref-type pair or thing for the thing-type pair), the continuous zero-centered 3K-LEx score, and interactions between each of these fixed effects. Random effects were participant and item (of the 32 target items). Default treatment contrasts were applied to the factor groups of grammaticality and PI, with the baseline levels being ‘No’ for grammaticality and ‘thing’ for PI, respectively. This model is outlined in R syntax as follows: lmer(RRT~ grammaticality*3KLEx*PI + (1 | participant) + (1 | item))
The lmerTest package was used to estimate p-values (Kuznetsova et al., 2016). The results of the three regression analyses are shown in Table 4.
Results of regression analyses.
p < .05
As shown in Table 4, at the PI region we find a significant two-way interaction between the fixed effects of grammaticality and PI type, and a significant three-way interaction between grammaticality, PI type, and 3K-LEx score. Recall that this is the region of the sentence in which the ungrammaticality would first appear. Later, in the spillover region, we find a significant negative modulator of grammaticality and a significant positive modulator of 3K-LEx score at the word este. These effects are visualized in Figure 4.
Regression analysis of residual reading times (RRTs) on grammaticality, 3K-LEx score, and polarity item (PI).
Figure 4 shows the linear regression for the PI region on the left, and for the este region on the right. RRTs are given on the y-axis and the zero-centered 3K-LEx score forms the x-axis. Each region analysis is split into two graphs for the two PI types, with solid lines representing grammatical phrases and dashed lines representing ungrammatical phrases. For the analysis of the PI region, the significant interactions are visualized in the regression plot for the ref-type PIs, where, as the 3K-LEx score decreases, there is an increasingly large gap in RRTs that opens between the grammatical and ungrammatical phrases: the grammatical phrases take relatively longer to read than the ungrammatical phrases at lower degrees of the 3K-LEx score. Note, though, that this only is evident for the ref-type PIs at the PI region.
Then, for the regression analysis of the este region, the significant effects for grammaticality and 3K-LEx score are visible for the thing-type PIs, where the ungrammatical sentences take relatively longer to read than grammatical sentences. The ungrammatical sentences also evoke significantly increasing RRTs as the 3K-LEx score increases along the x-axis, but only for the thing-type PIs. We discuss these results, and their significance to the research question, in the discussion that follows.
VII Discussion and conclusions
Our research question asked if L2 vocabulary size and individual PIs modulate the processing of sentences with NCIs and PPIs under the scope of clausal negation, in order to test whether certain types of PIs would influence the learning of polarity-sensitive phrase structures. The results indicate that individual PIs do indeed matter in the processing of these phrase structures, as supported by our results presented above, which show significant effects at both the critical PI region and the este region.
At the critical PI region, the significant three-way interaction between grammaticality, PI type, and 3K-LEx score tells us that RRTs decrease for the grammatical phrases as the 3K-LEx score increases, but particularly for the ref-type PIs. This is seen in Figure 4, at the PI region, where the lines representing grammatical and ungrammatical phrases are inversely oriented, but only for the ref-type PIs. This is to say that the participants at lower degrees of vocabulary size took relatively longer to read the grammatical phrases, which contained the NCI ninguno, than they did to read the ungrammatical phrases containing the PPI alguno. We interpret this finding to mean that the learners at lower degrees of vocabulary size were actually more surprised by the grammatical NC configuration (no tiene ninguno) than they were by the ungrammatical configuration (no tiene alguno). The results of the present study show that learners at lower degrees of Spanish vocabulary size also process the grammatical NC configuration slower than they do the ungrammatical configuration, suggesting that they have not acquired NC in Spanish to the extent that the [
Later in the sentence, at the este region, we found a significant negative modulator for grammatical phrases. Keeping in mind that the baseline level of the regresssion comprises the thing-type PIs, this tells us that the grammatical phrases provoked lower RRTs than the ungrammatical phrases, for the thing-type PIs in particular. This is visible in Figure 4, where, at the este region, participants took relatively longer to read the este word after seeing the ungrammatical structure (no tiene algo) than they did for the grammatical structure (no tiene nada). Furthermore, the significant positive modulator of 3K-LEx score shows that the increasing RRTs are associated with increases in the 3K-LEx score. In other words, although participants did not seem to notice the ungrammaticality of no tiene algo immediately upon encountering the PI, an effect does manifest itself upon reading through the spillover region. Importantly, such an effect is not to be found for the ref-type PIs, where no differences between the two lines are visible at the este region in Figure 4. Thus, this result shows that learners are capable of detecting an ungrammaticality in the placement of the PPI under the scope of negation, but primarily for the thing-type PIs.
Taking these results together, it seems that there is something about the thing-type PIs that helps learners to acquire their polarity sensitivity constraints better than with the ref-type PIs; again, this is evident at the PI region, where ninguno under the scope of negation elicits higher RRTs than alguno, and at the este region, where algo under the scope of negation elicits higher RRTs than nada. We would point to the differences in frequency between the two PI types, as demonstrated above in Table 3, which showed a clear frequency advantage, both in general and in postverbal position, for the thing-type PIs. If thing-type PIs are so much more frequent in the input than ref-type PIs, it is logical that their licensing conditions would be more salient to L2 learners. The relative infrequency of the ref-type PIs renders their licensing contexts less easy to acquire. Because ninguno is much less frequent in the input, learners at initial stages of L2 Spanish are not as familiar with its usage in NC constructions as they are with the word nada. It is only at later stages of acquisition that learners recognize ninguno under the scope of negation as grammatical and alguno under the scope of negation as ungrammatical.
There are a few more comments to be made regarding these results in relation to the previous studies highlighted above. First, our results support the findings of Mitsugi (2022) and Marsden et al. (2018), in the sense that learners are capable of acquiring polarity sensitivity in the L2, even if it differs from that of their L1. Here, we have shown that this acquisition correlates with increasing L2 vocabulary size. We point out, though, that this acquisition may not take place equally across all types of PIs. The present results are also enlightening with respect to Yamada (2022), who found no significant correlation between accurate NC syntax and PI type, comparing NCIs nada and nadie. We would contend that Yamada did not find a significant difference between those lexical items either because they do not differ enough in terms of usage frequency, or because they are simply too alike in orthographic and phonetic form.
In total, while previous studies have argued that licensing context and L2 proficiency are determining factors that influence the acquisition of polarity sensitivity, the present work has shown that the individual lexical PI itself is also a critical factor. Specifically, we have argued that L2 learners acquire the polarity sensitivity conditions of certain classes of PIs differently, when these PIs differ significantly in usage frequency and syntactic function. That is, if these PIs have widely differing distributions in the target language input, we should not expect their constraints to be acquired in an equal fashion, even if they form part of the same syntactic configuration. The contribution of the present study, thus, is a defense of the heterogenous nature of the acquisition of polarity sensitivity, based on the heterogenous nature of the PIs themselves. We conclude by recommending that researchers take the role of individual lexical items into consideration as they continue to broaden the literature on polarity in SLA.
Footnotes
Acknowledgements
We would like to thank the anonymous reviewers for their very helpful comments on this article. All errors are our own.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
1
Admittedly, these data represent an isolated snapshot of the expression of negation in Spanish. For discussion on how gesture and prosody also factor into Spanish negation and negative concord, see Prieto and Espinal, 2020.
2
For novel approaches to Spanish NCIs as fragment answers – which for some speakers can provide a double negation interpretation – and their implications for theoretical approaches to negative concord, see Espinal et al., 2015; Espinal and Tubau, 2016.
3
For Spanish, the appearance of NCIs in non-negative interrogatives has garnered different opinions. 
argues that postverbal NCIs in non-negative questions and if-phrases are disallowed in Spanish, as in (i).
(i) * ¿Quieres nada? ‘Do you want anything?’
However, Espinal (2000), drawing upon data from , argues that Spanish NCI usage in non-negative contexts is acceptable in the context of certain rhetorical questions.
4
Spanish NPIs (following the definition of NPI presently given) will not be discussed in the present work. We would direct the reader to Gutiérrez-Rexach and Schwenter’s (2003) analysis of the Spanish que digamos locution as an NPI, in addition to analysis of Spanish ‘expletive’ no as a kind of polarity item.