Research at the interface of semantics and pragmatics has advanced steadily since the field of Experimental Pragmatics began over two decades ago (Noveck and Sperber, 2012, Noveck, 2018). Arguably the most studied phenomenon therein concerns inferences derived from the quantifier some of the some, all> Horn scale (Horn, 1972, Horn, 1989). These scales are comprised of lexical items ordered in terms of informational strength. Consider (1):(1)

You got some of the gumballs.

(Degen & Tanenhaus, 2015: 668)

The strictly semantic content in (1) provides information that the listener got at least one of the gumballs (i.e., the lower-bound meaning for some). However, upon hearing (1) in conversation, a listener typically generates a scalar implicature (SI), as in (2):(2)

You got some, but not all, of the gumballs.

SIs arise because it is otherwise taken for granted that speakers are rational actors abiding by conversational maxims that promote mutually intelligible exchanges (Grice, 1975, Grice, 1991). In (1), the speaker is believed to be providing just the right amount of information required (maxim of Quantity) and to be speaking truthfully and with adequate evidence (maxim of Quality). Under these auspices, the listener reasons that the speaker could have chosen a more informative expression such as most or all but, since they did not, the implicature in (2) is derived.

Traditionally, it is argued that consideration of a stronger scalar alternative alone (e.g., most, all) is sufficient grounds for implicature derivation. Under this view, quantifiers that are logically entailed by some such as a few or none remain inactive. Nevertheless, recent theoretical and experimental work suggests that weaker and even cross-categorical quantifiers may be considered as alternatives whenever a scalar expression is used, especially if the quantifier appears amidst a clearly demarcated context where a more attractive descriptor may serve as a better fit (Noveck, 2018: p. 96). According to Gotzner’s Alternative Activation Account, a broad range of semantically related meanings are activated during pragmatic enrichment processes involving some and other scalar expressions, with contextual and grammatical constraints subsequently narrowing this set to the alternatives relevant for derivation of the some but not all meaning (Gotzner, 2017, Gotzner and Lacina, 2024).2

Though this filtering process ultimately yields the negation of stronger alternatives specifically, other paradigms may encourage participants to anticipate or regard certain words as more appropriate representations than some, even though they may not be thought of as informationally stronger in the classical sense of Horn scales and entailment relations. For example, Peloquin and Frank (2016) revealed that participants consider both weaker scalar terms and antonyms when interpreting scalar expressions from a host of different scale types (including the <some, all> scale). Skordos and Papafragou (2016) report that the presence of the antonym none biases children towards a pragmatic interpretation of some much in the same way that all does. Degen and Tanenhaus, 2015, Degen and Tanenhaus, 2016 demonstrated that when explicit numeral alternatives (e.g., four, five) accompany some in quantity descriptions, participants rate the corresponding some statements as less natural. Similarly, van Tiel (2014) revealed that some as an upper-bound construal is ill-suited for describing small set sizes. This is to suggest that even alternatives from different categories reduce the felicity of scalar expressions – an effect best interpreted as pragmatic enrichment triggered by alternative availability.3

These results notwithstanding, the consideration of alternatives is widely regarded as a core component of SI derivation proper. Chierchia (2006: 545-546) conceptualizes scalar alternatives as part of the lexical knowledge associated with a scalar term. Accordingly, whenever such an expression is used, its corresponding alternatives (i.e., scalemates) may be activated, with the degree of activation shaped by the contextual conditions wherein the term appears (De Carvalho et al., 2016, Ronai and Xiang, 2023). Therefore, it is generally assumed that hearers will consider (1) against a backdrop of relevant and contextually salient alternatives, as in (3) where the function || ||ALT serves to compare the quantifier phrase with iterations containing lexical competitors.(3)

||You got some of the gumballs||ALT =

(adapted from Chierchia, 2006: 546)

In this way, some can be viewed as having two levels of meaning: a basic logical interpretation and a scalar layer. Thus, the computation of its literal meaning and the activation of relevant alternatives constitute two essential prerequisites for generating inferences with some. There are several theoretical models which offer competing accounts of how individuals move from a basic meaning to a pragmatically enriched one. Neo-Griceans (e.g., Levinson, 2000) posit that SIs are derived automatically by default, unless context suppresses them. Relevance-theorists (Wilson & Sperber, 1995) instead argue that implicatures arise only when justified by contextual relevance, given the cognitive effort they require. A third approach – the grammatical view (e.g., Chierchia, 2006, Chierchia et al., 2012) – treats implicatures as the output of formal mechanisms such as silent exhaustivity operators and structured alternatives embedded locally in the grammar.

Experimental research has aimed to adjudicate among these models (a.o., Bott and Noveck, 2004, Breheny et al., 2006, De Neys and Schaeken, 2007, Marty and Chemla, 2013, Huang and Snedeker, 2018; for a review see Chemla & Singh, 2014) often through truth-value judgment tasks with underinformative sentences such as Some elephants are mammals (Smith, 1980). These tasks interpret rejection of such statements as evidence of scalar enrichment (some but not all), and acceptance as a lower-bound reading (some and possibly all). Studies using this method have consistently shown that pragmatic enrichment is associated with longer processing times, suggesting that implicature derivation incurs a cognitive cost (Bott and Noveck, 2004, Breheny et al., 2006; Huang & Snedeker, 2009). These findings have been widely taken to support Relevance-based or two-stage accounts in which the semantic interpretation is accessed first, and the pragmatic one is effortfully derived.4

However, recent critiques (e.g., Kissine & De Brabanter, 2023; but also Guasti et al., 2005) have questioned the assumption that rejecting an underinformative sentence necessarily reflects implicature computation. These studies suggest that such rejections may instead arise from general expectations about informativeness, task-specific strategies, or experimental context, rather than the derivation of a some but not all inference. This has motivated the development of alternative paradigms designed to assess more directly the interpretation of pragmatic some in more fine-grained, gradient contexts.

One such paradigm is the gumball task introduced by Degen and Tanenhaus, 2015, Degen and Tanenhaus, 2016 (hereafter D&T), which we adopt in the present study. D&T examined how context and alternatives influence implicature derivation under a Constraint-Based model, arguing that inferences with some arise (or are sublimated) as a result of probabilistic processes which accrue during the knowledge accumulation phase. Using a gumball paradigm – a visual judgment task where participants assess quantifier use based on gumballs distributed across containers – participants evaluated the naturalness of upper-bound some statements in contexts describing partitioned quantities of gumballs across a numerical distribution. Participants rated the statement You got some gumballs as most natural (and processed it most quickly) when it described intermediate set sizes rather than large or small ones (i.e., in regions where a few or most would have served as better descriptors). However, perceptions about the naturalness of some were reduced when more exact number phrases were placed in the context as alternatives and processing times increased at ranges where some was rated as unnatural. D&T concluded that the likelihood of implicature derivation is therefore probabilistic, affected by the availability of cues in the discourse and listeners’ consideration of alternatives. Thus, rather than some having a “core” or default interpretation based on hierarchical relations, its meaning is highly adaptable and subject to revision.5

While debates about processing cost for SIs and the default status of some have framed much of the recent experimental literature, they have also shaped the trajectory of research on implicature derivation in a second language (L2). Indeed, as explained in the next section, many L2 studies are motivated by the question of whether learners compute SIs less readily than native speakers, and whether this reflects underlying processing limitations.

The present study does not take a position in the ongoing theoretical debate between Neo-Gricean and Relevance-based accounts of scalar implicature. Rather, it addressed a different question: whether variability in L2 inferencing routines arise from cross-linguistic differences in quantifier semantics and in the thresholds that define their interpretations. Using a paradigm that elicits gradient judgements of quantifier meaning across numerical contexts, we examine where L2 speakers locate the lower and upper bounds of some and whether their judgements align with established semantic definitions. Although this inquiry concerns pragmatic judgment, we also investigate how learners further enrich the meaning of some when explicit competitors such as a few and most are introduced, and whether their response patterns reflect semantic transfer from the L1. In this way, we map the cross-linguistic dynamics of quantifier systems from both semantic and pragmatic perspectives and highlight how shifting scalar thresholds uniquely shape the pragmatic interpretation of some in the interlanguage.

L2 speakers represent a compelling test group for theories of pragmatic inference, particularly SI, because they occupy a cognitively heterogeneous space. On one end of the spectrum, adult L2 learners often face increased processing demands – slower lexical access, limited working memory, and heightened inhibitory control effort – making them especially vulnerable to cognitive bottlenecks (Green, 1998, Clahsen and Felser, 2006, Liu and Cao, 2016, Sorace, 2011). On the other end, highly proficient or balanced bilinguals may enjoy cognitive advantages, including enhanced executive control and linguistic flexibility (Adesope et al., 2010, Bialystok and Senman, 2004, Yow and Li, 2015), which could support pragmatic inferencing even under high cognitive load. This duality positions L2 populations as uniquely suited to adjudicate between competing accounts of SI computation. If, as Neo-Griceans propose, SIs are automatic and default, then even cognitively constrained L2 learners should derive them reliably. In contrast, Relevance Theory maintains that implicature derivation is context-sensitive and effortful, which has led researchers to propose that such inferences may be more susceptible to disruption in L2 contexts.

Early experimental studies sought to test these predictions using truth-value judgment tasks with underinformative statements (e.g., Some elephants have trunks) drawn from Noveck (2001) and Smith (1980). They probed whether L2 learners, like native speakers, would reject such sentences on pragmatic grounds – interpreting some as some but not all. Initial findings by Slabakova (2010) have been cited in support of the Neo-Gricean view, as L2 learners in her study frequently gave pragmatic responses. Slabakova interpreted this as evidence that scalar implicatures are easier to compute than logical interpretations. In her view, because L2 speakers are generally assumed to have more limited processing resources, their frequent pragmatic responses could not plausibly result from effortful inferencing. Rather, she argued, such results indicate that the pragmatic, upper-bounded interpretation is the default reading of some, while the logical, lower-bounded interpretation requires additional processing effort. However, this interpretation overlooks an important characteristic of her participants – highly proficient immersed L2 users who may have benefited from enhanced executive control (as later noted by Mazzaggio et al., 2021; see also Bouton, 1992). Such individuals are unlikely to face the same processing limitations as classroom learners. In fact, according to Cummins’s (1976) Threshold Hypothesis, bilingual cognitive advantages emerge once proficiency surpasses a certain threshold. It is therefore plausible that Slabakova’s participants were operating above that threshold: their pragmatic bias may have reflected enhanced executive control linked to advanced bilingualism, rather than the automaticity of implicature computation.

Notably, subsequent studies have not consistently replicated her findings. Research by Miller et al. (2016) with L1-English L2-Spanish learners, Snape and Hosoi (2018) with L1-Japanese L2-English speakers, and Dupuy et al. (2019) with French learners of Spanish and English found that L2 learners performed comparably to native speakers, suggesting that learners can access the necessary semantic and pragmatic representations without incurring substantial processing costs.

To complicate the picture, other studies have produced contrasting results that underscore the role of task demands. Research by Mazzaggio et al. (2021) with Italian learners of English and Spanish, and Khorsheed et al. (2022) and Khorsheed and van Tiel (2024) with Malay learners of English, indicates that under increased cognitive load or limited processing time – such as when test items are presented auditorily or under time pressure – L2 learners are less likely to derive implicatures and more likely to settle for literal interpretations (see also Cho, 2024). Recently, Mognon et al. (2025) showed that even when using a ternary judgment task, Dutch L2 learners of English were still sensitive to time pressure, favoring literal interpretations over pragmatic ones – particularly at lower proficiency levels. However, this trend is not uniform: Lin (2016), studying L1-Chinese L2-English learners, found robust implicature computation even under auditory time constraints. Overall, the mixed findings across online and offline paradigms highlight the need for more systematic investigation into how L2 speakers navigate the demands of pragmatic inference.

A key limitation that emerges from the L2 literature is the lack of clear hypotheses regarding the semantics of the quantifiers being examined across the learners’ L1 and their L2 (Mazzaggio et al., 2025). Most studies tacitly assume that the quantifiers responsible for triggering the implicatures are equivalent across the tested languages. However, though quantified expressions with systematic entailment patterns exist in many languages, they vary widely in their lexical, syntactic, and semantic realizations (von Fintel and Matthewson, 2008, Katsos et al., 2016). To illustrate, in a study involving 31 languages, Katsos et al. (2016) observed considerable cross-linguistic variation in adults’ SI rates involving underinformative some and most statements. Moreover, from a structural standpoint, proportional quantifiers are often derived through nominalization in Romance and Slavic languages, yielding expressions like “a majority of” or “the largest part.” In Japanese, quantifiers are commonly embedded in predicate constructions but can also function as independent noun modifiers (Imani, 1990). Additionally, partitive constructions such as some of are possible in some languages but not in others.

Quantifiers also vary considerably at the semantic level. In English, a few generally denotes more than two (Solt, 2007), whereas its Dutch counterpart enkele can refer to a single entity (Debrouwer & Schaeken, 2015). Similarly, while both quelques and certains translate to some in French, certains triggers fewer scalar inferences, at least in children (Pouscoulous et al., 2007). In Japanese, the quantifier hotondo, often linked to English most, is more accurately translated as nearly all (Hayashishita and Ueyama, 2012, Kamiya, 2012). Stateva et al. (2019) documented significant variation in the numerical bounds of existential quantifiers few, some, half, most, and almost all equivalents across Germanic, Romance, and Slavic languages. Disparate performance in a follow-up picture selection task designed to test scalar implicature derivation prompted the researchers to conclude that the point at which participants engage in inferential reasoning with some and its translational equivalents varies widely as a result of underlying cross-linguistic differences in the distributional semantics of the quantifier itself, altering the trigger points for pragmatic enrichment. These findings highlight the need to account for both lexical and conceptual differences in L2 quantifier acquisition, rather than presuming direct equivalence across languages.

The role that quantifiers play in pragmatic inference generation presents an intriguing case for L2 acquisition of SIs. Learners must master how quantifier meanings are realized in their L2, but this may be affected by semantic transfer from the L1 (a.o., Jarvis & Pavlenko, 2008). Indeed, in L2 acquisition, cross-linguistic influence (also known as language transfer) is a pervasive phenomenon whereby learners’ L1s affect the acquisition and use of the L2 across virtually all linguistic domains (Odlin, 2003). Such influence can be facilitative – yielding positive transfer when the L1 and L2 share analogous structures or patterns – or it can be obstructive, resulting in negative transfer when L1-specific forms are inappropriately applied to the L2, often leading to systematic errors (Treffers-Daller, 2009, Bardovi-Harlig and Sprouse, 2017). For instance, L2 learners often map L1 conceptual distinctions onto L2 words despite misaligned semantic boundaries, and they may carry over L1 pragmatic norms (e.g., politeness strategies or speech act conventions) into L2 communication. Jarvis and Pavlenko (2008) have used the term “conceptual transfer” to highlight how learners’ underlying conceptualization (shaped by the L1) can subtly influence L2 word meanings and interpretations. While negative transfer has been extensively studied in phonology (Goldstein & Bunta, 2012) and morphosyntax (Sorace & Filiaci, 2006), its role in L2 quantifier acquisition remains mainly underexplored.

Investigating transfer at the level of semantic specification – particularly how L2 quantifier representations may map onto conceptual meanings in a learner’s L1 – is central to the present study. It remains an open question whether learners initially associate newly acquired L2 quantifiers with specific lexical counterparts in the L1, or whether they map them more loosely onto a conceptual range occupied by an L1 equivalent. If the latter, learners may initially anchor L2 meanings to familiar L1 quantifiers and gradually refine these mappings with increased proficiency and exposure. Nevertheless, such recalibration processes may give rise to non-targetlike judgments, potentially accounting for some of the variability observed in earlier studies of L2 pragmatic inference.

Slabakova (2010) suggested that since the implicature calculation mechanism is universal, L1 transfer effects are likely not responsible for disparate performance vis-à-vis native speakers. This conclusion is based primarily on results from truth-value judgement tasks using underinformative some statements in isolation, which have been the dominant methodology in L2 SI research. However, while these tasks reveal whether learners compute implicatures in a given context, they offer limited insight into whether underlying cross-linguistic differences in quantifier semantics might contribute to variability in pragmatic inferencing behavior. Mazzaggio and Stateva (2024), for example, provide compelling evidence of negative pragmatic transfer at the semantics-pragmatics interface in the interpretation of “many”-equivalents: Italian-Slovenian bilinguals, when tested in Slovenian, failed to distinguish between the quantifiers precej and veliko, treating them as interchangeable despite their known pragmatic differentiation in native Slovenian speakers (see also Stateva & Stepanov, 2017).6 This pattern is consistent with transfer from Italian, where the equivalents molto and tanto lack this pragmatic contrast and are generally interpreted as synonymous (Mazzaggio & Stateva, 2023). It is this type of semantic-level variation that the current study investigates – namely, whether L1-specific quantifier representations influence how L2 learners interpret some in the midst of pragmatic alternatives, potentially shaping the probabilistic reasoning that underlies implicature derivation.

The present study investigates how L1-Japanese L2-English learners pragmatically enrich the meaning of some, focusing on the semantic factors that underlie this enrichment. Rather than examining the derivation of scalar implicatures in the classical sense – where some contrasts with and negates stronger alternatives such as most or all – our goal is to capture how learners adjust the interpretation of some in response to contextual and lexical alternatives more broadly. To this end, we compare some with its partitive form (some of) and with related quantifiers (a few, most) to determine how these expressions delimit the perceived lower and upper bounds of some across quantities. Crucially, across two experiments we manipulate the availability of alternatives: Experiment 1 presents some statements in relative isolation, while Experiment 2 introduces explicit competitors such as a few and most. This design allows us to test how increasing the salience of alternatives modulates the perceived naturalness of some and reveals how pragmatic enrichment adapts to the structure of the available lexicon. Finally, rather than seeking to adjudicate between competing theories of implicature derivation, we adopt a Constraint-Based perspective, which emphasizes how multiple interacting factors such as transfer, contextual support, and the availability of alternatives jointly shape pragmatic enrichment in L2 speakers.

To operationalize this investigation, we employ D&T’s (2015) gumball paradigm which elicits naturalness judgements of quantifier expressions across a continuous range of set sizes. While such ratings do not directly index implicature computation, they provide a sensitive measure of how L2 speakers map quantificational meanings onto referential contexts. This method allows us to pinpoint where quantifiers are perceived as most felicitous, offering insight into the scalar thresholds learners associate with each term. Importantly, it also enables us to examine how some is pragmatically enriched when contextually stronger or weaker alternatives are present, revealing how the availability of competitors shapes L2 speakers’ inferential behavior. In this way, the paradigm offers a controlled means of testing whether semantic transfer from the L1 affects the anchoring of scalar meanings in the L2 – an essential precursor to interpreting any observed variability in implicature-related behavior.

Japanese is a particularly informative test case given its distinct quantifier system, which differs from English in both lexical realization and scalar structure (Imani, 1990, Minagawa, 2008, Sawada, 2013, Sawada, 2016, Hayashishita and Ueyama, 2012). By comparing responses from native and L2 speakers across matched experimental contexts, we assess whether differences in quantifier semantics influence learners’ interpretations in systematic ways. In doing so, we challenge the assumption that scalar expressions are interpreted in a universal way. Instead, we propose that scalar reasoning in the L2 is shaped by language-specific semantic constraints inherited from the L1 and that learners’ probabilistic expectations about the applicability of alternatives is similarly derived from their native language system.