2.1. On scalar implicatures in adult and child language

In communication, participants often rely on inferences. Consider, for instance, the example in (5), where the sentence uttered by B is interpreted as in (6), an assumption inferentially derived based on what the speaker said. While some is generally argued to mean “some and possibly all”, in (5), its meaning becomes enriched – strengthened to “not all”. This additional meaning is not part of what the speaker has said, yet it arises in communication due to the hearer’s ability to go beyond what is said and derive the intended meaning. In fact, the strengthening of some to “not all” is so pervasive that participants tend to do it even with sentences that are logically true such as (7), thus ending up rejecting these sentences as false 60% of the time (Bott & Noveck, Reference Bott and Noveck2004).

Similarly to the sentence containing some, the sentence containing the disjunction marker or in (8) is also interpreted as giving rise to the strengthened meaning in (9), an additional meaning that is also arrived at via an inference.

A common assumption since Grice (Reference Grice, Cole and Morgan1975) is that the basic meaning of some is “some, possibly all” and that the strengthened “some, but not all” interpretation is derived from the basic one via an implicature; similarly, the basic meaning of or is “or, possibly and,” and the strengthened “or, but not and” interpretation is derived from the basic one via an implicature. Implicatures are inferences that we draw in conversation, arising from the apparent non-observance of certain maxims of communication (Grice, Reference Grice, Cole and Morgan1975, Reference Grice1989). Interlocutors are typically expected to provide truthful, informative, relevant, and appropriately phrased contributions, “to make their conversational contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which they are engaged” (Grice Reference Grice, Cole and Morgan1975, p. 45). However, in certain cases, speakers appear to flout certain maxims, expecting their hearers to figure out the intended meaning themselves. In the case of scalar implicatures, according to Grice (Reference Grice, Cole and Morgan1975, Reference Grice1989) and neo-Griceans (e.g., Gazdar, Reference Gazdar1979; Horn, Reference Horn1972), speakers flout the Maxim of Quantity, in particular the submaxim presented in (10).

Scales (also called Horn scales) order items in terms of informational strength, such that for the scales $ \langle $ all, some $ \rangle $ or $ \langle $ and, or $ \rangle $ ,Footnote ¹ a sentence containing all entails a sentence containing some but not vice versa, and a sentence containing and entails a sentence containing or but not the other way around. When a speaker utters a weak term such as some or or, this means that they have chosen not to articulate the stronger, more informative alternatives all and and. This may happen either because the speaker is ignorant about whether all/and can be used or knows that it cannot.

While strengthened meanings of the type “some, not all” and “or, not and” are often explained pragmatically by means of implicatures, there are multiple possible accounts in the literature (pragmatic, lexical, and grammatical). Even pragmatic theories do not fully agree on how implicatures are generated. According to Gricean and Neo-Gricean pragmatic theories (e.g., Horn, Reference Horn1972; Levinson, Reference Levinson1983, Reference Levinson2000), scalar implicatures are generated automatically for certain lexical items and independently of context. In contrast, according to Relevance Theory (Carston, Reference Carston and Kempson1988; Sperber & Wilson, Reference Sperber and Wilson1995), scalar implicatures are derived only when the context makes them relevant. This latter view is supported by findings from Bott and Noveck (Reference Bott and Noveck2004), showing that participants are less accurate and take longer to answer correctly in case of a “some but not all” interpretation rather than of a “some and possibly all” interpretation. According to Lexical Theories (Chierchia, Reference Chierchia and Belletti2004; Levinson, Reference Levinson2000), scalar terms have two meanings: a weak meaning and a strong meaning. Interestingly, according to these approaches, the strong implicature meaning is stored in the lexicon, and the weak meaning is derived by means of implicature cancellation. Finally, grammatical theories (Chierchia, Reference Chierchia2006; Fox, Reference Fox, Sauerland and Stateva2007) argue that implicatures are derived by inserting a silent exhaustification operator EXH. EXH(P) asserts the meaning of P and negates the stronger alternatives of P, as below.

There is an ongoing debate about which of these theories best accounts for implicature derivation, with no clear consensus having been reached (Sauerland, Reference Sauerland2012).

As far as implicatures in child language are concerned, the literature seems to suggest that children struggle with implicatures at an early stage, preferring the weak logical meaning of scalar terms (“some, possibly all” and “or, possibly and”) over the strong one. Noveck (Reference Noveck2001) was the first to conduct a systematic investigation of the development of scalar implicatures with quantifiers and modals like x might be y by means of the Truth Value Judgement Task (TVJT). His experiments suggest that French- and English-speaking children tend to be more logical than adults. Papafragou and Musolino (Reference Papafragou and Musolino2003) also found that, in a regular TVJT, Greek children were only adult-like with numerical scales but not with the scales $ \langle $ all, some $ \rangle $ and $ \langle $ finish, start $ \rangle $ . Similar findings were uncovered for Romanian by Stoicescu et al. (Reference Stoicescu, Sevcenco, Avram, Burada, Tatu and Sinu2015), Bleotu et al. (Reference Bleotu, Benz and Gotzner2021a), and Bleotu et al. (Reference Bleotu, Benz and Gotzner2025a): young children struggle with implicatures but gradually become more adult-like as they get older (between ages 7 and 9); nevertheless, they are more adult-like with cardinals from earlier on (Bleotu, Reference Bleotu, Tǎnase-Dogaru, Tigǎu and Zamfirescu2021).

Children’s challenges with scalar implicatures have received multiple accounts in the literature (pragmatic, lexical, grammatical, and processing). According to Neo-Gricean pragmatic accounts (Horn, Reference Horn1972; Levinson, Reference Levinson1983, Reference Levinson2000), children may experience pragmatic delay, that is, they simply lack certain pragmatic abilities enabling them to derive implicatures, and they are more logical at this stage.

According to Relevance Theory (Carston, Reference Carston and Kempson1988; Sperber & Wilson, Reference Sperber and Wilson1995), children’s lower rates of implicatures compared to adults have to do with difficulties in accessing the strong scalar alternatives. Importantly, the prediction is that in a context where the stronger alternative is made relevant, children should derive implicatures. According to Lexical Theories (Chierchia, Reference Chierchia and Belletti2004; Levinson, Reference Levinson2000), children know the meaning of some/or; they have just not yet associated the lexical entry with the scale $ \langle $ all, some $ \rangle $ / $ \langle $ and, or $ \rangle $ (Guasti et al., Reference Guasti, Chierchia, Crain, Foppolo, Gualmini and Meroni2005). According to Grammatical Theories (Chierchia, Reference Chierchia2006; Fox, Reference Fox, Sauerland and Stateva2007), implicatures are derived through a covert exhaustivity operator, EXH (equivalent to silent only), which affirms a proposition and excludes its stronger alternatives. Children are not adult-like in their construction of the set of alternatives they can exhaustify over. Nevertheless, they seem to be able to employ EXH, given the fact that they fail to derive implicatures with only some but do not fail with only the cat and the dog (Barner et al., Reference Barner, Brooks and Bale2011). Additionally, according to a Processing Theory, children’s difficulty with implicatures may also be related to task processing demands, such as the presence of distractors in the visual context, the use of tasks that impose additional difficulties on children (e.g., the TVJT), and the failure to contextualise utterances. Once task demands are simplified, children seem to perform in a more adult-like manner (Guasti et al., Reference Guasti, Chierchia, Crain, Foppolo, Gualmini and Meroni2005; Pouscoulous et al., Reference Pouscoulous, Noveck, Politzer and Bastide2007).

In addition to the theories of implicature above, various recent studies have proposed an Alternatives-Based Account, according to which children’s difficulties with implicatures stem from their inability to access the lexical alternatives required to compute the relevant implicatures. Once the stronger alternative is made available, children seem to perform in a more adult-like manner (Barner et al., Reference Barner, Brooks and Bale2011; Bill et al., Reference Bill, Pagliarini, Romoli, Tieu and Crain2021; Chierchia et al., Reference Chierchia, Crain, Guasti, Gualmini, Meroni, Amy, Dominguez and Johansen2001; Pagliarini et al., Reference Pagliarini, Bill, Romoli, Tieu and Crain2018; Singh et al., Reference Singh, Wexler, Astle-Rahim, Kamawar and Fox2016; Tieu et al., Reference Tieu, Romoli, Zhou and Crain2016). Note that multiple variants of the Alternatives-Based Account are possible, which may place children’s difficulty with alternatives in different domains – pragmatic, lexical, and grammatical.

However, while children are generally assumed not to derive implicatures to the same extent as adults, it bears mentioning that this is a matter of empirical debate, as there appears to be significant variation in implicature rates across different tasks, with children deriving more implicatures in tasks that involve a higher level of engagement compared to TVJTs. Binary TVJTs tend to be more challenging for children, whereas more interactive tasks – such as giving rewards, moving objects, or colouring and erasing – often result in higher implicature rates.

Reward tasks have been found to boost implicatures. In their study, Papafragou and Tantalou (Reference Papafragou and Tantalou2004) used a reward task in which children had to reward an animal for how well it had described its own actions. For example, an animal had to colour the stars (four in number); it coloured all of them and then reported this as I coloured some. Children were found to perform in a rather adult-like manner: they computed scalar implicatures and refused to reward the elephant with a prize in such cases. In another reward task conducted by Katsos and Bishop (Reference Katsos and Dorothy2011), children were asked to offer a “small”, “big”, or “huge” strawberry as a reward to Mr. Caveman, depending on how good the speaker’s responses were. Children rewarded fully informative responses by giving the speaker “huge” strawberries, underinformative ones by giving “big” strawberries, and false responses by giving “small” strawberries. Their responses indicated sensitivity to underinformativeness, previously obscured in binary tasks. Children have also been found to be more adult-like in act-out tasks. Pouscoulous et al. (Reference Pouscoulous, Noveck, Politzer and Bastide2007) conducted an act-out task targeting implicatures with the existential quantifiers quelques “ some₁” and certains “some₂” in French, where a puppet utters I would like some boxes to contain a token when the scenario displays each of five boxes already containing a token. If participants take some to be compatible with all, then they should leave the boxes unchanged; otherwise, they should remove at least one token. The task increased implicature production with quelques “some₁” for children of various ages (4, 5, and 7). In an act-out task (a colouring task) conducted by Bleotu (Reference Bleotu2019, Reference Bleotu, Sevcenco, Stoica, Stoicescu, Tănase-Dogaru, Tigău and Tomescu2024) in Romanian, children have also been found to perform in an adult-like manner (see also Bleotu et al., Reference Bleotu, Panaitescu, Benz, Nicolae, Bîlbîie and Tieu2025d, for a recent extension of this method to disjunction). Felicity judgement tasks, where children have to choose between a weak scalar sentence (containing some or or) and a strong scalar sentence (containing all or and), have also been shown to lead to more adult-like behaviour (Chierchia et al., Reference Chierchia, Crain, Guasti, Gualmini, Meroni, Amy, Dominguez and Johansen2001; Foppolo et al., Reference Foppolo, Guasti and Chierchia2012). Additionally, situating the utterances children have to evaluate in a story context also seems to boost implicature rates (Guasti et al., Reference Guasti, Chierchia, Crain, Foppolo, Gualmini and Meroni2005; Papafragou & Musolino, Reference Papafragou and Musolino2003).

The increase in implicatures could be due to a generally higher level of engagement with all the above-mentioned tasks in a more concrete way than with TVJTs. It could also be due to more contextual relevance (in the act-out, reward, and story-based tasks), as well as access to stronger alternatives and contrast (in felicity judgement tasks). In the next subsection, we focus on contextual relevance and alternatives, drawing on further evidence from the literature and considering its theoretical implications.

2.2. On the role of contextual relevance and alternatives

On the role of relevance: Relevance has been argued to impact the derivation of pragmatic inferences. As remarked by Papafragou and Musolino (Reference Papafragou and Musolino2003), “if preschoolers, unlike adults, cannot readily infer the pragmatic nature of the task, and are not given adequate motivation to go beyond the truth-conditional content of the utterance, they may readily settle for a statement which is true but does not satisfy the adult expectations of relevance and informativeness” (p. 269). We shall refer to this account as the Relevance-Based Account.

Both adults and children have been found to derive more implicatures in a context that makes the strengthened meaning more relevant (Bleotu et al., Reference Bleotu, Benz and Nicole2022b; Degen, Reference Degen2013; Guasti et al., Reference Guasti, Chierchia, Crain, Foppolo, Gualmini and Meroni2005; Ronai & Xiang Reference Ronai and Xiang2021a, b; Skordos & Papafragou, Reference Skordos and Papafragou2016; Yang et al., Reference Yang, Minai and Fiorentino2018; Zondervan et al., Reference Zondervan, Meroni and Gualmini2008).

Contextual relevance has typically been manipulated either (i) through a background story or (ii) through a question.

Stories that focus on the quantity of the participants or objects in the story have been found to boost implicatures. For instance, in Guasti et al. (Reference Guasti, Chierchia, Crain, Foppolo, Gualmini and Meroni2005), children were adult-like, deriving implicatures at a rate of 75%. In one story, there were five soldiers who had to go collect a treasure somewhere far away, and they could either go by motorbike or ride a horse. The soldiers started to discuss among themselves: some soldiers said they would go by motorbikes, which are fast, while other soldiers argued that they would ride a horse, which would be less expensive. After this discussion, they all decided to ride horses. Carolina, a puppet, was then asked to say what was happening in the story. Carolina’s utterance was Some soldiers are riding a horse. Children then had to say whether what Carolina had said was “right” or “wrong”. Similarly to adults, children rejected Carolina’s underinformative utterances, deriving implicatures.

In another important experimental study, which served as a source of inspiration for our current investigation of disjunction, Skordos and Papafragou (Reference Skordos and Papafragou2016) probed the role of relevance and alternatives in children’s derivation of implicatures associated with the quantifier some. In their first experiment, similarly to Foppolo et al. (Reference Foppolo, Guasti and Chierchia2012), they exposed children to utterances where all was used in both true and false contexts and to utterances where the use of some was either pragmatically appropriate or not. These utterances were presented in various orders, for example, a Mixed order, where some and all trials were intermixed in a pseudorandomised order, making the stronger lexical scale member all highly accessible during the evaluation of some statements, and a Some-First order, where the utterances with some preceded those containing all, thus making the stronger lexical scale member unavailable to children prior to evaluating some. Skordos and Papafragou (Reference Skordos and Papafragou2016) found that children derived more implicatures in the Mixed order than in the Some-First order, a finding suggesting that access to stronger alternatives is important for implicature derivation.

Their second experiment revealed, however, that what affects the rate of implicature calculation is actually whether the alternative is made relevant in the context. The second experiment tested children’s interpretation of utterances such as Some of the blickets have a scarf in two conditions: a Quantity condition, where participants were guided through linguistic and visual cues to pay attention to whether all or some of the creatures had a scarf (the implicit QUD was Do all or only some of the blickets have a crayon?), and an Object condition, where they were encouraged to pay attention to the kinds of objects the blickets had (the implicit QUD was Do the blickets have a crayon or another object?). For example, in the Quantity condition, All of the blickets have a scarf would be false because three out of four blickets would have a scarf, but in the Object condition, it would be false because all four blickets would have a shovel. The results revealed that children derived more implicatures from potentially underinformative utterances containing some in the Quantity condition, which made the contrast between some and all relevant in the context. This suggests that alternatives matter for implicature derivation, but only in a context that makes them relevant.

In a third experiment, Skordos and Papafragou went one step further, exposing children to utterances containing none or all prior to utterances containing some. Interestingly, children derived implicatures at a similar rate in both the None-First and the All-First conditions. This finding led Skordos and Papafragou (Reference Skordos and Papafragou2016) to argue that explicitly providing stronger alternatives is not necessary, as long as the context makes them relevant.

Contextual relevance can also be manipulated through questions, a manipulation motivated by the idea that, ultimately, any sentence is to be understood as an answer to a question that introduces the QUD (Gualmini et al., Reference Gualmini, Hulsey, Hacquard and Fox2008; Hulsey et al., Reference Hulsey, Hacquard, Fox, Gualmini, Csirmaz, Gualmini and Nevins2004). This idea has been formulated as the Question–Answer Requirement (Hulsey et al., Reference Hulsey, Hacquard, Fox, Gualmini, Csirmaz, Gualmini and Nevins2004):

To give some concrete examples, adults have been shown to derive more implicatures with single-scale utterances that represent answers to explicit questions, such as those in (13) and (14) (Ronai & Xiang Reference Ronai and Xiang2021a, Reference Ronai and Xiangb), particularly if the question contains a strong scalar term, such as excellent in (13) or all in (14).

Additionally, Bleotu and Benz (Reference Bleotu and Benz2024) recently investigated adults’ derivation of embedded implicatures, which arise when an utterance contains more than one scalar term. For instance, an utterance such as Some meals are adequate contains weak terms from two different scales: $ \langle $ all, some $ \rangle $ and $ \langle $ good, adequate $ \rangle $ . Bleotu and Benz (Reference Bleotu and Benz2024) found that adults tended to derive more implicatures from the weak scalar term that appeared in the second/embedded position (i.e., deriving Some meals are adequate and not good) when the QUD included the weak term of the first scale (e.g., some) and the strong term of the second (embedded) scale (e.g., good), as in (15), compared to when there was no such QUD.

Building on Bleotu et al. (Reference Bleotu, Benz and Gotzner2021b, Reference Bleotu, Benz and Gotzner2022a) and Bleotu et al. (Reference Bleotu, Benz and Gotzner2021a), which was further developed into Bleotu et al. (Reference Bleotu, Benz and Gotzner2025a), Bleotu et al. (Reference Bleotu, Benz and Nicole2022b) employed a Shadow Play Paradigm (Figure 1) to test the influence of a scalar question introducing a QUD upon Romanian adults’ and children’s interpretation of utterances such as those in (16), embedding a scalar term belonging to the scale $ \langle $ all, some $ \rangle $ under a scalar term belonging to the scale $ \langle $ certain, possible $ \rangle $ . Participants saw eight dogs enter the game, four of which remained in front, while the silhouettes of four others could be seen hidden behind a curtain. Participants had to infer the identity of these silhouettes based on an utterance they heard (16) and visual cues (the dogs that are in front of the curtain).

Figure 1. Example picture from Bleotu et al. (Reference Bleotu, Benz and Nicole2022b).

Bleotu et al. (Reference Bleotu, Benz and Nicole2022b) conducted two experiments: Experiment 1, where the question involved the $ \langle $ certain, possible $ \rangle $ scale, and Experiment 2, where the question involved the $ \langle $ all, some $ \rangle $ scale (see (17)).

Romanian children and adults were both found to derive more global implicatures of the type It is not certain that some dogs are blue (GI $ {}_{NotCertainSome} $ ) in the $ \langle $ certain, possible $ \rangle $ QUD experiment than in the $ \langle $ all, some $ \rangle $ QUD one. Thus, contextual manipulations seem to matter for implicature derivation both for adults and children.

On the role of access to alternatives: One possible explanation for children’s challenges with implicatures could have to do with the failure to retrieve the stronger scalar mate from the lexicon (Barner et al., Reference Barner, Brooks and Bale2011; Chierchia et al., Reference Chierchia, Crain, Guasti, Gualmini, Meroni, Amy, Dominguez and Johansen2001; Singh et al., Reference Singh, Wexler, Astle-Rahim, Kamawar and Fox2016; Tieu et al., Reference Tieu, Romoli, Zhou and Crain2016). When encountering some or or, children could simply fail to retrieve all or and from the lexicon. Once these alternatives are made accessible, children’s implicature rates are expected to increase. We shall refer to this explanation as the Alternatives-Based Account, an account compatible with multiple explanations for the source of implicatures (pragmatic, lexical, or grammatical).

In this vein, Chierchia et al. (Reference Chierchia, Crain, Guasti, Gualmini, Meroni, Amy, Dominguez and Johansen2001) conducted a Felicity Judgement Task in which children heard both or and and descriptions of the same contexts. For instance, children were told a story about some farmers who were cleaning their animals. After looking at all of the animals, each farmer decided to clean a horse and/or a rabbit. At this point, the two puppets provided an alternative description of the story (see (18)).

Interestingly, children were 93.3% accurate, choosing the utterance containing and in situations where every farmer cleaned both a horse and a rabbit and the utterance containing or in situations where every farmer cleaned only one animal. This result shows us that, when given two utterances in contrast, children are able to identify the most informative one. However, it does not alone provide evidence that children have the ability to derive scalar implicatures, a more complex process that additionally involves the participants’ understanding that some conveys the meaning not all.

The idea that children struggle with implicatures because they have a hard time retrieving the stronger scale mates from the lexicon is also supported by the work of Barner et al. (Reference Barner, Brooks and Bale2011). They provide experimental evidence that children are able to derive implicatures when only modifies lexical Determiner Phrases (DPs), but they find it much harder to do so when only modifies some (see (19)).

Thus, children are not unable to derive implicatures; rather, they are simply unable to lexically associate the weak scalar term with the stronger scalar term.

Recent research on free choice inferences (from The hen is allowed to push the train or the boat to The hen is allowed to push the train and the hen is allowed to push the boat) also suggests that children have no difficulty deriving implicatures when they do not have to retrieve alternatives from the lexicon, as is generally assumed to be the case with free choice inferences (Barner et al., Reference Barner, Brooks and Bale2011; Chierchia et al., Reference Chierchia, Crain, Guasti, Gualmini, Meroni, Amy, Dominguez and Johansen2001; Gualmini et al., Reference Gualmini, Crain, Meroni, Chierchia and Guasti2001; Tieu et al., Reference Tieu, Romoli, Zhou and Crain2016).

Importantly, when children have to retrieve lexical alternatives, being exposed to them helps. In multiple experiments on some, Foppolo et al. (Reference Foppolo, Guasti and Chierchia2012) have shown that children are more adult-like in their interpretation when they have access to stronger alternatives containing all. While in a classical TVJT (Experiment 1), children derived implicatures at a rate of only 42%, in a Felicity Judgement Task (Experiment 5), their performance was adult-like at a 95% rate: when all the chipmunks in a picture were taking a shower and two puppets were describing this situation, one using an appropriate, informative sentence (All chipmunks are taking a shower), and the other using an underinformative but true sentence (Some chipmunks are taking a shower), children were able to correctly pick the puppet uttering the all sentence. Moreover, in Experiment 6, where, before being exposed to underinformative utterances with some, children were exposed to correct and incorrect uses of all, the rates of implicatures were also relatively high (72.5%) compared with 42% in Experiment 1. More precisely, participants were exposed to (i) a situation in which all was used correctly (e.g., the sentence All the Smurfs went on a boat describing a situation in which five out of five Smurfs went on a boat), as well as (ii) a situation in which all was incorrectly used, and where some would have been appropriate (e.g., the sentence The dwarf picked up all the carrots describing a situation in which a dwarf picked up three out of five carrots). Only after being acquainted with these correct and incorrect uses of all did participants have to evaluate utterances with some. This double exposure helped children interpret some in a more adult-like manner.

Finally, as detailed previously, Skordos and Papafragou (Reference Skordos and Papafragou2016) also investigated whether the presence of stronger alternatives helps children derive more implicatures with the quantifier some. Importantly, they found that, while access to alternatives may matter (Experiment 1), it is not enough to boost implicatures in the absence of a relevant context (Experiment 2). Moreover, relevance alone may have this effect even in the absence of the stronger all alternatives (Experiment 3). The nuanced perspective upon alternatives put forth in Skordos and Papafragou (Reference Skordos and Papafragou2016) allows us to distinguish between two versions of the Alternatives-Based Account: an Alternatives-Only Account, which assumes exposure to alternatives is enough to boost implicatures, and a Relevant Alternatives Account, which assumes that access to alternatives matters only in relevant contexts, and that relevance plays a crucial part in boosting implicatures.

Interestingly, while Skordos and Papafragou (Reference Skordos and Papafragou2016) try to tease apart the role of context and alternatives in interpreting some, most of the literature conflates the role of context and the role of access to alternatives. This is mainly because, in many of the experiments conducted on implicatures, multiple manipulations are typically implemented simultaneously. For instance, when the context manipulation occurs via the QUD, as in Ronai and Xiang (Reference Ronai and Xiang2021a) and Ronai and Xiang (Reference Ronai and Xiang2021b), the question itself contains a strong scalar term, such that both context and lexical access to the stronger scale-mate may be argued to boost implicatures.

2.3. Disjunction in child language

As far as the interpretation of disjunction in child language is concerned, there is a vast literature that shows that children behave differently from adults. Notably, adults tend to interpret simple disjunctions exclusively and inclusively while showing a preference for exclusive interpretations in the case of complex disjunctions (Chierchia et al., Reference Chierchia, Crain, Guasti, Gualmini, Meroni, Amy, Dominguez and Johansen2001; Gualmini et al., Reference Gualmini, Crain, Meroni, Chierchia and Guasti2001; Nicolae & Sauerland, Reference Nicolae, Sauerland, Bade and Schöller2016; Nicolae et al., Reference Nicolae, Petrenco, Tsilia and Marty2024, Reference Nicolae, Petrenco, Tsilia, Marty, Baumann, Gutzmann, Koopman, Liefke, Renans and Scheffler2025, among others). In contrast, children have been shown to interpret both simple and complex disjunctions in more varied ways: inclusively, conjunctively, and exclusively. As observed by Singh et al. (Reference Singh, Wexler, Astle-Rahim, Kamawar and Fox2016) for English, Tieu et al. (Reference Tieu, Yatsushiro, Cremers, Romoli, Sauerland and Chemla2017) for French and Japanese, and Bleotu et al. (Reference Bleotu, Ivan, Nicolae, Bîlbîie, Benz, Panaitescu and Tieu2023) for Romanian, children interpret disjunction inclusively (The hen pushed one and possibly both) or conjunctively (The hen pushed both). However, German children interpret these inclusively or exclusively (Sauerland & Yatsushiro, Reference Sauerland and Yatsushiro2018). Interestingly, children generally treat simple and complex disjunctions on a par, unlike adults, who are generally more prone to exclusive readings with complex disjunctions (Nicolae et al., Reference Nicolae, Petrenco, Tsilia and Marty2024, Reference Nicolae, Petrenco, Tsilia, Marty, Baumann, Gutzmann, Koopman, Liefke, Renans and Scheffler2025; Spector, Reference Spector2014).

In theoretical terms, children’s inclusive behaviour has been typically explained as a logical interpretation of disjunction, along the lines of Noveck’s (Reference Noveck2001) claim that children are generally more logical than adults.

According to the Relevance-Based Account (Degen, Reference Degen2013; Guasti et al., Reference Guasti, Chierchia, Crain, Foppolo, Gualmini and Meroni2005; Ronai & Xiang Reference Ronai and Xiang2021a, Reference Ronai and Xiang2021b; Skordos & Papafragou, Reference Skordos and Papafragou2016; Yang et al., Reference Yang, Minai and Fiorentino2018; Zondervan et al., Reference Zondervan, Meroni and Gualmini2008, among others), children should become more exclusive, deriving more implicatures with disjunction, if the disjunctive utterance is embedded in a pragmatically relevant context. According to the Alternatives-Based Account (Barner et al., Reference Barner, Brooks and Bale2011; Tieu et al., Reference Tieu, Romoli, Zhou and Crain2016, Reference Tieu, Yatsushiro, Cremers, Romoli, Sauerland and Chemla2017, among others), children should become more exclusive if they have access to stronger conjunctive alternatives, which increase their awareness of the $ \langle $ and, or $ \rangle $ scale.

In contrast, no consensus has yet been reached with respect to children’s conjunctive interpretation of disjunction. Several possible explanations have been proposed, however. One such proposal put forth by Singh et al. (Reference Singh, Wexler, Astle-Rahim, Kamawar and Fox2016) and endorsed by Tieu et al. (Reference Tieu, Yatsushiro, Cremers, Romoli, Sauerland and Chemla2017) is the implicature account, according to which this interpretation is derived by children as an implicature. One way to implement this is via recursive exhaustification, that is, children are actually able to exhaustify, but, differently from adults, they do so over a different set of alternatives. While adults have lexical access to the alternatives A, B, and A and B, children only have access to the alternatives A and B (see Table 1).

Table 1. Alternatives accessed by children and adults

According to Singh et al. (Reference Singh, Wexler, Astle-Rahim, Kamawar and Fox2016), adults derive implicatures from disjunction by accessing and then negating the stronger conjunctive alternative, as in (20):

In contrast, children only have access to the individual disjunct alternatives, and consequently exhaustify over them. Exhaustification proceeds recursively (21): at Step 1, children exhaustify separately over each of the disjunct members ((21b) and (21c)); at Step 2, they exhaustify over each disjunct member separately once again, and then conjoin the resulting meanings (21d). In conjunction with the initial disjunctive meaning, the result is the conjunctive interpretation in (21e).

Another proposal put forth in the literature is that disjunction is ambiguous between disjunction and conjunction (Sauerland & Yatsushiro, Reference Sauerland and Yatsushiro2018). It may be that, at an early stage in development, children disambiguate between these two meanings by observing the Strongest Meaning Principle (22), reminiscent of the Semantic Subset Principle (Crain et al., Reference Crain, Ni, Conway, Clifton and Rayner1994).

Sauerland and Yatsushiro (Reference Sauerland and Yatsushiro2018) argue that, contrary to the implicature account, the ambiguity approach predicts the absence of conjunctive interpretations for complex disjunctions, which are unambiguous and tend to generally express exclusive meanings. Their claim, however, is not supported by the data from German children, whose responses varied between inclusive and exclusive (see Sauerland & Yatsushiro, Reference Sauerland and Yatsushiro2018, for a more detailed discussion).

A related, yet different proposal is that children’s conjunctive interpretation of disjunction is a semantic default, that is, children start out thinking that disjunction has the same meaning as a conjunction (Aloni et al., Reference Aloni, Klochowicz and Sbardolini2024), essentially because of two cognitive biases: a neglect zero bias, driving children towards avoiding empty configurations, and a no split bias, driving children not to split states that involve entertaining several alternatives. Children gradually develop from a conjunctive interpretation to an inclusive one, where the ability to split states has been acquired, and then to an exclusive one, where children additionally develop scalar reasoning. The proposal differs from the ambiguity account in that it assumes that children initially start out only with the conjunctive meaning rather than both conjunctive and inclusive meanings.

Finally, another proposal is that the conjunctive interpretation of disjunction is not a real interpretation grounded in grammar but rather an experimental artefact, a repair strategy motivated by the experimental set-up (Huang & Crain, Reference Huang and Crain2020; Skordos et al., Reference Skordos, Feiman, Bale and Barner2020). Both Huang and Crain (Reference Huang and Crain2020) and Skordos et al. (Reference Skordos, Feiman, Bale and Barner2020) make this claim as a comment on the experimental design used in Tieu et al. (Reference Tieu, Yatsushiro, Cremers, Romoli, Sauerland and Chemla2017): a TVJT in a predictive mode, where a puppet would make a guess about the actions of a character with respect to two objects, and participants had to evaluate whether the puppet guessed well or not. According to Huang and Crain (Reference Huang and Crain2020), making a guess in the form of a disjunction when there are only two objects in the background is not felicitous in a guessing game context, as it does not add new information to the discourse context. Consequently, as a repair strategy, children will default to the more informative conjunctive interpretation. Huang and Crain (Reference Huang and Crain2020) and Skordos et al. (Reference Skordos, Feiman, Bale and Barner2020) argue that adding more objects to the background leads to the disappearance of conjunctive interpretations from children’s responses, a claim supported by evidence from experiments with disjunctive statements and three objects in the background instead of two.

Interestingly, of the accounts presented above (the implicature account, the ambiguity account, the conjunctive default account, and the experimental artefact account), only the implicature account clearly predicts that lexical access to alternatives leads to more exclusivity (as a result of an implicature boost). We elaborate on this in Section 4.1. But before doing so, it is important to dwell a bit on the importance of investigating disjunction in Romanian.

4.1. Predictions

Overall, we expect that adults should be exclusive in all three experiments, given that they have been shown to be exclusive with disjunction even in the absence of a stronger conjunctive alternative or a relevant QUD (see Bleotu et al., Reference Bleotu, Ivan, Nicolae, Bîlbîie, Benz, Panaitescu and Tieu2023; Tieu et al., Reference Tieu, Yatsushiro, Cremers, Romoli, Sauerland and Chemla2017). Potentially, we could see an increase in implicatures in Experiment 2 (Alternatives) and Experiment 3 (Alternatives & QUD) compared with Experiment 1 (Baseline), based on previous findings that even adults may derive more implicatures in the presence of stronger alternatives (see, e.g., Bleotu & Benz, Reference Bleotu and Benz2024; Ronai & Xiang, Reference Ronai and Xiang2021a).

As far as children are concerned, we predict that children should in principle interpret disjunction inclusively in Experiment 1 (Baseline), given that they are assumed to start out by being more logical in their interpretation (Noveck, Reference Noveck2001). In light of previous findings from Singh et al. (Reference Singh, Wexler, Astle-Rahim, Kamawar and Fox2016) and Tieu et al. (Reference Tieu, Yatsushiro, Cremers, Romoli, Sauerland and Chemla2017), some children might also display evidence of conjunctive interpretations of disjunction.

As for Experiment 2 (Alternatives) and Experiment 3 (Alternatives & QUD), different accounts make different predictions. The Alternatives-Based Account assumes children’s difficulty with implicatures stems from a difficulty in the retrieval of lexical alternatives. In an Alternatives-Only version of the account, mere access to alternatives should be enough to boost implicatures, so children should be more exclusive in both Experiments 2 and 3, given that they are exposed to conjunction (as part of assertions in Experiment 2 and questions in Experiment 3). Importantly, given previous findings from Foppolo et al.’s (Reference Foppolo, Guasti and Chierchia2012) Experiment 6 (children derived more implicatures with some when they had access to both true and false stronger all alternatives), we expect a significant boost in implicatures in Experiment 2 (Alternatives) compared to Experiment 1 (Baseline), given that in Experiment 2, participants are made more aware of the use of and by being exposed to both true and false conjunctive utterances. They should thus be more likely to derive a not both implicature for or. We also expect a boost in implicatures in Experiment 3 (Alternatives & QUD), given that the questions participants are exposed to contain conjunction.Footnote ⁵ In the Relevant Alternatives version of the account, where relevance plays a specified role in implicature derivation, children should show a boost in exclusivity implicatures in Experiment 3, but not necessarily in Experiment 2.

Finally, the Relevance Account, which assumes that children’s difficulty with implicatures stems from a failure to accommodate the utterances at issue in a relevant context, is similar to the Relevant Alternatives version of the Alternatives-Based Account. It also predicts that children should be (more) exclusive with disjunction only in Experiment 3 (Alternatives & QUD), where the disjunctive utterance is presented as an answer to an explicit question.

4.2. Participants

We collected data from 500 participants: 257 typically developing monolingual Romanian-speaking 5- and 6-year-old children and 243 adult native-speaker controls. Experiment 1 (Baseline) was conducted with 85 children (mean age 5;04) and 71 adults. More specifically, we tested 27 children and 21 adults on marked sau, another group of 28 children and 27 adults on sau…sau, and a different group of 30 children and 23 adults on fie…fie. Experiment 2 (Alternatives) was conducted with 86 children (mean age 5;04) and 83 adults. More specifically, we tested 32 children and 23 adults on marked sau, a different group of 27 children and 30 adults on sau…sau, and a different group of 27 children and 30 adults on fie…fie. Experiment 3 (Alternatives & QUD) was conducted with 86 children (mean age 5;06) and 89 adults. More specifically, we tested 27 children and 35 adults on marked sau, 27 children and 26 adults on sau…sau, and a different group of 32 children and 28 adults on fie…fie. Footnote ⁶

4.3. Methodology and materials

Building on the materials and design of Tieu et al. (Reference Tieu, Yatsushiro, Cremers, Romoli, Sauerland and Chemla2017), we ran three modified TVJTs presented in Prediction Mode rather than Description Mode (Singh et al., Reference Singh, Wexler, Astle-Rahim, Kamawar and Fox2016). Such a task licenses ignorance inferences, which often characterise disjunctive statements. Participants were introduced to a puppet, whose statements were pre-recorded. The task proceeded in three steps:

1. (i) For each story, Bibi made a guess about what would happen.

2. (ii) Participants then saw the outcome.

3. (iii) They then had to say whether Bibi had guessed well.

As detailed previously, we employed multiple disjunctions: marked sau (which is felicitous in utterances that represent answers to questions, unlike neutral sau), sau…sau, and fie…fie. These disjunctions were tested in a between-subjects design, each in a different variant of the experiment, in order to ensure that participants’ interpretation of disjunction would not be influenced by being exposed to the other disjunction types. In short, no participant completed more than one experiment, and no participant saw more than one disjunction type.

In Experiment 1 (Baseline), participants started with two practice trials, where the puppet made two guesses (one good and one bad). The test phase consisted of 13 experimental trials, in which participants heard disjunctive statements such as The hen pushed the bus or the airplane, presented in two kinds of contexts: 1-disjunct-true (1DT) (×4), in which, for example, the hen pushed only the bus, and 2-disjunct-true (2DT) (×4), in which, for example, the hen pushed both the bus and the airplane. We also included a false control condition, 0-disjunct-true (0DT) (×2), in which the hen pushed neither object. Participants also heard three fillers that contained no disjunction. Table 2 provides examples of experimental items for the 1DT condition, where the character acted only upon one object.

Table 2. Example experimental item with marked sau for the 1-disjunct-true (1DT) condition in Experiment 1

Importantly, to address the potential objection that children’s conjunctive interpretations are an experimental artefact related to the number of objects in the context, we introduced additional objects in the background so that four objects were present, even though the test sentences mentioned only two.

In Experiment 2 (Alternatives), participants heard disjunctive statements after hearing unrelated conjunctive statements. For example, participants might hear a conjunctive statement referring to the actions of a deer in one story, and then hear a disjunctive statement in the subsequent story referring to the actions of a hen.

Experiment 2 was inspired by Foppolo et al. (Reference Foppolo, Guasti and Chierchia2012) and Skordos and Papafragou (Reference Skordos and Papafragou2016). It contained 26 sentences: 2 warm-up sentences and 24 test sentences presented in four pseudo-randomised blocks (4 × 6 sentences) where disjunctive statements were always preceded by conjunctive statements presented in 2-conjunct-true (2CT) and 1-conjunct-true (1CT) contexts. A sentence block thus consisted of a true 2CT conjunctive statement, a false 1CT conjunctive statement, a 1DT disjunctive statement, a 2DT disjunctive statement, a 0DT disjunctive statement, and a true/false filler. Table 3 provides examples of conjunctive items employed in Experiment 2. The disjunctive utterances were identical to those in Experiment 1 (see Table 2 for an example of an item in the 1DT condition).

Table 3. Example experimental items with marked sau for the 2-conjunct-true (2CT) and 1-conjunct-true (1CT) conditions in Experiment 2

Experiment 3 (Alternatives & QUD) differed from Experiment 1 (Baseline) in that the disjunctive statement represented an answer to a conjunctive question, as illustrated in (25). The question was presented with a natural rising intonation.

While all experiments involved a story with a potential implicit QUD (e.g., What objects did the hen push?), Experiment 3 (Alternatives & QUD) employed an explicit question that made the stronger conjunctive alternative available and contextually relevant. Otherwise, the experiment employed the same design and materials as Experiment 1 (see Table 2).

4.4. Data analysis

The data and scripts for the statistical analyses below are available on OSF.Footnote ⁷

Only participants who displayed above 50% accuracy on the fillers and 0DT controls were included in the analyses. This criterion led to the exclusion of seven child participants and five adult participants. For Experiment 2, we also assessed participants’ accuracy on the conjunctive statements. Children’s accuracy on the conjunctive statements was 91.5% (86.9% for 1CT and 96.1% for 2CT), whereas adults’ accuracy on the conjunctive statements was 98.15% (97.5% for 1CT and 98.8% for 2CT).

Figure 2 displays the percentage of yes-responses to the 1DT and 2DT conditions, across groups, experiments, and disjunction types.

Figure 2. Percentage of yes responses from children and adults to 1DT and 2DT conditions, across disjunction types and experiments.

The 1DT and 2DT responses were used to categorise each participant for the planned analyses, as illustrated in Figure 3 and explained below:

• • Participants who accepted more than 50% each of both 1DT and 2DT items were categorised as inclusive.

• • Participants who accepted more than 50% of 1DT items but less than 50% of 2DT items were categorised as exclusive.

• • Participants who accepted more than 50% of 2DT items but less than 50% of 1DT items were categorised as conjunctive.

• • Participants who accepted less than 50% of both 1DT and 2DT items were categorised as contradictory (since this behaviour amounts to treating disjunctions as contradictions).Footnote ⁸

• • Finally, participants who accepted exactly 50% of items in either 1DT or 2DT condition were labelled as mixed.

Figure 3. Categorisation of participants in the sau sau Baseline task, to illustrate how participants were classified based on their responses to 1DT (x-axis) and 2DT (y-axis) trials. In this case, 18 children were categorised as inclusive, 4 children were categorised as conjunctive, 5 as mixed, and 1 as exclusive. 3 adults were inclusive, 2 were mixed, and the remaining 22 were exclusive.

Statistical analyses were carried out on the counts of participants falling in each category for the different factor combinations. We decided against analysing raw responses with a mixed-effects logistic regression. The main reason is that most implementations of mixed models assume a cantered normal distribution for the random effects (even for logistic regression, the random effects are assumed to be normally distributed on the log-odds scale). This assumption goes directly against the observation that most participants consistently adopt one or another interpretation for ambiguous target sentences. Indeed, if the latter is correct, the random effects would follow a multimodal distribution, not a Gaussian one. After discussion with the reviewers and editors, we did fit a logistic regression with maximal random effect structure, and while the model converged, the by-participants random effects did not follow a cantered multivariate Gaussian distribution. The estimated fixed effects from this model were therefore uninterpretable. The counts of the “mixed” participants were not analysed, since they do not constitute a consistent category (but as explained below, they were counted when determining the total number of participants per task). We used the “contradictory” participants as the reference level (in this way, any category that is significantly higher than the reference level can be interpreted as a reading of disjunction that is significantly more available than chance errors).

The counts were analysed with a Poisson model using log-link and the total number of participants who completed a given task with a given disjunction as the offset. The offset in a Poisson model is a correction for the fact that the total number of participants may differ between conditions. Mixed participants contributed to the offset, even though the model was fitted on data excluding these participants, as the estimated rates for the other categories would be overestimated otherwise. Effects were tested by comparing models with and without the predictor of interest, using likelihood ratio tests.Footnote ⁹ To summarise, our dependent variable will be the count of participants, and, unless otherwise specified, our predictors are Category (four levels), Disjunction type (three levels), and Task (three levels). Although Group (two levels) is an important factor, prior research has shown that adults and children interpret disjunctions very differently. Because of this, and given the already high number of predictors, we decided to analyse the child and adult data separately rather than include Group as a predictor.

Finally, we corrected for multiple comparisons (since both manipulations have a chance to lead to an increase in implicatures with children). Since we are not interested in main effects but rather interactions with Category, we applied a conservative Bonferroni correction with $ m=6 $ (2 degrees of freedom from Task  $ \times $  3 degrees of freedom from Category). Where relevant, corrected p-values are given as $ {p}^{\prime } $ .

4.5. Results

Table 4 gives the full distribution of participants across the different categories, for each combination of task and disjunction type. Figure 4 summarises this data in a graph.

Table 4. Count of participants in each category by task and disjunction type

Figure 4. Distribution of participants across categories (excluding the Mixed category) by group, task, and disjunction type.

Children’s results: We found no triple interaction among Disjunction, Category, and Task ( $ {\chi}^2(12)=14,p=0.3,{p}^{\prime }=1 $ ), so we dropped it from the model for subsequent tests. We found no significant difference between the Baseline and Alternatives tasks ( $ {\chi}^2(3)=2.1,p=0.55,{p}^{\prime }=1 $ ), but a clear difference between the Baseline and Alt & QUD tasks ( $ {\chi}^2(3)=41.6,p=4.8\times {10}^{-9},{p}^{\prime }=2.9\times {10}^{-8} $ ).

These tests looked for any difference in the distribution of child participants among categories. As a post hoc analysis, we looked at which categories are responsible for the difference between the Baseline and Alt & QUD tasks using the emmeans package. The marginal means showed that the significant interaction is driven by a combination of lower rates of inclusive ( $ 0.64 $ to $ 0.47 $ ) and conjunctive readings ( $ 0.20 $ to $ 0.06 $ ) and a higher rate of exclusive readings ( $ 0.02 $ to $ 0.32 $ ) in the Alt & QUD task, none of which is significant on its own.

We also found a significant interaction between Category and Disjunction types ( $ {\chi}^2(6)=58.7,p=8.1\times {10}^{-11} $ ). A post hoc inspection of marginal means showed a very small difference between marked sau and sau...sau; the latter seems to elicit a bit more conjunctive readings. Importantly, fie...fie had lower rates of inclusive participants, no exclusive participants, and higher rates of conjunctive participants than all sau-disjunctions.

Since we did not find any evidence of a triple interaction, there is no point in further exploring how the different readings vary across tasks for each disjunction separately.

Adults’ results: The triple interaction was not significant for adults either ( $ {\chi}^2(12)=17.4,p=0.13,{p}^{\prime }=0.80 $ ). This time we found no significant differences between either the Alternatives or the Alt & QUD task and the Baseline ( $ {\chi}^2(3)=2.31,p=0.51,{p}^{\prime }=1 $ and $ {\chi}^2(3)=2.65,p=0.45,{p}^{\prime }=1 $ , respectively).

However, we did find a significant interaction between Disjunction and Category ( $ {\chi}^2(6)=17.8,p=0.0066 $ ). For adults, the marginal means paint a very different picture. Unlike with children, where the interaction was driven by a categorically different behaviour of the fie...fie disjunction, with adults we observe small quantitative differences between the three types of disjunction, with fie…fie positioned right in the middle between marked sau and sau...sau. In particular, the estimated marginal rates of exclusive interpretations are 0.56 for marked sau, 0.73 for fie...fie, and 0.85 for sau...sau.