Scalar implicature is not a default process: An ERP study of the scalar implicature processing under the effect of focus factor

Highlights

• This study examines the processing mechanism of SIs under the effect of the focus factor.

• The scalar term in the focus position is more favourable for some listeners to acquire SIs.

• Pragmatic responders evoke a larger N400 and sustain negativity during SI acquisition.

• Semantic responders do not evoke an ERP effect in the processing of the SI term.

• SI acquisition is not a default process.

Abstract

This study explored the processing mechanism of scalar implicatures under different focus conditions by the picture-sentence verification paradigm. Through the guidance of different types of incorrect sentences, scalar terms were controlled to adjust whether they were at the focus position and divided the focused and non-focused conditions. The behavioural results showed that when the content of the picture was “all squares have…”, more participants under the focused condition could judge that the underinformative sentence “some squares have…” was not appropriate. Moreover, according to whether participants could stably determine that the underinformative sentences were inappropriate, they were divided into pragmatic and semantic responders. The ERP results showed that pragmatic responders evoked a P200 effect, an N400 effect and a sustained negativity effect according to the scalar terms, and a P200 effect and a sustained positivity effect according to the sentence-final words. In contrast, semantic responders did not elicit any ERP effect. These results indicate that the generation of scalar implicatures is not completely determined by the scalar terms and that the focus factor plays an important role in the scalar implicatures inference.

Introduction

In many scenarios of verbal communication, a speaker’s utterance always conveys much more information than what is in its literal meaning. For example, the speaker B in example (1) expressed “he/she does not like all the pictures” by the word “some”.

• A: Do you like these pictures?

• B: I like some of them.

“Some” in example (1) means “not all”. This level of meaning is not the explicit literal meaning of the word, but it is the implication resulting from pragmatic enrichment. This type of nonliteral meaning is referred to as scalar implicature (SI) in Pragmatics. SIs can be derived on the basis of a variety of scales, such as <all, some>, <love, like>, <must, may>, <and, or>, etc. A characteristic of these scales is that their elements are ordered by entailment relations. Elements on the left side of the scale are informationally stronger, and therefore logically entail the weaker ones. The elements on the right side of the scale often imply a negative meaning of the strong words and are known as scalar terms or SI-trigger words. SIs provide researchers with a favourable observation window to explore the semantic-pragmatic interface mechanism.

The processing mechanism of the SIs inference can be summarized into two main viewpoints in theoretical linguistics research: the Default view (e.g., Horn, 2004, Horn et al., 2012, Levinson, 2000) and the Context-Driven view (e.g., Sperber and Wilson, 1986). The Default view holds that an SI is the salient, unmarked, and presumed meaning of the trigger word (e.g., Levinson, 2000). In the presence of SI-trigger words, listeners are able to rapidly acquire the preferred interpretations of the trigger words prior to completion of processing of the entire proposition. Therefore, the SIs process has the characteristics of being unconscious, automatic, effortless, and independent of context. The Context-Driven view argues that the activation of SIs is based on contextual requirements and that listeners will be driven to reason SIs only when the SIs are related to the context (Sperber and Wilson, 1986). Therefore, the SI inference is conscious and non-automatic, requires effort, and is dependent on the context.

In response to the debate on the processing mechanism of SIs in theoretical linguistics, experimentalists have carried out a large number of experimental studies. Most of the early studies have adopted the truth-value judgement paradigm (Noveck, 2001, Noveck and Posada, 2003, Bott and Noveck, 2004, Feeney et al., 2004, De Neys and Schaeken, 2007, Noveck and Reboul, 2008, Pijnacker et al., 2009). Participants were required to make true or false judgements on sentence propositions, especially on underinformative sentences. For example, consider the underinformative sentence “Some elephants are mammals”. When “some” uses the semantic interpretation, “at least some and possibly all elephants are mammals” is a true proposition; when “some” uses the pragmatic interpretation, “some but not all elephants are mammals” is a false proposition. Studies have found that when participants used the pragmatic meaning of “some” to make a sentence judgement, the response time significantly increased. Most researchers viewed that it reflects the need for more cognitive effort in the activation of SIs rather than effortless processing.

Recently, experimentalists started paying more attention to studies on SI inference-related influencing factors, including the following. (1) Contextual factor: For example, compared with the lower bound contexts (e.g., Mary was surprised to see John cleaning his apartment and she asked the reason why.) in which SIs and context correlation are low, when sentences with scalar terms (e.g., John intended to host some of his relatives.) appear in upper bound contexts (e.g., Mary asked John whether he intended to host all of his relatives in his tiny apartment.) in which SIs and context correlation are high, participants tend to obtain SIs (Breheny et al., 2006, Bergen and Grodner, 2012, Dupuy et al., 2016). (2) Syntactic factor: For example, down-entailing sentences, such as conditional sentences, have lower calculation rates of SIs (Hartshorne et al., 2015). (3) Factors related to scalar terms: For example, the availability of the scalemates (Van Tiel et al., 2016) and the alternatives (Degen and Tanenhaus, 2015, Rees and Bott, 2018) to SI-trigger words; the forms of SI-trigger words (Bennett and Goodman, 2018). (4) The listener’s factors, including pragmatic competence (Nieuwland et al., 2010, Zhao et al., 2015), working memory resources (De Neys and Schaeken, 2007, Dieussaert et al., 2011, Marty et al., 2013, Marty and Chemla, 2013), and background knowledge (Bergen and Grodner, 2012), etc. For example, compared with the group with low pragmatic competence, the group with high pragmatic competence exhibits faster acquisition of SIs (Nieuwland et al., 2010), and the SI process for the group with high pragmatic competence is likely to be automated (Zhao et al., 2015). Additionally, listeners are less likely to make SIs when they believe that the speaker is not motivated to be informative (Bonnefon et al., 2009) or is unlikely to know whether the more informative statement is true (Bergen and Grodner, 2012). The above study has positive promoting effects on refined research on the SI processing mechanism. However, few studies are focused on the effect of focus on SI processing.

During language communication, important information can be highlighted and emphasized through focus. There are diverse methods for highlighting focus information, including wh-word in question–answer pairs (e.g., what did Ronnie buy to cook?/who bought salmon to cook?), cleft structure (e.g., It was Ronnie who bought the salmon to cook./It was the salmon that Ronnie bought to cook.), pitch accent (Ronnie bought salmon to cook./Ronnie bought salmon to cook.), focalizer (e.g., Ronnie ONLY bought salmon to cook.), word order (e.g., Salmon, Ronnie bought to cook.), information contrast (e.g., Ronnie did not buy beef, she bought salmon to cook.), typography and layout (e.g., italics, capitals, font size, underlining and boxed text, etc. See the summarized table in Sanford et al., 2006, Wang et al., 2014). Many studies have demonstrated the important effect of focus factor on language comprehension. For example, in behavioural studies, when cleft structure (Bredart and Modolo, 1988) is used and anomalous words are capitalised or underlined (Bredart and Docquier, 1989) to be the focus information of the sentence, the probability of participants detecting semantic anomalies is significantly increased. When the anomalous words are not focus information, semantic illusion frequently occurs: anomalies and inconsistencies in sentences are often overlooked. Similar results were obtained in studies that employed the text change detection paradigm. In this paradigm, participants were continuously presented with a text twice. During the second trial, individual words in the text were changed. The participants were required to determine if any words were changed when the text was presented the second time. The results showed that the change in detection rate was significantly increased when the changed words were highlighted by the cleft structure (Sturt et al., 2004), a wh-word in the context (Ward and Sturt, 2007), or italicisation and stress (Sanford et al., 2006). Electrophysiological studies also proved that there are differences in the processing mechanism of focused constituent and non-focused constituent in sentences. An example is the study by Wang et al. (2009) who used wh-words in questions to manipulate whether anomalous words are located at the focus positions (e.g., What vegetables did mum buy for dinner?/Who bought vegetables for dinner? Mum bought eggplant/beef for dinner.). ERP results showed that a clear N400 effect was triggered when the anomalous words are located at focus positions in the answer, while the N400 effect was reduced when the anomalous words are located at non-focus positions in the answer. Li and Ren (2012) used accent to label focus and found that compared with semantically appropriate words, the N400 effect was invoked by semantically inappropriate words when accented but not when de-accented. These studies showed that processing of input words during language comprehension is non-uniform. Compared with non-focus information, more attentional resources are allocated to focus information for deep processing, such that the interpretation of the focus information is more elaborate, while non-focused constituents are often overlooked, and their processing is an incomplete semantic and pragmatic analysis (Sanford and Sturt, 2002, Sanford, 2002).

In SI studies, some researchers hold that focus factor plays an important role in SI inference. An example is in question–answer pairs in which SI is only produced when the SI-trigger word corresponds to the answer for the wh-question and is the focused constituent (Van Rooij and Schulz, 2004). Therefore, as shown in example sentence (2), the numeral “two” in the answer in B will not generate SI. This is because the “who” in the question determines that the word “John” in the answer is the focused constituent, and “two”, a non-focused constituent, does not generate SI.

• A: Who has two children?

• B: John has two children.

Currently, only 3 empirical studies examined the role of focus factor on SI inference. Zondervan, 2010, Degen and Goodman, 2014, Exp 1 & 2) employed behavioural research methods to design wh-questions in context to control whether the SI-trigger word is the focused constituent of the sentence. The experimental results showed that when the SI-trigger word was located at the focus of the sentence, it could help the participants calculate more SIs. However, the aforementioned study results do not directly present an online processing of SI under the effect of focus factor. Nieuwland et al. (2010) is another ERP study related to focus factor that aimed to examine the relationship between pragmatic abilities and the SI processing mechanism, but the design of experimental materials is associated with the focus factor. Example 3 cites the experimental materials of Nieuwland et al. (2010) in which the researchers hold that in the sentence “Some people have lungs,” (Exp. 1), the comma signals clausal wrap-up and the end of the quantifier scope, which brings the clause-final words “lungs” clearly into focus. In contrast, in “Some people have lungs that are diseased by viruses” (Exp. 2), the scope of the quantifier encompasses the whole relative clause construction (“lungs that are diseased by viruses”), and the focus of the utterance is not “lungs” but “diseased by viruses”. The experimental results showed that pragmatically skilled participants showed a larger N400 to underinformative statements, but this effect disappeared in experiment 2. The researchers suggested that this is because in experiment 2, the critical words were unfocused so that the local underinformativeness went unnoticed, i.e., relatively shallow processing for critical words, which causes the disappearance of the N400 effect.

• (Exp. 1) Some people have lungs/pets, which require good care.

• (Exp. 2) Some people have lungs/pets that are diseased by viruses.

In addition to Nieuwland et al. (2010), many experimental studies have employed the ERP technique to examine the SI processing mechanism (Noveck and Posada, 2003, Nieuwland et al., 2010, Hunt et al., 2013, Politzer-Ahles et al., 2013, Zhao et al., 2015, Hartshorne et al., 2015, Barbet and Thierry, 2018). The advantage of the ERP technique lies in its high temporal resolution that can facilitate the tracking of dynamic cognitive activities in the brain and reveal different cognitive processing stages. These studies adopted different paradigms (see, e.g., picture-sentence verification paradigm: Hunt et al., 2013, Politzer-Ahles et al., 2013; MMN paradigm: Zhao et al., 2015; Stroop paradigm: Barbet and Thierry, 2016, Barbet and Thierry, 2018; priming paradigm: Rees and Bott, 2018); targeted different variables (e.g., pragmatic abilities: Nieuwland et al., 2010, Zhao et al., 2015; syntactic factors: Hartshorne et al., 2015; conversational contexts: Holtgraves and Kraus, 2018); and considered ERPs elicited by different target words, such as the SI-trigger words (e.g., Politzer-Ahles et al., 2013), the centre of the scalar term (e.g., some of the steaks, see Hunt et al., 2013), the relative words to the scalar term (e.g., some…the rest, see Hartshorne et al., 2015), and the sentence-final words (see Nieuwland et al., 2010, Noveck and Posada, 2003). Therefore, it is difficult to clarify only one that is the most primitive course of SI processing.

However, we are still able to partly understand ERP responses that are associated with SI processing through these experiments: N400 may reflect the acquisition of scalar implication of trigger words (e.g., Nieuwland et al., 2010, Hunt et al., 2013); sustained negativity may reflect cancellation of inappropriate SI (e.g., Politzer-Ahles et al., 2013); sustained positivity may reflect processing of plausible words that have low correlation with SIs; and P300 may reflect revision processes on inappropriate SI in face-threatening scenarios (e.g., Holtgraves and Kraus, 2018), etc.

In summary, our study employed ERP technology to examine the SI processing mechanism under the effect of the focus factor. As many studies found that the judgement responses by different participants towards underinformative sentences are heterogeneous, our study also examined online processing of SI-trigger words under different interpretation strategies. The experimental materials were roughly as follows (refer to Section 5.2 for details). There were six squares on the pictures presented to the participants, with two dimensions of colour and character in all or part of the squares with the same attribute characteristics. The participants were asked to judge whether the sentences were appropriate according to the pictures. The judgement sentences were divided into correct sentences, incorrect sentences, felicitous-some sentences, and infelicitous-some sentences. Among them, the latter two types of judgement sentences both contained the Chinese scalar term “有的” (equivalent to “some” in English), and “infelicitous-some sentences” represented underinformative sentences, which were the target sentences of the experiment. Whereas “felicitous-some sentences” represented informative sentences, which were the control sentences. The experiment utilized the incorrect sentences to adjust whether the scalar terms were at the focus positions and divided the focused and non-focused conditions accordingly. The experimental design methods were as follows. (1) Operating the incorrect sentences refers to whether they are relevant to the quantitative errors. Under the focused condition, the incorrect sentences do not match the pictures in the quantity description, while under the non-focused condition, the incorrect sentences do not match the pictures in the colour description. As shown in Fig. 1, under the focused condition, an incorrect sentence was designed as “四个正方形带字母” (in English: four squares with letters). The sentence content did not match the three squares with letters appearing in the picture, forming an error in quantity expression. Under the non-focused condition, an incorrect sentence was designed as “四个正方形带红色” (in English: four squares with red color), which did not match the four squares with yellow color in the picture, forming an error in color expression. Under the focused and non-focused conditions, the attention of the participants is guided to a judgement of the number or colour expression by setting different types of incorrect sentences, thereby controlling whether the scalar term is the focus component. (2) During each block, ten incorrect sentences of the same type are presented consecutively before the presentation of other types of judgement sentences so that the participants form a stable attracting focus on the quantity or colour words. That is, under the focused condition, each block first presented ten sentences with quantity errors consecutively; under the non-focused condition, each block first presented ten sentences with color errors consecutively. (3) The number of incorrect sentences was twice that of other types of judgement sentences. Under the two experimental conditions, the total number of incorrect sentences was 100 and there were 50 judgement sentences of each of the other types. Through the appearance of a large probability of incorrect sentences, the attracting focus of the participants was strengthened.

Based on related SI and focus empirical study results, we predict the main results of this study as follows. When the SI-trigger word is a focused constituent, more attention is used by the participant and processing is deepened; therefore, the SI inference rate should be higher. Conversely, when the trigger word is not the focused constituent, less attention is given, and processing is shallow, which may cause the pragmatic meaning of the SI-trigger word to be easily overlooked. Therefore, in behavioural performance, more participants may obtain the scalar implicature under the focus condition, thereby employing pragmatic interpretations for judgement of underinformative sentences. In ERP responses, many experiments found that the pragmatic interpretation of trigger words requires more cognitive resources. Therefore, the amplitude of ERP components evoked by trigger words in pragmatic interpretation could be larger. In addition, the quantity expression of underinformative sentences under pragmatic interpretation is not consistent with the stimulation pictures, which increases the difficulty of integrating sentence-final words under such conditions and can similarly evoke ERP responses with high amplitude. Regarding evoked components, as past ERP studies differ greatly in experimental paradigms and analysed keywords, it is difficult to accurately predict the evoked components of SI-trigger words and sentence-final words in this study. However, according to ERP studies on language processing and related ERP studies on SI, the following major ERP components may be elicited in this study: (1) The N400 component, which has the typical right centroparietal distribution. N400 is, in principle, elicited by every content word. There is a consensus that N400 reflects the meaning processing of an incoming word (Kutas et al., 2006, Kutas and Federmeier, 2011). Regarding the study findings of SI processing, the acquisition of pragmatic meaning of scalar term will evoke N400 with a greater amplitude (Nieuwland et al., 2010, Hunt et al., 2013). (2) The sustained negativity (SN), which is reported to be located at the posterior brain region. Politzer-Ahles et al. (2013) used the picture-sentence verification paradigm to record sustained negative waves evoked by pragmatically inconsistent quantifiers. As the paradigm and keywords of that study are similar to this study, this component could be induced in this study. (3) The sustained positivity (SP), which is widely distributed. Hartshorne et al. (2015) reported that this component was triggered by the phrase “the rest” in conditional sentences, which has low correlation with scalar implicature but is plausible. In this study, sentence-final words in underinformative sentences have unexpected but plausible characteristics and thus SP could be elicited. Furthermore, it can be expected that the ERP effect predicted above might only be evoked in the ERP responses of pragmatic responders, while semantic responders only interpreted the trigger word “some” semantically, so there might not be significant difference in brain potential evoked by the target word in semantic responders, that is, there was no ERP effect.