Lexical cue and perceptual clue to the interpretation of animal-monster hybrids: Classifier selection as a probing stimulant

Abstract

We explored how label cue and body structure impacted the mental construal of animal-monster hybrids through the choice of Chinese classifiers. Taking Chinese classifiers zhi and ge as the probe, we manipulated label cue and perceptual input from monster structure and detected how these two factors influenced the interpretation of the hybrids. Our rationale is that if the choice of zhi is indicative of animal analogousness, and, conversely, that of ge is indexical of non-animal likeness, the difference of effect between using zhi and ge for monstrous animal hybrids should reflect the variation in their animacy interpretation. We found that the effect difference between zhi and ge across the higher- and lower-animacy monsters was remarkable in the unlabeled situation but, on the contrary, was attenuated in the labeled situation. In a word, linguistic cuing lumped the monster figures into the same category and biased them toward non-animate construal.

Introduction

Does language affect cognition and perception and enable language users to access and manipulate mental representations? This question, involving understanding the core of human cognition (Carruthers, 2002), has long intrigued philosophers (Lee, 1996). Many scholars (Clark, 1998; James, 1890; Vygotsky, 1962; Whorf, 1956) have conjectured about the transformative power of language on cognition. Pursuing the same avenue of research, we set out to explore the effect of label information on the mental categorization of legendary monstrous beasts through the choice of Chinese classifiers. The legendary creatures discussed herewith are visual hybrids consisting of genuine animals and novel monsters. They are categorically ambiguous and indeterminate; what's more, they are not linguistically represented by specific Chinese labels. With these characteristics, they are suited for exploring the role of verbal information in perceptual categorization since they are amenable to the experimental manipulation on label provision and monstrosity.

A hybrid, according to Shen and Gil (2017), is a creature created through the merging of components from two or more separate entities, known as the hybrid's parents. The hybrid itself is a fresh and unfamiliar creature whose categorical membership is not transparent, while its parents are often well-known creatures belonging to recognizable classes. Prototypical and well-known examples of visual hybrids are Centaurs, part-human part-horse, and Mermaids, combining the half of a woman with the half of a fish. Visual hybrids are a fascinating border case of category ambiguity that must be addressed by any categorization theory. On the one hand, the hybrid's parents are well-known individuals; in this respect, the hybrid is far from being a strange, meaningless figure. The hybrid does not reflect any known idea in its entirety and, at least at first, lacks conventional lexical representation. As Shen and Gil (2017) noted, in the attempt to conceive or verbally depict these visual hybrids, the speaker is put under linguistic and mental strain (Gentner & France, 1988). Due to such characteristics, examining the conceptual and linguistic processing of visual hybrids under such strain can help us reveal underlying cognitive processes that aren't easily visible in observing more conventional objects, or, alternatively, completely novel ones.

Studies have been administered to investigate the conceptual hierarchy of humans−animals−plants−non-animate objects by using novel hybrids. For instance, Mashal, Shen, Jospe, and Gil (2014) and Shen and Gil (2017) addressed whether a hybrid's parent is more central to its categorical conceptualization. More specifically, they probed the tendency for hybrids to be categorized conforming to the parent that is higher on the Ontological Hierarchy (e.g., man-lion conceptualized as a kind of man rather than as a kind of lion). These studies provided evidence that the Ontological Hierarchy plays a role in the conceptualization of hybrids. Given the studies exploring hybrids encoding fused hierarchical features, we wonder whether the provision of lexico-semantic information takes effect on hybrid conceptualization as well. In the meantime, since classifiers encode perceptual categorization in the Chinese language system, the interconnection of lexical cue, novel animal hybrids, and Chinese classifiers deserves significant attention in categorization research.

The mapping of linguistic meaning and non-verbal conceptual knowledge has long been under discussion. Gleitman and Papafragou (2005) pinpointed that conceptual structure is not isomorphic to linguistic representation. In their view, words are often semantically broad, so they are not able to reflect the differentiation presented in the conceptual structure. For instance, the English word uncle does not semantically specify whether the referent is a relative by blood or marriage. That is, linguistic system is not as rich and complex as conceptual construct. In a word, there does not seem to be a direct or corresponding relation between language and thought.

Another issue that relates to our discussion is how language affects non-verbal cognition. Roberson and Davidoff (2000) and Lupyan (2009) demonstrated that the effect of words on conceptual representations and perception is reduced when verbal interference is introduced. Lupyan (2008) revealed that memory for exemplars was impoverished if category labels were provided during encoding. Regier and Kay (2009) pointed out that language impacts the perception of color: there is indeed an impact of color language on color categorical discrimination. In other words, people are less able to differentiate colors when they do not have separate words for them. Additionally, Dolscheid, Shayan, Majid, and Casasanto (2013) observed that language influences the recognition of musical pitch. Still, Borghi and Binkofski (2014) found that words may be particularly helpful in learning abstract concepts whose members are typically diverse, which means they are used in more differential contexts. For example, the concrete word cabbage tends to occur in contexts relating to cooking and eating and thus characterizes relatively low semantic diversity while the abstract word life can appear in a larger number of different contexts and thus features higher semantic diversity. Linguistic input can therefore facilitate the acquisition of abstract words. Lastly, Edmiston and Lupyan (2015) discovered that hearing or saying an object's name facilitated recognition of the object. In sum, language has been proposed to organize or shape non-linguistic conceptual representations, including colors, music, or objects.

Languages have developed a broad variety of nominal classification systems such as count-mass grammar, gender marking grammar, and classifier grammar systems (Aikhenvald, 2006; Senft, 2000). Chinese is a classifier language where classifiers tend to combine with a certain class of entities (Pinker, 1999). The influence of classifiers, critical components of the Chinese language system, on cognitive activities is well documented. Saalbach and Imai (2007) examined the influence of classifiers with experiments that required different cognitive processes, including categorization, similarity judgment, property induction, and word–picture matching, which revealed how strong the impact of classifiers was, and how they interacted with distinct cognitive capacities. In their research, an impact of the same-classifier relation was found in the similarity judgment task, in which participants were allowed to evaluate similarity between the target and the test items, and in the inductive reasoning task, in which participants were required to decide whether a novel property could apply to same-classifier items. In a word, they found a same-classifier effect such that when the target and test items shared the same classifier, they were judged more similar to each other and more likely to share a novel property. In addition, Chen, Bowerman, Huettig, and Majid (2010) investigated the impact of Chinese classifiers on overt attention in eye-tracking experiments, probing what participants looked at while listening to a pre-specified target noun. In their finding, when participants heard a classifier, they altered attention more often to classifier-matching objects than to distractor objects, whereas when the classifier was not explicated in speech, the effect being discussed was absent. In short, Saalbach and Imai (2007) and Chen et al. (2010) illuminated that the impact of classifier categories interacts with differential and specific cognitive constraints.

While the idea that language impacts how we categorize and perceive objects finds some empirical support, the position that nonverbal cognition is unaffected by linguistic knowledge also has many supporters (Gleitman & Papafragou, 2005; Li & Gleitman, 2002; Snedeker & Gleitman, 2004). Some researchers have attributed cross-linguistic differences in arithmetic performance (Pica, Lemer, Izard, & Dehaene, 2004) as well as spatial cognition (Levinson, Kita, Haun, & Rasch, 2002) to disparities in cultural emphasis and experience rather than differences in language (Gleitman & Papafragou, 2005; Li & Gleitman, 2002). Furthermore, Papafragou, Massey, and Gleitman (2006) showed that, in regard to the differential encoding of path and manner, which has been taken as a prime example of the effects of language encoding on thought, linguistic encoding does not faithfully represent nor forcibly affect the mental portrayal of events. Although language-thought relation has been well scrutinized, scant attention has been paid to the role of classifiers as a reflector of the linguistic effect on non-verbal cognition. The purpose of this paper, therefore, is to determine how linguistic information and perceptual prompts play their parts in object interpretation through the observation of classifier choices. The theoretical thrust and rationale behind our research are concisely stated as follows.

As the focus of our attention, monster characters are featured with their indeterminate status: they are made-up creatures, which we have never seen or encountered and can be interpreted on a par with animals or as a totally odd entity. With this feature, they are suitable for the study on the linguistic influence on object representation in the sense that they are characterized with diverse bodily structures. Their structural variation serves as a stimulator of perceptual input for interpretation as more or less animal-like. In addition, what makes monster figures more eligible for this line of research is that, as hybrid animal beings, they are not identified by specific Chinese names. The most appropriate name for them is the umbrella term guaishou, which is literally glossed as ‘weird beast’. This generic term provides the cue that, although these monsters are morphing animals, they are not genuine animal species. With the cue of label, we can detect the competition effect between perceived body structure of monster figures and linguistic interference in the mental construal of categorically ambiguous entities.

Chinese speakers always have to choose a classifier or a measure word when they speak of a number or an amount of something. Consider the examples in (1) as follows

As in (1a), the classifier (abbreviated as CL afterwards) ben is selected for the noun shu ‘book’. In (1b), the measure word (abbreviated as MW hereafter) qun denotes a pack of animals. While Chinese measure words are used to create a unit of quantification, Chinese classifiers tend to combine with a certain class of entities (Pinker, 1999).

Related to our discussion are Chinese specific classifier zhi and default classifier ge. Many speakers use the classifier zhi within the category of animals. Zhi is employed with animal words in wide and general context. In the account of Myers (2000), zhi is more robust than other specific classifiers to combine with animal words. Witness the examples in (2):

For Myers, the classifier zhi is generally used for animals for it has undergone an analogical inference. The reason why zhi is a good candidate for analogical extension is due to its constant semantics and highly frequent appearances with animal terms. According to Myers (2000), unlike specific classifiers, the use of the classifier ge is not controlled by analogy. It is selected by a default rule: it is used when the speaker is not able to draw on memory or an analogy with something in memory. In other words, it kicks in as the physical properties of the object do not fit with any classifier. Let us consider the examples in (3):

If ge has a complement function for classifier selection, then it is expected that speakers are likely to choose ge for nouns without analogical exemplars. This has been confirmed by corpus data. Myers (2000) showed that ge is frequently used for atypical nouns. As in (3a) and (3b), ge goes well with abstract nouns tiyan ‘experience’ and xiwang ‘wish’. The fact that nouns like tiyan do not have exemplars for analogical inference indicates that ge is used by default when no specific classifiers are available.

Li and Lin (2017), who probed the conceptual profiling of monstrous hybrid creatures through Chinese classifier selection, lent support to the discrepancy between linguistic and conceptual means to entity profiling. First, in a corpus study, they found that the noun guaishou with the second character shou glossed as ‘beast/animal’ extensively co-occurs with Chinese animal classifier zhi, which is often used with animal words (Myers, 2000). By contrast, the noun guaiwu with the second character wu interpreted as ‘object/thing’ expansively collocates with Chinese default classifier ge, which is used where no analogical exemplars are found (Myers, 2000). Second, in a picture-phrase matching study, Li and Lin demonstrated that the monster figures with an increased number of body parts and thus endowed with a weaker animal-like property are less felicitous to match the classifier zhi than those without. By contrast, the same characterization did not invite different rates of acceptability for default classifier ge. Taken together, the two studies showed that the (non-)/animacy distinction linguistically represented in nouns guaishou and guaiwu is encoded by the compatibility with zhi versus ge. On the contrary, the (non-)/animal-like property conceptually manifested in the monster features is codified by the gradient compatibility with the classifier zhi alone. Li and Lin provided evidence that classifier behaviors indexical of the analogical proximity of monster characters with real-world animals diverge at the linguistic and conceptual realms. While Li and Lin established that the behaviors of Chinese classifiers pointed to the linguistic/conceptual division in monster construal, it is also tempting to explore the influence of linguistic interference on the conceptual profiling of objects.

For the purpose just stated, we decided to delve into the effect of labelling on the mental construal of categorically indeterminate animal-monster hybrids through the choice of Chinese classifiers. Taking Chinese classifiers zhi and ge as the indicator for animacy, we experimented on label prompting as well as bodily structure and detected how these two factors impacted the animacy interpretation of monstrous hybrids. Our rationale is that if the choice of zhi is indicative of animal analogousness, and, conversely, that of ge is indexical of non-animal likeness (Myers, 2000), the difference of effect between using zhi and ge for monstrous animal hybrids should reflect the variation in their animacy interpretation. As such, with this line of research, we hope to contribute to the role of lexical intervention in object interpretation.

Along our thread of research, we addressed the question of whether the cue of the category label guaishou affects the choices of Chinese classifiers zhi and ge, which indicate animacy at both linguistic and conceptual levels, for the hybrid monster creatures. In this research work, we devised a norming study and a phrase-picture verification study. In the norming study, we carried out an Animacy Rating Task to measure the animacy property of the target monster images. Once the animacy statuses of the monster figures were identified, we executed a Phrase-Picture Verification Task to detect whether the provision of the label guaishou impacted the acceptability of using zhi and ge with the monstrous creatures featuring different levels of animacy characteristics. Through our studies, we wished to assess whether and how labeling and perceived animacy interacted in the interpretation of categorically ambiguous and indeterminate objects.

Three theories are presented here in conjunction with formulated hypotheses and predictions. The three theories are Label Feedback Theory, Thinking-for-Speaking Theory, and Category-Based Inference Theory. They offer different views on how linguistic input exercises influence on human conception and perception and thus provide a basis for our discussion.

On the basis of the studies showing an effect of language on cognition, Lupyan (2012) advanced the label-feedback hypothesis, providing a theoretical account of the cognitive mechanisms through which words can influence conceptual representations and perception. According to this hypothesis, linguistic labels deliver top-down feedback to the levels of conceptual representations that further affect perception. In the view of Lupyan (2012), category labels become more strongly associated with properties that are typical or diagnostic of the category and become dissociated with those that are idiosyncratic features of specific category members, because some properties are more highly correlated with category membership than others (e.g., “having wheels” versus “being black” for cars). This top-down activation of the most diagnostic perceptual features of the category from the verbal label produces a transient “perceptual distorting” in which category members that share those features are drawn closer together and non-members are pushed away, facilitating categorization. Top-down feedback from a verbal label to perceptual features is claimed to occur even during passive processing of these objects. In sum, the up-regulating the effect of labels through their overt presentation benefits typical category members more than atypical ones in visual tasks. Effects of labels on perceptual processing appear to be stronger for more typical exemplars. The atypical exemplars, although having the most potential to be affected by the label, interact with the label more weakly than the more central exemplars (Lupyan, 2012).

On the basis on the Label Feedback Theory (Lupyan, 2012), we hypothesized that the generic label guaishou should induce a typicality augmenting effect: the label cue should render the lower-animacy items more monster-like than their higher-animacy counterparts. The rationale on the basis of the Label Feedback Theory (Lupyan, 2012) is that given that lower-animacy items are more characteristic than the higher-animacy ones as monstrous creatures, they should be perceived as more anomalous and unanimous in the label-cued condition. Therefore, viewing that the choice of zhi indicates animal analogousness, and that for ge reflects non-animal likeness, we predicted that the lower-animacy items should produce a greater effect difference between the cues of zhi and ge than their higher-animacy counterparts in the labeled context as opposed to in the unlabeled context.

In the view of Slobin (1996), our thoughts are transformed into lexical categories and grammatical structures in the use of language, thus changing preverbal conceptualization. That is, “thinking” adopts different forms depending on the verbal or visual mode that is being employed. In the view of “thinking for speaking”, the particular speech events that a language speaker experiences provide cues to non-linguistic categorization: a new linguistic label induces a language user to pay attention to certain types of classification criteria. In this line of thinking, speakers of different languages are predisposed to attend to certain aspects of experience due to obligatory categories in grammar. He observed that in a satellite-framed language like English, which indicates the manner information of the movement in the main verb, speakers use manner verbs more frequently when describing events in story descriptions. More specifically, English speakers have more ready access to manner and utilize this dimension more extensively both communicatively and cognitively. On the contrary, such a tendency does not appear in a manner-indifferent verb-frame language like French, in which manner is an adjunctive element to an already completed clause and is indicated when it is needed. Put simply, it is possible to talk about movement patterns in all these languages, but this dimension is a more regular part of thinking for people speaking satellite-framed languages.

It is worth noticing that “thinking for speaking” effects are hypothesized to occur prior to the moment of speaking—thoughts structure themselves according to the structure of the medium language. Empirical evidence has been provided that the “thinking for speaking” effect is a transient phenomenon, not extending beyond speech time (e.g., Papafragou, Hulbert, & Trueswell, 2008). In other words, “thinking for speaking” is distinguished from “habitual thinking”, the essence of the stronger Whorfian hypothesis (Whorf, 1956), which advocates that one's language determines the long-lasting habitual conceptualization of reality.

In accordance with the Thinking-For-Speaking Theory (Slobin, 1996, 2003), we hypothesized that the perception of monster figures should be affected by linguistic experience. According to Slobin (1996, 2003), mature speakers know by experience whether their language requires such categories or not, and select the appropriate information in building linguistic messages. In this vein of theory, the collocational patterns of the label guaishou are expected to play a role in the classifier choices for the monster characters. The reasoning behind this hypothesis is that since the noun phrase guaishou is notably more often used with zhi than ge (Li & Lin, 2017), these collocational patterns should direct Chinese speakers to attend to the association of animal-like monstrous creatures with the use of zhi. With zhi and ge as reflectors for (non-)/animal analogousness, we predicted that the choice of zhi should uniformly invite a higher acceptance rate in the labeled condition than in the unlabeled condition and, meanwhile, should incur a higher level of acceptability than that of ge within the labeled condition.

Gelman (2003) proposed that children view categorical membership as natural and inherent: knowing the category label helps children make important property inferences. Gelman advocated that category labels have an advantaged status relative to other information, such as non-label features. Specifically, category labels express category membership and inductive potential, allowing children to learn and use categories that extend beyond perceptually salient features. Young children appeal to inheritance and innate potential and display rather an elaborate belief that kinship overrides outward similarity.

Along the line of the Category-Based Inference theory, Gelman and Markman (1986) tested whether children generalize based on identical as well as semantically-similar labels, given that children treat labels as proxies for kinds. They tested 4-year-old children on property inductions with category information provided by either identical or synonymous labels. They found that children made inferences to categorically-related items at above chance level and that, more importantly, synonyms and identical labels gave rise to similar effects. Also, Gelman and Coley (1990) found that when 2-year-old children were shown images sharing category but not appearance similarity (e.g. a bluebird and a dodo, both labeled birds) and images sharing appearance but not categorical similarity (e.g. a bluebird and a pterodactyl, labeled bird and dinosaur), children made generalizations to objects with identical labels rather than appearances. These studies of category-based induction demonstrated that children reason that members of a category possess common underlying features, and language is a key source of information indicating category membership (Markman, 1989).

In the view of the Category-Based Theory, category labels have a privileged status in relation to other non-label information. This view led to the hypothesis that the label cue guaishou should cluster monster figures into the same class, overriding the impacts of their differential bodily structures. Considering that the choice of zhi indicates animal likeness, and that of ge mirrors non-animateness, we predicted that when no category label is provided, it is more likely for participants’ classifier choices to be associated with the perceived animacy as compared to when the category label of guaishou is provided (which is likely to override the perceptual cues).