More hard words: Learning emotion and mental state adjectives from linguistic context

ABSTRACT

How do young children learn the meanings of adjectives that label emotions and mental states, like happy or confident? The concepts behind these words seem easy to grasp, and yet the properties they denote are abstract and lack reliable visual correlates, much as with verbs such as think or know. There is robust evidence that the linguistic context in which mental state verbs appear supports their acquisition. Using a two-pronged approach, we explore whether the same is true for these adjectives. First, we present a comprehensive study of adjectives in child-directed speech (CDS) in CHILDES corpora, revealing that these adjectives have a unique distributional profile based on the semantics of co-occurring words, and the syntactic complements with which they appear, distinguishing them from other adjective types. Second, we present a word learning study with adults using a version of the Human Simulation Paradigm manipulating key variables from the corpus study (subject animacy and syntactic complements), while holding other features (morphosyntactic form and syntactic position of the adjective) constant. Participants’ guesses demonstrate that they recruit these cues to constrain the meaning of novel adjectives to one corresponding to an emotion or mental state. By presenting a thorough comparison of adjective types across key linguistic contexts and providing evidence that such contexts constrain adjective meaning, we systematically extend syntactic bootstrapping to the adjectival domain in a way that goes well beyond previous work, and paints a vibrant picture of what supports the acquisition of adjectives.

Acknowledgements

We are grateful to all of the members of the research team in the Rutgers Laboratory for Developmental Language Studies for their important and amazing foundational and sustained efforts with the corpus data, and their ability to locate and share comedic gems to maintain our spirits over time. Some of these research assistants also honed their acting skills for our videos. Feedback from members of the UNC Language Development Lab and audiences at the UMass Jabberwocky Words in Linguistics Workshop, UC-Irvine, Yale, and UConn was incredibly helpful in shaping our thinking about and presentation of this work. We thank Meg Gotowski for offering helpful comments during initial stages of our experiment brainstorming, and Kyle Parrish and Chris Wiesen for assistance with statistical analyses for the word learning experiment. Finally, we are grateful to Aaron Steven White and Jeff Lidz for their especially insightful comments on earlier versions of this manuscript. All errors, of course, remain our own.

Disclosure statement

No potential conflicts of interest are reported by the authors(s).

Declaration of Interests

The authors report there are no relevant competing financial or non-financial interests to declare.

Supplemental material

Supplementary materials can be found online at: https://osf.io/s2mrf/?view_only=1343bed6828742b78c05aff6765f134e

Data availability

The authors will share source data upon request.

Notes

¹ For example, animate objects trigger differential object marking in several languages, including Spanish (Aissen 2003), and a number of Australian languages use a special Accusative marker for objects that are animate or human-denoting (Blake 1977). In addition, Navajo, Chamorro, Jacaltec, and Japanese either prohibit or disprefer inanimate subjects, or subjects that are ‘less animate’ than the object (Chung 1983, Craig 1976, Hale 1972, Kuno 1973; cited in Becker 2014).

² Note that not all adjectives (or APs) or verbs (or VPs) denote properties of individuals. For example, raising predicates like seem and tough adjectives, both of which take an expletive, non-referential subject, but have very different meanings, certainly do not. This point is, in a way, tightly related to our discussion of animacy earlier, appearance with an inanimate subject signals that the adjective does not obligatorily denote a property of an individual, whereas consistent appearance of an animate subject, and no inanimate or expletive one, is a fairly reliable cue that this is the case. We thank our reviewers for gently pushing us on this point.

³ We exclude a handful of occurrences of remember about and like about, which were in wh-questions in which there would have been a gap or trace between the verb and about (i.e., {what do you remember/what did you like} t about X).

⁴ Our approach classifies a given adjective as labeling an emotion or mental state in part based on an intuitive definition of emotion words as labeling internal affective states, and we recognize that there may be disagreement over certain boundary cases (e.g., is interested an emotion or a non-emotion mental state?). However, in the absence of established truth-conditional diagnostics for discriminating emotion words from non-emotion words, we adopt this approach here and work toward outlining clear inferential diagnostics in the near future.

⁵ To be clear, these differences among adjective types, which are driven by conceptual and distributional differences, track deeper semantic differences. While this is a topic to be explored in more depth in a separate line of research, we note here specifically that for emotion and mental state adjectives, there are further distinctions that may or may not tightly parallel verbs as embedding predicates. Hacquard & Lidz (2019) divide attitude verbs into representationals (which take clauses that have the hallmarks of declarative clauses, such as finite clauses in English, e.g., know) and preferentials (which take clauses that have the hallmarks of imperatives, such as non-finite clauses, e.g., want). Some mental state adjectives (e.g., aware, confident) pattern with the former, and others with the latter (e.g., interested, reluctant, willing). However, some emotion/mental state adjectives appear with both (e.g., amazed, excited, happy, relieved, worried), and these predicates fall into the category of veridical preferential predicates, in that they are responsive, and when accompanied by about, can take an interrogative who/which complement. When they take a finite clause complement expressing a proposition p in English, they not only presuppose the truth of p, but also express that the subject believes that p, and that the degree to which the subject prefers that p at world w is greater than a standard threshold of a set of focus alternatives (see, e.g., Uegaki & Sudo 2019). We would like to propose that the commonality among all emotion/mental state adjectives, regardless of further divisions within these categories, is that they all express a relationship or stance between an experiencer (and therefore obligatorily animate) subject, and an eventuality expressed by the complement—and further, that for emotion adjectives in particular, the content of the complement can be taken as the cause of the affective state of the subject.

⁶ We use postnominal in the same way as Blackwell (2005). In their analysis of adjectives in child-directed speech in British English, including free play, book texts, and shared book reading, Davies et al. (2020) used the terms prenominal as we do for attributive uses, postnominal for predicative uses, postpositive for adjectives immediately following a noun, and isolated for adjectives appearing alone.

⁷ We include if clauses in this table to be thorough and accurately represent syntactic structures immediately following these adjective types, although we clarify in the following that not all of these if clause instances should be treated as the same sort of syntactic argument as the other complements. We therefore do not include these clauses in our experiment. We also leave aside the difference in interrogative/declarative status between if and other finite clause complements here.

⁸ We are grateful to a reviewer for highlighting this connection.

⁹ The extent to which this correspondence between syntax and semantics for emotion adjectives is universal, so as to be a reasonable expectation that any learner holds about whatever language they are presented with, must await future crosslinguistic comparison.