Seeing and Believing: The Relationship between Perception and Mental Verbs in Acquisition

ABSTRACT

Perception verbs and mental verbs have significant overlap in their syntax and semantics; both reference mental representations when taking embedded clauses, as in I see that Maria was here and I think that Maria was here. Some have suggested that perception is more accessible for young children than mental states, raising the question of whether perception verbs could serve as a semantic model for the acquisition of mental verbs via their shared syntax. Since embedded clauses are key to referencing mental states for both verb classes, we examine the developmental trajectory of perception vs. mental verbs in these constructions and others. Using a sample of 5,884 child-produced utterances and 8,313 parent-produced utterances from the Brown and Gleason corpora of CHILDES, we analyze children’s production of perception and mental verbs in their syntactic frames, as well as that of their parents. We find that children begin to produce embedded frames for both perception and mental verbs around the same time, but produce embedded frames with mental verbs more often, especially as they get older, despite greater use of perception verbs overall. These patterns do not reflect parental input: parents produce both verb types with similar frequency and use embedded frames more often than their children. These findings suggest that perception verbs are unlikely to serve as a model for mental verbs, and instead that mental verbs and their regular occurrence with embedded frames may provide a model for perception verbs when the latter reference mental states. We propose a semantic updating account for children’s acquisition of perception verbs, arguing that children’s early knowledge of perception verbs may not include mental state representations as a component of their meaning, and that this may only develop later as children learn the propositional syntax that is shared by and regularly occurs with mental verbs.

Disclosure statement

No potential conflict of interest was reported by the authors.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Notes

¹ Perception verbs – especially verbs of vision – might seem like they map primarily to physical actions (e.g., look = point one’s eyes toward), but in fact, the semantics of verbs like see and hear typically require that the perceiver achieves a mental representation as a result of a perceptual experience (Landau & Gleitman, 1985; Viberg, 1983). This is perhaps best exemplified by the acceptability of statements like She looked right at it but didn’t see it.

² In the perception training condition, children were trained using the same perception task from Gopnik et al. (1994); children in the belief training condition were asked to report their own or another person’s false belief in two separate deceptive appearance tasks (e.g., a golf ball that turned out to be soap).

³ Viberg (1983) (as well as Rogers (1971), using different labels) describes two other classes of perception verbs in his typology, distinct from experience verbs. These other classes of perception verbs are less syntactically productive than experience verbs, and do not overlap much, if at all, with mental verbs in the complements they license. In particular, they do not take sentential complements. For that reason, the focus in this paper is on experience perception verbs. Activity perception verbs, such as look (at) and listen, take as their subject an agent that consciously controls some action or process related to perception.

1. John looked at the house.

2. Roberta listened to the conversation.

Copulative perception verbs, such as look and sound, take as their subject the entity that is the source of some perceptual property.

• (iii) The table looked red.

• (iv) The music sounded lovely.

⁴ The discussion here focuses primarily on de dicto attitude reports, rather than de re attitude reports, which can contain descriptions of individuals or events that are (extensionally) correct but that the attitude holder would not assent to. For example, consider a scenario in which Sally watches someone dressed in a bear costume maliciously knock a child’s ice cream cone out of their hand. Sally thinks the person in the bear costume is a jerk, but unbeknownst to her, that individual is her brother, who she believes is a kind person.

• (v) Sally thinks that the person in a bear costume is a jerk. (de dicto)

• (vi) Sally thinks that her brother is a jerk. (de re)

Since the descriptions the person in the bear costume and her brother pick out the same individual, (vi) is true on a de re reading, even though Sally would not agree with it. There are many discussions and analyses of the de re and de dicto distinction in the semantics literature (e.g., Burge, 1977; Keshet & Schwarz, 2019; Lewis, 1979; Quine, 1956; inter alia). However, both de dicto and de re attitude reports are epistemically non-neutral, and the data presented here do not require any unique treatment of the de re/de dicto distinction.

⁵ For example, perception verbs take small clause complements, but mental verbs do not.

• (vii) David saw Nancy leave.

• *David thought Nancy leave.

⁶ Embedded complements represent propositions whose truth-values must be evaluated relative to some possible world (Hintikka, 1962; Lewis, 1986; Montague, 1974); the world of evaluation is specified by the matrix clause under which the complement is embedded.

⁷ The truth of (6) and (7) also depends on the complement being true, since both see and know are factive verbs that presuppose the truth of their clausal complements.

⁸ We found no instances of activity or copulative perception verbs used with embedded complements in our data set.

⁹ Perception verbs were also coded as experience, activity, or copulative, following Viberg’s (1983) typology. For polysemous perception verbs like look, feel, taste, and smell this classification was determined by context and/or frame. This coding was not used in our analysis, however, beyond separating out experience perception verbs, as discussed in the Results section.

¹⁰ Verbs (especially mental verbs) that occurred without a complement (e.g., I don’t know) could be considered to have elided complements. However, attempting to recover a possible complement structure from discourse context would be quite difficult since it would not always be clear what the elided complement might have been. Moreover, it is still the case that the verb was produced without a complement in such instances. It can also be difficult to distinguish between potentially elided complements (e.g., I don’t know [how to tie my shoes]) and formulaic uses that may not carry true mental content or be intended to express full propositions (see Diessel & Tomasello, 2001).

¹¹ For example, an utterance transcribed as “look (.) Mommy”, where (.) indicates a possible sentence/utterance boundary.

¹² For example, an utterance such as “watch going” when Adam and his parents were discussing wristwatches.

¹³ In the analyses of Higginbotham (1983), Moulton (2009), and Davis (2016), small clause complements are treated as non-embedded complements. From a strictly syntactic point of view, small clauses are a type of embedded clause; however, it can be argued that semantically they function quite differently from sentential complements since they do not carry propositional content (Higginbotham, 1983; inter alia). For this reason, they can be treated as similar to noun phrase complements (Parsons, 1990). See Davis (2016; and references therein) for further discussion.

¹⁴ We conducted separate analyses on children’s verb type and frame production using age and MLU as predictors in our models. MLU showed the same trends and significance levels as age as a predictor of verb type and frame production.

¹⁵ Since feel, taste, and smell are polysemous, this subset included only uses that were coded as experience perception verbs. (See footnote 9.)

¹⁶ We carried out the same analyses reported in the previous section to ensure that the general characteristics of the full data set held for just experience perception verbs and mental verbs. We found the same overall trends for this subset as for the full sample containing all perception verbs: a linear regression model showed no interaction between subject group (child vs. parent) and verb type (experience perception vs. mental) for overall verb production (β = 0.019, SE = 0.013, t = 1.51, p > 0.05), and a multinomial logistic regression showed that the estimated effect of age for experience perception verbs (β = 0.11) compared to mental verbs (β = 0.74) was similar to the effect of age for all perception verbs compared to mental verbs.

¹⁷ Verb type was included as a fixed, rather than random, effect in our models as we are not attempting to generalize beyond the specific verbs or verb classes examined here (see Clark, 1973).