Children’s (in)ability to recover from garden paths in a verb-final language: Evidence for developing control in sentence processing

https://doi.org/10.1016/j.jecp.2010.01.003Get rights and content

Abstract

An eye-tracking study explored Korean-speaking adults’ and 4- and 5-year-olds’ ability to recover from misinterpretations of temporarily ambiguous phrases during spoken language comprehension. Eye movement and action data indicated that children, but not adults, had difficulty in recovering from these misinterpretations despite strong disambiguating evidence at the end of the sentence. These findings are notable for their striking similarities with findings from children parsing English; however, in those and other studies of English, children were found to be reluctant to use late-arriving syntactic evidence to override earlier verb-based cues to structure, whereas here Korean children were reluctant to use late-arriving verb-based cues to override earlier syntactic evidence. The findings implicate a general cross-linguistic pattern for parsing development in which late-developing cognitive control abilities mediate the recovery from so-called “garden path” sentences. Children’s limited cognitive control prevents them from inhibiting misinterpretations even when the disambiguating evidence comes from highly informative verb information.

Introduction

When children and adults interpret an utterance, they appear to do so in “real time,” rapidly forming a hypothesis about the utterance’s meaning as each word is perceived (see Pickering and Van Gompel, 2006, Tanenhaus, 2007, Trueswell and Gleitman, 2007). The real-time nature of interpretation places specific processing demands on listeners. Because a listener cannot fully predict how a sentence will be completed, the listener’s belief about the meaning of a phrase may turn out to be incorrect with additional linguistic input; For adults, these temporary misinterpretations are just that temporary (e.g., the speaker meant Mary, not Susan, when he said “she”). Adults are able to exert rapid control over their interpretive processes, for example, to inhibit hypotheses proven wrong and to promote hypotheses consistent with newer input. But what about children? In cognitive tasks, even 4- to 6-year-olds have difficulty in exerting control over “prepotent”/automatic responses (e.g., Davidson, Amso, Anderson, & Diamond, 2006). Does this extend to real-time language comprehension? The initial evidence suggests that perhaps it does (e.g., Trueswell et al., 1999, Weighall, 2008; see also Novick, Trueswell, & Thompson-Schill, 2005). Yet, as discussed below, most of this evidence comes from a single language—English. The use of English raises many concerns, most notably because the grammar of any one particular language naturally forces some classes of linguistic evidence to appear earlier in a sentence than others, making it difficult to generalize to other language learners. Here we asked about the development of sentence processing in another language, Korean, which runs its grammar essentially in the opposite direction of English. Will the same developmental patterns emerge? Or is there something specific about the ordering of linguistic information in English that drives what has been learned so far about children’s control over interpretive processes?

Developmental changes in executive function abilities and “cognitive control” are known to occur throughout childhood (Davidson et al., 2006, Diamond et al., 2002, Müller et al., 2005). Executive function (EF) refers to a set of cognitive processes that underlie goal-directed behaviors, including mental flexibility, planning, working memory, and inhibition (Hill, 2004, Huizinga et al., 2006, Miller and Cohen, 2001). EF is crucial in one’s ability to control thoughts and actions so as to generate adapting behaviors to changing needs of the environment (Hill, 2004, Müller et al., 2005, Stuss and Knight, 2002). Preschool-age children frequently show deficits in various cognitive control tasks involving these EF components, particularly inhibitory control (i.e., the ability to inhibit or select a representation under conditions of conflict) (Zelazo, Müller, Frye, & Marcovitch, 2003). They often fail to adapt to the changing rules and tend to perseverate on one rule or dimension in the Dimensional Change Card Sorting Task (DCCST) (Zelazo & Frye, 1998). Also, they perform poorly on tasks that tap into inhibitory control such as the Go/No-Go task (Durston et al., 2002), the Stroop task (Zysset, Müller, Lohmann, & von Cramon, 2001), and other related tasks (Carlson and Moses, 2001, Carlson et al., 2002, Davidson et al., 2006, Sabbagh et al., 2006). In fact, numerous studies have demonstrated that EF and inhibitory control do not fully develop until late adolescence (Anderson, 2002, Anderson et al., 2001, Diamond et al., 2002). This slow development has been linked to the protracted maturation of the associated brain regions, in particular, the prefrontal cortex (Huttenlocher & Dabholkar, 1997; see also Mazuka, Jincho, & Oishi, 2009).

Novick and colleagues (2005) proposed that EF, in particular inhibitory control, plays an important role in language processing when the initial interpretation of a sentence is required to be inhibited in place of a later alternative interpretation. As mentioned above, revision of misinterpretations is a necessary consequence of the serial nature of linguistic input and the incremental nature of spoken language processing. That is, listeners evaluate incoming linguistic evidence in real time, assigning provisional syntactic and semantic analyses essentially on a word-by-word basis (e.g., Altmann and Kamide, 1999, Kjelgaard and Speer, 1999, Sedivy et al., 1999, Tanenhaus et al., 1995). Sometimes, however, these provisional analyses turn out to be incorrect, as evidenced by later information, resulting in processing difficulty. This has often been described as the “garden path” phenomenon, where the listener has been led down the proverbial garden path by misleading evidence and must “back up” to pursue a different path. For instance, consider what occurs when adults hear a sentence like Example (1):

  • (1)

    Put the apple on the towel into the box.

The first prepositional phrase (PP), on the towel, is technically ambiguous; it could be a destination phrase for the verb put (i.e., telling the listener where the apple should be put)1 or a modifier phrase for the noun phrase (NP) the apple (i.e., telling the listener more about the apple). Listeners tend to initially pursue a destination interpretation when they first encounter the ambiguous phrase on the towel. When the sentence continues with an additional PP, such as into the box, the “morphosyntactic” information of into requires the new PP to be the destination phrase, forcing the previous PP (on the towel) to be a modifier of the apple. The initial (and in this case erroneous) parsing preference for the destination analysis of on the napkin is due in large part to the strong syntactic and semantic expectations associated with put; it takes a destination, typically in the form of a PP almost always occurring after a direct object NP.

Evidence for these conclusions comes from eye movement studies of spoken language processing in which listeners act on objects in response to spoken instructions (cf. Tanenhaus et al., 1995, and Trueswell & Tanenhaus, 2005, for a review). This method, often referred to as the “visual world” paradigm, provides a moment-by-moment record of listeners’ real-time interpretive commitments by recording their eye fixations on (ir)relevant objects as they listen to each instruction. Listeners’ eye positions tend to be closely time-locked with the speech they hear (e.g., they will typically fixate on an apple 0 to 200 ms after hearing the word apple) (see Tanenhaus et al., 1995). Tanenhaus and colleagues (1995) also found that eye position could capture listeners’ temporary consideration and revision of garden path sentences. While hearing Put the apple on the towel … as adults viewed a scene containing an apple on a towel, an empty towel, a pencil on a plate, and an empty box, they showed an increased tendency to look over at the empty towel (a potential destination for the putting action) as they heard the word towel. Such an increased fixation tendency was not observed when listeners heard unambiguous sentences such as Put the apple that’s on the towel … Furthermore, on hearing into the box after Put the apple on the towel, listeners tended to show some general confusion, looking around more as compared with the unambiguous sentences. Yet adults ultimately arrived at the correct interpretation, moving the apple into the box in most cases.2

As noted by Novick and colleagues (2005), these language processing conditions are quite similar to those found in EF tasks; listeners must use newly arriving information to inhibit a representation (based on an early input) that is prepotent (i.e., a familiar analysis given past experience with similar input). Adults have developed the ability to modulate and inhibit early-arriving constraints when later countervailing evidence is encountered and thus are able to recover from temporary misanalyses (e.g., considering on the towel as the verb’s destination).

Slow development of EF predicts that children with limited cognitive control abilities would not deal well with these parsing conditions, which require inhibiting an initial prepotent interpretation. In fact, comprehension studies that have used sentences such as Example (1) have found that 4- and 5-year-olds, as compared with adults and older children, have much greater difficulty in revising their initial parsing choice; that is, they often cannot recover from a garden path (hence named the “kindergarten path effect”) (Trueswell et al., 1999; see also Hurewitz et al., 2000, Kidd and Bavin, 2005, Weighall, 2008). These studies also used the visual world eye-tracking paradigm and observed that when children heard an instruction with a temporary ambiguity, Put the frog on the napkin into the box, their early eye fixation patterns resembled those of adults,3 indicating real-time commitment to the destination analysis. Among the four different objects in the scene—a frog on a napkin (target), a frog on a towel (competitor), an empty napkin (incorrect destination), and an empty box (correct destination)—there were significantly more looks to the competitor and to the incorrect destination as compared with the unambiguous condition (Put the frog that’s on the napkin into the box). However, unlike adults, children’s ultimate interpretations (as revealed by their actions on the toy animals) were often consistent with this initial destination analysis rather than the modifier analysis; on more than 60% of temporarily ambiguous trials, children made errors such as moving a frog first to the incorrect destination (the empty napkin) and then to the correct destination (the empty box). Children made very few errors (5%) with unambiguous items. Strikingly, all errors for temporarily ambiguous instructions involved moving an object to the incorrect destination, suggesting that children were honoring the destination analysis of on the napkin. Also, the competitor animal was as likely to be involved in their actions as the target animal, indicative of their failure to reconsider on the napkin as a noun modifier after encountering the second PP. Together, these data suggest that children formed an initial destination interpretation of the ambiguous phrase based on the verb put and had difficulty in revising this commitment even after encountering countervailing linguistic information, with this difficulty perhaps being due to their immature cognitive control abilities.

However, it is also possible that children’s interpretation errors may be arising for other reasons. In particular, they may reflect developmental changes in children’s reliance on particular sources of information to structure in the input. Many studies with children (3–5 years of age or younger), using a wide range of linguistic material, suggest that their parsing system is organized and operates in a way similar to the mature parsing system; that is, it engages in rapid real-time parsing and referential interpretation and can weigh validity and reliability of evidence probabilistically even in the face of ambiguity (Arnold et al., 2007, Epley et al., 2004, Nadig and Sedivy, 2002, Sekerina et al., 2004, Thothathiri and Snedeker, 2008). Yet these studies also reveal changes over developmental time in information use. For instance, some studies suggest that children may rely disproportionately on verb information over extrasentential contextual information when resolving structural ambiguities, perhaps due to the higher reliability of verbs to predict structure (Snedeker and Trueswell, 2004, Snedeker and Yuan, 2008, Trueswell and Gleitman, 2004, Trueswell and Gleitman, 2007). Snedeker and Trueswell (2004) compared children’s use of verb cues and contextual cues to resolve the meaning of globally ambiguous sentences such as Tickle the frog with the feather in which with the feather could be an instrument of the verb tickle or a modifier of the noun frog. The 5-year-olds’ interpretation choices showed great sensitivity to the type of verb used in the sentence (consider Choose the frog …), whereas manipulations of the context to support one interpretation over the other showed no effects even for verbs that were neutral with regard to their semantic/syntactic preferences for instrument phrases. Adults, by contrast, showed good use of these contextual cues to inform their interpretive commitments.

Such disproportionate reliance on verb information may be pertinent to the acquisition order of various cues in which more reliable cues to structure are discovered and acquired earlier (Bates and MacWhinney, 1987, Snedeker and Yuan, 2008). Therefore, the information that regularly predicts syntactic structure more reliably, such as verb-specific information (Boland & Cutler, 1996), becomes employed first developmentally and dominates children’s early parsing processes (Trueswell and Gleitman, 2004, Trueswell and Gleitman, 2007).

This may explain why preschoolers appeared to fail to revise their initial sentence interpretation. It is possible that children in these studies were not necessarily unable to revise the initial choice but instead were reluctant to use morphosyntactic evidence from a preposition (into) to override what they know about a common verb (put). When hypotheses generated from different information sources compete, children may lean more toward those that are suggested by a more reliable information source on syntax (i.e., verb information) than others (i.e., morphosyntactic information), similar to the idea presented by MacWhinney, Pleh, and Bates (1985). It is possible, then, that children’s difficulty in revising the temporarily ambiguous items reflects an information reliability difference in their parsing system. Under this view, as children gain more experience with their language, they become attuned to exceptional parsing circumstances that require reevaluation of cue validity and become better able to evaluate a diverse set of evidence to determine sentence structure and meaning.

With English data, both accounts outlined above—immature cognitive control and differences in children’s information use—are equally plausible. In English, verb-specific lexical information tends to arise early in a sentence relative to other information. In fact, the verb appears first in the type of instructions (i.e., imperative sentences) used in these visual world studies. This early access to verb information in English sentences makes it hard to determine whether 5-year-olds’ verb-consistent interpretation patterns were because a verb-based representation was formed early and thus became a hypothesis that needed to be inhibited in favor of a later-formed representation or because children were reluctant to override a verb-generated analysis with that indicated by less predictive morphosyntactic information. Languages such as Korean, however, would permit us to pull apart these two possibilities. As we discuss below, Korean, a head-final language, offers an exceptionally strong test of the generality of children’s failure to inhibit earlier interpretations as opposed to their reliance on particular information such as verb-specific lexical constraints.

Because of its head-final nature (Baker, 2001), the order by which information arrives to the ear in Korean is, to a first approximation, often opposite to that of English. For example, compare Example (2) with its Korean equivalent in Example (3):Verbs appear sentence/clause finally in Korean (Kim, 1999), whereas verb information tends to arise much earlier in a sentence in English. What this means in the current context is that verb information, which is used to guide parsing commitments in English, instead is used to confirm and elaborate sentence parses in Korean (and, as we discuss below, sometimes to disambiguate parses). Case markers (-ey and -ul above) in principle serve as cues to the structure of the input and, as such, guide processing.

Adult sentence processing studies support the idea that adult listeners and readers of head-final languages such as Korean (and Japanese) parse sentences incrementally just as head-initial language listeners do and thus do not await the verb to determine the sentence meaning/structure (Kamide et al., 2003, Kamide and Mitchell, 1999, Konieczny et al., 1997; see Mazuka & Nagai, 1995, for detailed proposals). Listeners/Readers of head-final languages tend to begin projecting the structure of an upcoming sentence as soon as they gain access to structure-relevant evidence such as morphosyntactic information from case markers (Kamide & Mitchell, 1999) or argument structure information concerning the number and type of NPs (e.g., Aoshima et al., 2004, Kamide et al., 2003). As this evidence suggests, we believe that Korean adults as well as children incrementally parse sentences, starting to build their interpretations in real time as soon as they encounter linguistic evidence.

Like English (and all other natural languages for that matter), Korean contains temporary syntactic ambiguity. For instance, even the simple sentence in Example (3) above contains a temporary ambiguity that is illustrated below in Examples (4a) and (4b):When the PP naypkhin-ey appears prior to an NP kaykwuli-lul, as in Examples (4a) and (4b), it remains ambiguous until the verb arrives because it can be interpreted either as a modifier or as a destination phrase, depending on the verb.4 The case marker -ey here can be either a locative marker, a genitive marker, or a reduced form of a full relative -ey issnun (i.e., on/in-). When a verb requires a destination phrase (put in Example 4a), naypkhin-ey serves as the destination of put, indicating where the frog should be placed. Here the marker -ey is used as a locative, indicating a destination. By contrast, if the verb is pick up as in Example (4b), no destination phrase is allowed and naypkhin-ey becomes the modifier of the following noun, specifying the target referent as the frog that is on the napkin, not the one on the book. The marker -ey, in this case, is either a reduced form of a full relative, -ey issnun, or a genitive, -uy (although spelled differently, this marker is pronounced the same as the locative -ey). Thus, the phrase naypkhin-ey kaykwuli-lul (we use napkin-on frog-Acc to refer to these hereafter) in Examples (4a) and (4b) is temporarily ambiguous until the verb becomes available.

Theories of initial parsing preferences almost uniformly predict that such a temporary ambiguity would be resolved toward a destination interpretation. That is, listeners should initially interpret napkin-on frog-Acc as two separate arguments (the napkin is a destination to which the frog will be going). For instance, minimal attachment theory (Frazier, 1987) predicts that listeners have a preference against complex NPs such as the sort required if napkin-on frog-Acc were treated as a single NP with napkin-on serving as a modifier of frog-Acc. Other parsing theories, which instead emphasize the frequency of syntactic alternatives (e.g., constraint-based parsing theories of MacDonald et al., 1994, Trueswell and Tanenhaus, 1994), make the same prediction but for a different reason; the ambiguous marker -ey is much more commonly used as a locative (destination) than as a genitive (modifier) in the child-directed linguistic input. In a prior corpus analysis and two sentence-completion experiments, we found a strong tendency for the destination use of these PPs. An analysis of a corpus of Korean child-directed speech identified 115 occurrences of this type of phrase with -ey, with 108 of these marking a destination and only 7 cases where the phrases were modifying the following NP.5 This tendency for the destination interpretation was also observed in written and auditory sentence completion studies.6

Thus, there is a strong expectation that Korean listeners will have a bias to initially interpret napkin-on as a destination phrase for an upcoming verb. Ending the sentence with the verb put would confirm this interpretation. However, ending it with pick up does not confirm this analysis and instead forces the modifier interpretation (e.g., the frog that’s on the napkin). Thus, the sentence involving put (Example 4a above) should be an easy sentence for Korean children to understand, whereas the one involving pick up (Example 4b) should not; it should cause a garden path.

On the face of it, however, simple sentences such as Pick up the frog on the napkin and Put the frog on the napkin should be easy for Korean children to understand. What children know about these common verbs should tell them what to do with the phrase on the napkin. Indeed, if children rely disproportionately on verb information to determine structural interpretations (as the “change in information use” account predicts), Korean children should show little difficulty in parsing these sentences. According to this view, despite the fact that the verb becomes accessible later than other sources in Korean, young Korean children would be more willing to override their initial case marker (morphosyntactic)-based interpretation with that proposed by a verb because verbs are a highly reliable source of sentence structure. As a result, their interpretations should tend to be more consistent with verb information than with case-marking information, predicting that both pick up and put sentences would be unlikely to elicit errors.

Yet the cognitive control account makes a strong prediction contrary to this intuition. Because the pick up sentence (Example 4b) above requires revision, children, but not adults, should make errors when they hear pick up and produce nonadult actions consistent with napkin-on being the verb’s destination phrase. They ought to, for instance, move the frog over to another napkin rather than raising it in the air. By contrast, the put version is not expected to flummox Korean children because ending the sentence this way is consistent with the initial interpretation. Under this account, although Korean children know the meanings of these verbs, cognitive control limitations sometimes prevent using this knowledge to override an initial parsing choice.

To test these predictions, the current experiment examined how Korean adults and 4- and 5-year-olds parse these temporarily ambiguous sentences described above ending with the verb put or pick up, using the visual world eye-tracking paradigm.

Section snippets

Participants

Participants were 16 adult native speakers of Korean (13 women and 3 men) and 16 children (4 years 0 months to 5 years 4 months of age, mean = 4 years 9 months, 10 girls and 6 boys). Adult participants were recruited via a subject participant pool at Ajou University in Suwon, Korea, and they received course credit for their participation. All participating children were monolingual Korean speakers recruited from preschools in Suwon. The experimenter visited the preschools and tested children

Eye gaze patterns from onset of verb

The central question of interest here is whether Korean children also show difficulty in revising their initial parsing as compared with adults. To assess this, we examined looks to the destination object (the empty napkin) on hearing the sentence-final verb—pick up or put. Looks to the destination object can be taken as evidence for listeners’ consideration of the destination analysis of the PP (e.g., Spivey et al., 2002). It is expected that Korean-speaking adults will use their knowledge of

General discussion

The central finding of the current study is that Korean-speaking children show substantial difficulty in recovering from garden path sentences even though the disambiguating evidence at the end of the sentence was a verb. Eye gaze patterns demonstrated that, unlike adults, Korean-speaking children had considerable difficulty in using their knowledge of pick up to block consideration of the initial destination interpretation of napkin-on; both sentences ending with pick up and sentences ending

Closing remarks

The results of our study suggest general difficulty for children when recovering from garden path sentences. It appears that young children (regardless of the language they are learning) rely disproportionately on early-arriving constraints on structure to resolve ambiguity, whereas postambiguity constraints exert less influence even when they are highly informative. Such a pattern is expected cross-linguistically under constraint-based accounts of language development that incorporate notions

Acknowledgments

This research was supported by National Institutes of Health (NIH) Grant R01-HD37507 (awarded to John Trueswell and Lila Gleitman). We thank David January, Allison Wessel, Alon Hafri, and other lab members at the University of Pennsylvania for their help in coding the data and their valuable comments. We thank Young-jin Kim and Kwangill Choi, at Ajou University, as well as Doyun Lee, the director at Rainbow Child Care in Korea, for their help in data collection. We are also grateful to Emily

References (65)

  • M.M. Kjelgaard et al.

    Prosodic facilitation and interference in the resolution of temporary syntactic closure ambiguity

    Journal of Memory and Language

    (1999)
  • B. MacWhinney et al.

    The development of sentence interpretation in Hungarian

    Cognitive Psychology

    (1985)
  • U. Müller et al.

    Executive function and children’s understanding of false belief: How specific is the relation?

    Cognitive Development

    (2005)
  • M.J. Pickering et al.

    Sentence processing

  • J. Sedivy et al.

    Achieving incremental semantic interpretation through contextual representation

    Cognition

    (1999)
  • J. Snedeker et al.

    The developing constraints on parsing decisions: The role of lexical-biases and referential scenes in child and adult sentence processing

    Cognitive Psychology

    (2004)
  • J. Snedeker et al.

    Effects of prosodic and lexical constraints on parsing in young children (and adults)

    Journal of Memory and Language

    (2008)
  • M.J. Spivey et al.

    Eye movements and spoken language comprehension: Effects of visual context on syntactic ambiguity resolution

    Cognitive Psychology

    (2002)
  • M. Thothathiri et al.

    Give and take: Syntactic priming during spoken language comprehension

    Cognition

    (2008)
  • J.C. Trueswell et al.

    The kindergarten-path effect: Studying on-line sentence processing in young children

    Cognition

    (1999)
  • A.R. Weighall

    The kindergarten path effect revisited: Children’s use of context in processing structural ambiguities

    Journal of Experimental Psychology

    (2008)
  • S. Zysset et al.

    Color-word matching Stroop task: Separating interference and response conflict

    NeuroImage

    (2001)
  • P.D. Allopoenna et al.

    Tracking the time course of spoken word recognition using eye movements: Evidence for continuous mapping models

    Journal of Memory and Language

    (1998)
  • G.T.M. Altmann et al.

    Now you see it, now you don’t: Mediating the mapping between language and the visual world

  • V.A. Anderson

    Executive function in children: Introduction

    Child Neuropsychology

    (2002)
  • V. Anderson et al.

    Development of executive functions through late childhood and adolescence in an Australian sample

    Developmental Neuropsychology

    (2001)
  • J.E. Arnold et al.

    Children’s use of gender and order-of-mention during pronoun comprehension

    Language and Cognitive Processes

    (2007)
  • M.C. Baker

    The atoms of language

    (2001)
  • E.A. Bates et al.

    Competition, variation, and language learning

  • Boersma, P., & Weenink, D. (2004). Pratt: Doing phonetics by computer. Available from:...
  • S.M. Carlson et al.

    Individual differences in inhibitory control and children’s theory of mind

    Child Development

    (2001)
  • S.M. Carlson et al.

    How specific is the relation between executive function and theory of mind? Contributions of inhibitory control and working memory

    Infant and Child Development

    (2002)
  • Cited by (0)

    View full text