Christiansen & Chater (C&C) argue that higher-level chunks preserve information from lower-level chunks albeit in a much impoverished form. However, they also suggest that there is no obligatory relationship between low-level chunks and high-level chunks. To support their claim, they cite the case of SF (cf. Ericsson et al. Reference Ericsson, Chase and Faloon1980), who could accurately recall as many as 79 digits after grouping them in locally meaningful units (e.g., historical dates and human ages). Moreover, they argue that the Now-or-Never bottleneck forbids broad parallelism in language at the expense of avoiding ambiguities (e.g., “garden path” sentences). In brief, C&C propose that chunks are only locally coherent and that their gist, however contradictory, is being safely kept track of at higher levels. Unfortunately, the authors remain silent about the mechanisms underlying higher-level representation formation.
C&C also declare themselves agnostic about the nature of chunks. Indeed, although there is ample psychological evidence for the existence of chunks in various types of experimental data, from pause durations in reading to naive sentence diagramming, chunks remain notoriously difficult to define. However, we have reasons to reject the possibility, which follows naturally from the Chunk-and-Pass framework, that chunks are arbitrary and may depend exclusively on memory limitations. To wit, chunks correspond most closely to intonational phrases (IPs) (cf. Gee & Grosjean Reference Gee and Grosjean1983), which, in turn, are hard to capture by grammatical rules. For example, the sentence “This is the cat / that chased the rat / that ate the cheese” contains three IPs (separated by slashes) that fail to correspond to syntactic constituents (noun phrases or verb phrases). Yet IPs are not entirely free of structure, as they must begin at the edge of a syntactic constituent and end before or at the point where a syntactic constituent ends (cf. Jackendoff Reference Jackendoff2007). Moreover, although a given utterance can be carved up in several ways (hence, contain a variable number of IPs), carvings are not arbitrary and license only certain IP combinations and not others. We may therefore conclude that IPs and corresponding chunks must be globally coherent (i.e., fit well with each other) and depend on the meaning conveyed.
Furthermore, I believe that chunking is driven not by memory limitations nor by language structures, but by an overall need for coherence or meaningfulness (cf. Dumitru Reference Dumitru2014). Indeed, evidence from memory enhancement techniques suggests that chunking must rely on global coherence. So, for example, memory contest champions who use the so-called mind palace technique (e.g., Yates Reference Yates1966) often achieve impressive results. The method requires them to commit to long-term memory a vivid image associated with each item to be remembered (e.g., faces, digits, and lists of words) as well as a familiar walk through the palace where these images are stored at precise locations. Whenever necessary, contestants can retrace the walk through the palace (i.e., rely on global coherence) to recall a huge number of unrelated facts.
I also claim that coherence is grounded in a model of reality that people instantly build when recalling items or understanding language based on their experience with frequent patterns of perception and action (Barsalou Reference Barsalou2008). Indeed, as shown in Altmann (Reference Altmann2002) and in Altmann and Kamide (Reference Altmann and Kamide2009), for instance, people use available lexical information at the earliest stages of processing to anticipate upcoming words. Furthermore, as reported in Kamide et al. (Reference Kamide, Altmann and Haywood2003), people target a larger sentential context during online processing, hence would look more readily toward a glass of beer when hearing “The man will taste the …” but towards candy when hearing “The girl will taste the …,” for instance, although the verb itself combines equally well with “beer” and with “candy.” Subsequently, Dumitru and Taylor (Reference Dumitru and Taylor2014) reported that disjunction words like “or” cue knowledge about expected argument structure and sense depending on grounded sentential context. More important, language processing may reflect knowledge of the world that goes beyond people's awareness and beyond language structures (cf. Dumitru et al. Reference Dumitru, Joergensen, Cruickshank and Altmann2013). In particular, when understanding conjunction and disjunction expressions, people rapidly establish grounded connections between the two items mentioned (i.e., the concepts evoked by the nouns linked by conjunction or by disjunction) in the form of Gestalts.
Accordingly, people shifted their gaze faster from the picture of an ant to the picture of a cloud when hearing “Nancy examined an ant and a cloud” than when hearing “Nancy examined an ant or a cloud”; they could instantly evoke a single Gestalt in conjunction situations (where they usually select both items mentioned) and two Gestalts in disjunction situations (where they usually select one of the items). As expected, their attention shifted faster between two object parts (in this case, two representations belonging to the same Gestalt) than between two objects (two representations belonging to different Gestalts). Subsequent work by Dumitru (Reference Dumitru2014) confirmed that Gestalts generate perceptual compatibility effects such that visual groupings of a particular set of stimuli (e.g., two different-coloured lines) had complementary effects on validation scores for conjunction descriptions as opposed to disjunction descriptions. Importantly, language users need not be aware of the dynamics of the concept-grounding process (i.e., why they shift their gaze between stimuli at a certain speed), and there are no language-related constraints that might explain these differences in behaviour.
To summarise, I have questioned the proposal by C&C that Chunk-and-Pass processing is exclusively driven by memory constraints and obeys the rules of the language system. Instead, I discussed recent evidence suggesting that chunking is driven by a need for global coherence manifested as Gestalt-like structures, which in turn underlie memory organisation and mirror real-world phenomena. I further suggested that Gestalts hijack Chunk-and-Pass processing when they generate online effects that are not language-related.
Christiansen & Chater (C&C) argue that higher-level chunks preserve information from lower-level chunks albeit in a much impoverished form. However, they also suggest that there is no obligatory relationship between low-level chunks and high-level chunks. To support their claim, they cite the case of SF (cf. Ericsson et al. Reference Ericsson, Chase and Faloon1980), who could accurately recall as many as 79 digits after grouping them in locally meaningful units (e.g., historical dates and human ages). Moreover, they argue that the Now-or-Never bottleneck forbids broad parallelism in language at the expense of avoiding ambiguities (e.g., “garden path” sentences). In brief, C&C propose that chunks are only locally coherent and that their gist, however contradictory, is being safely kept track of at higher levels. Unfortunately, the authors remain silent about the mechanisms underlying higher-level representation formation.
C&C also declare themselves agnostic about the nature of chunks. Indeed, although there is ample psychological evidence for the existence of chunks in various types of experimental data, from pause durations in reading to naive sentence diagramming, chunks remain notoriously difficult to define. However, we have reasons to reject the possibility, which follows naturally from the Chunk-and-Pass framework, that chunks are arbitrary and may depend exclusively on memory limitations. To wit, chunks correspond most closely to intonational phrases (IPs) (cf. Gee & Grosjean Reference Gee and Grosjean1983), which, in turn, are hard to capture by grammatical rules. For example, the sentence “This is the cat / that chased the rat / that ate the cheese” contains three IPs (separated by slashes) that fail to correspond to syntactic constituents (noun phrases or verb phrases). Yet IPs are not entirely free of structure, as they must begin at the edge of a syntactic constituent and end before or at the point where a syntactic constituent ends (cf. Jackendoff Reference Jackendoff2007). Moreover, although a given utterance can be carved up in several ways (hence, contain a variable number of IPs), carvings are not arbitrary and license only certain IP combinations and not others. We may therefore conclude that IPs and corresponding chunks must be globally coherent (i.e., fit well with each other) and depend on the meaning conveyed.
Furthermore, I believe that chunking is driven not by memory limitations nor by language structures, but by an overall need for coherence or meaningfulness (cf. Dumitru Reference Dumitru2014). Indeed, evidence from memory enhancement techniques suggests that chunking must rely on global coherence. So, for example, memory contest champions who use the so-called mind palace technique (e.g., Yates Reference Yates1966) often achieve impressive results. The method requires them to commit to long-term memory a vivid image associated with each item to be remembered (e.g., faces, digits, and lists of words) as well as a familiar walk through the palace where these images are stored at precise locations. Whenever necessary, contestants can retrace the walk through the palace (i.e., rely on global coherence) to recall a huge number of unrelated facts.
I also claim that coherence is grounded in a model of reality that people instantly build when recalling items or understanding language based on their experience with frequent patterns of perception and action (Barsalou Reference Barsalou2008). Indeed, as shown in Altmann (Reference Altmann2002) and in Altmann and Kamide (Reference Altmann and Kamide2009), for instance, people use available lexical information at the earliest stages of processing to anticipate upcoming words. Furthermore, as reported in Kamide et al. (Reference Kamide, Altmann and Haywood2003), people target a larger sentential context during online processing, hence would look more readily toward a glass of beer when hearing “The man will taste the …” but towards candy when hearing “The girl will taste the …,” for instance, although the verb itself combines equally well with “beer” and with “candy.” Subsequently, Dumitru and Taylor (Reference Dumitru and Taylor2014) reported that disjunction words like “or” cue knowledge about expected argument structure and sense depending on grounded sentential context. More important, language processing may reflect knowledge of the world that goes beyond people's awareness and beyond language structures (cf. Dumitru et al. Reference Dumitru, Joergensen, Cruickshank and Altmann2013). In particular, when understanding conjunction and disjunction expressions, people rapidly establish grounded connections between the two items mentioned (i.e., the concepts evoked by the nouns linked by conjunction or by disjunction) in the form of Gestalts.
Accordingly, people shifted their gaze faster from the picture of an ant to the picture of a cloud when hearing “Nancy examined an ant and a cloud” than when hearing “Nancy examined an ant or a cloud”; they could instantly evoke a single Gestalt in conjunction situations (where they usually select both items mentioned) and two Gestalts in disjunction situations (where they usually select one of the items). As expected, their attention shifted faster between two object parts (in this case, two representations belonging to the same Gestalt) than between two objects (two representations belonging to different Gestalts). Subsequent work by Dumitru (Reference Dumitru2014) confirmed that Gestalts generate perceptual compatibility effects such that visual groupings of a particular set of stimuli (e.g., two different-coloured lines) had complementary effects on validation scores for conjunction descriptions as opposed to disjunction descriptions. Importantly, language users need not be aware of the dynamics of the concept-grounding process (i.e., why they shift their gaze between stimuli at a certain speed), and there are no language-related constraints that might explain these differences in behaviour.
To summarise, I have questioned the proposal by C&C that Chunk-and-Pass processing is exclusively driven by memory constraints and obeys the rules of the language system. Instead, I discussed recent evidence suggesting that chunking is driven by a need for global coherence manifested as Gestalt-like structures, which in turn underlie memory organisation and mirror real-world phenomena. I further suggested that Gestalts hijack Chunk-and-Pass processing when they generate online effects that are not language-related.