TL;DR
This paper reviews how humans and other animals learn to track relationships between words or elements that are not next to each other in a sequence. It highlights that while this ability is crucial for language, it's cognitively demanding and often requires specific cues to facilitate learning, suggesting it might be an ancient cognitive capacity that predates language itself.
What Problem Does This Paper Address?
Natural language relies heavily on understanding how words and phrases relate to each other, even when separated by other words. These "nonadjacent dependencies" are fundamental to sentence structure (e.g., subject-verb agreement across clauses). However, tracking these relationships is more cognitively taxing than processing adjacent elements, requiring working memory to hold information over time. The paper aims to understand how this crucial ability develops in humans (infants to adults) and whether it's present in other animals, shedding light on its evolutionary origins and its role as a potential precursor to language.
Methods
The primary method discussed is artificial grammar learning (AGL). This involves exposing participants (humans, infants, or animals) to sequences of artificial stimuli (sounds, visual patterns) that follow specific rules, including nonadjacent dependencies. Participants are then tested on their ability to distinguish between sequences that conform to the learned rules and those that violate them. The paper also discusses structured sequence processing tasks, which are closely related. The research categorizes different types of nonadjacent dependencies (e.g., identity relations, arbitrary associations, category-based relationships) and examines various "cues" that facilitate learning, such as perceptual similarity, prosodic cues, pauses, and "edge effects" (positioning dependent elements at the beginning or end of a sequence).
Datasets
Not discussed in the paper. The studies reviewed use artificial stimuli generated according to specific grammar rules, rather than pre-existing natural language datasets.
Key Insights and Ideas
- Nonadjacent dependency learning is cognitively demanding: It requires more effort than learning adjacent relationships, likely due to working memory demands.
- Facilitating cues are crucial: Learning nonadjacent dependencies is significantly enhanced by cues that highlight the relationship between the distant elements. These cues can be physical, perceptual (e.g., similar sounds), attentional, or structural (e.g., pauses, edge effects).
- Early emergence in humans: Infants as young as 7 months can detect simple nonadjacent dependencies (identity relations), and even more complex arbitrary ones by 3 months, suggesting a foundational cognitive ability.
- Cross-species parallels: Many nonhuman animals, including primates and birds, show evidence of learning nonadjacent dependencies, particularly identity relations and some arbitrary ones. This suggests a shared evolutionary origin for this ability.
- Domain-generality: The ability to learn nonadjacent dependencies is not limited to linguistic material and applies to other sequence-based learning tasks, indicating it might be a general cognitive mechanism.
- Language as a potential driver: While the ability is not exclusive to language, its importance for language processing and acquisition suggests it may have been a crucial cognitive precursor that was co-opted and refined for linguistic communication.
Key Results
- Human infants and adults can learn nonadjacent dependencies, but learning is often conditional on the presence of facilitating cues. For instance, Gomez (2002) found that infants and adults could learn "AXB" dependencies only when the intervening "X" elements were highly variable, making the adjacent relationships uninformative.
- Nonhuman animals show varying degrees of success. Squirrel monkeys and chimpanzees can learn identity-based nonadjacent dependencies ("AXA"). Studies with tamarins and rats suggest they can learn arbitrary dependencies ("AXB") and category-based ones, sometimes showing different sensitivities than humans (e.g., tamarins being sensitive to vowel sounds).
- Mixed results in primates: Some studies with macaques showed they could learn adjacent dependencies but not nonadjacent ones in mixed-complexity grammars, possibly due to the salience of adjacent relationships. However, EEG studies suggest macaques might be sensitive to nonadjacent dependencies under certain conditions.
- Birds (zebra finches) can learn nonadjacent dependencies through a "starting small" approach, where they first learn adjacent relationships and then gradually increase complexity.
Limitations and Future Work
- Cross-species differences: While similarities exist, there are also differences in how humans and nonhuman animals learn nonadjacent dependencies, suggesting potential species-specific adaptations or constraints. Further research is needed to clarify these differences.
- Neurobiological underpinnings: While the paper discusses the cognitive aspects, more research is needed to understand the neural mechanisms and computations that support this learning across species.
- Evolutionary origins: The paper suggests a common evolutionary origin or convergent evolution for these abilities. Future work could leverage comparative studies to further explore the phylogenetic roots of these cognitive capacities.
- Domain specificity: While evidence points to domain-generality, further research is needed to fully delineate the boundaries of this ability and its specific role in language acquisition versus other cognitive domains.
Summary updated