Auditory Artificial Grammar Learning in Macaque and Marmoset Monkeys

Benjamin Wilson, Heather Slater, Yukiko Kikuchi, Alice E. Milne, William D. Marslen‐Wilson, Kenny Smith, Christopher I. PetkovJournal of Neuroscience 2013 · DOI

Abstract

Artificial grammars (AG) are designed to emulate aspects of the structure of language, and AG learning (AGL) paradigms can be used to study the extent of nonhuman animals' structure-learning capabilities. However, different AG structures have been used with nonhuman animals and are difficult to compare across studies and species. We developed a simple quantitative parameter space, which we used to summarize previous nonhuman animal AGL results. This was used to highlight an under-studied AG with a forward-branching structure, designed to model certain aspects of the nondeterministic nature of word transitions in natural language and animal song. We tested whether two monkey species could learn aspects of this auditory AG. After habituating the monkeys to the AG, analysis of video recordings showed that common marmosets (New World monkeys) differentiated between well formed, correct testing sequences and those violating the AG structure based primarily on simple learning strategies. By comparison, Rhesus macaques (Old World monkeys) showed evidence for deeper levels of AGL. A novel eye-tracking approach confirmed this result in the macaques and demonstrated evidence for more complex AGL. This study provides evidence for a previously unknown level of AGL complexity in Old World monkeys that seems less evident in New World monkeys, which are more distant evolutionary relatives to humans. The findings allow for the development of both marmosets and macaques as neurobiological model systems to study different aspects of AGL at the neuronal level.

TL;DR

This study establishes a new quantitative framework to compare artificial grammar learning (AGL) across species, revealing that Rhesus macaques (Old World monkeys) can master complex, branching auditory structures that elude common marmosets (New World monkeys). While both species detect simple rule-breaking, only macaques show evidence of "deep" learning, processing nondeterministic transitions similar to those found in human language and animal song.

What Problem Does This Paper Address?

To understand the evolutionary origins of language, researchers often use Artificial Grammars (AG)—sets of rules that generate sequences of stimuli—to see if nonhuman animals can learn structural patterns. However, the field has been plagued by a "comparison problem": different studies use wildly different grammar structures, making it nearly impossible to say if a songbird is "smarter" at syntax than a primate.

Furthermore, most animal AGL research has focused on very simple, deterministic patterns (e.g., AABB). Human language is rarely that predictable; it is nondeterministic and branching. This paper seeks to bridge these gaps by creating a unified complexity scale and testing whether primates can handle more lifelike, branching auditory grammars.

Figure 1: Quantitative parameter space for comparing AG complexity across species.

Methods

The researchers first developed a quantitative parameter space to map AG complexity based on two factors: the number of stimulus classes/elements and structural linearity (a measure of predictability). They then tested 13 Rhesus macaques and 4 common marmosets using a "Reber-like" grammar consisting of five naturally spoken nonsense words (Klor, Biff, Jux, Cav, Dupp).

During the habituation phase, monkeys listened to "correct" sequences. In the testing phase, they were presented with both "Correct" (well-formed) sequences and "Violation" sequences that broke the branching rules. Responses were measured using blind-coded video analysis of orienting behaviors. Crucially, the team developed a novel infrared eye-tracking task for the macaques to pinpoint exactly when in a sequence a monkey noticed a mistake.

Key Insights and Ideas

  • The Evolutionary Gradient: Macaques (Old World monkeys) demonstrated significantly more sophisticated learning than marmosets (New World monkeys). This suggests that the capacity for complex structural learning may have increased along the primate lineage leading to humans.
  • Strategies vs. Structure: Marmosets appeared to rely on "simple" strategies—they noticed if a sequence started with the wrong word or if the sequence was entirely unfamiliar. Macaques, however, could identify violations even in novel sequences that followed the correct "starting" rules, suggesting they learned the internal branching logic.
  • Decoupling Perception from Production: Interestingly, macaques show high AGL capacity despite having very limited vocal flexibility. This suggests that the brain mechanisms for perceiving complex syntax-like structures evolved independently of the mechanisms for producing complex speech or song.

Key Results

The results highlighted a clear divergence in learning depth between the two species. Macaques showed a robust preference for violation sequences ($p < 0.001$), and eye-tracking data proved they were sensitive to subtle violations occurring late in a sequence (positions 3-5), even when the beginning of the sequence was perfectly legal.

In contrast, while marmosets did respond to violations ($p = 0.032$), post-hoc analysis showed this was driven almost entirely by sequences that broke the very first rule (the "starting" element). When the beginning was correct, they failed to notice branching errors later in the string.

Figure 5: Behavioral results comparing Rhesus macaques (A, B) and common marmosets (C, D) in response to correct and violation sequences.

Limitations and Future Work

  • Sample Size and Technology: The study included only four marmosets, and eye-tracking was not possible for them due to their small size and head-movement patterns. A larger cohort and improved tracking would strengthen comparative claims.
  • Neural Mechanisms: While this paper establishes behavioral capacity, it does not record neural activity. The authors suggest that both species can now serve as "neurobiological model systems" to study how neurons encode these grammatical rules in the auditory cortex and frontal lobes.

Summary updated

Abstract

Artificial grammars (AG) are designed to emulate aspects of the structure of language, and AG learning (AGL) paradigms can be used to study the extent of nonhuman animals' structure-learning capabilities. However, different AG structures have been used with nonhuman animals and are difficult to compare across studies and species. We developed a simple quantitative parameter space, which we used to summarize previous nonhuman animal AGL results. This was used to highlight an under-studied AG with a forward-branching structure, designed to model certain aspects of the nondeterministic nature of word transitions in natural language and animal song. We tested whether two monkey species could learn aspects of this auditory AG. After habituating the monkeys to the AG, analysis of video recordings showed that common marmosets (New World monkeys) differentiated between well formed, correct testing sequences and those violating the AG structure based primarily on simple learning strategies. By comparison, Rhesus macaques (Old World monkeys) showed evidence for deeper levels of AGL. A novel eye-tracking approach confirmed this result in the macaques and demonstrated evidence for more complex AGL. This study provides evidence for a previously unknown level of AGL complexity in Old World monkeys that seems less evident in New World monkeys, which are more distant evolutionary relatives to humans. The findings allow for the development of both marmosets and macaques as neurobiological model systems to study different aspects of AGL at the neuronal level.
BibTeX
@article{wilson2013auditory,
  author = {Benjamin Wilson and Heather Slater and Yukiko Kikuchi and Alice E. Milne and William D. Marslen‐Wilson and Kenny Smith and Christopher I. Petkov},
  title = {{Auditory Artificial Grammar Learning in Macaque and Marmoset Monkeys}},
  journal = {Journal of Neuroscience},
  year = {2013},
  doi = {10.1523/jneurosci.2414-13.2013},
}
Figure