LAG-1 Human

What is LAG-1 in the context of human research?

LAG-1 refers to two distinct but important concepts in human research: (1) a reinforcement schedule used in behavioral interventions, particularly for individuals with autism spectrum disorder (ASD), and (2) a computational neural field model that integrates learning, attention, and gaze behaviors in humans.
In behavioral science, LAG-1 refers to a schedule of reinforcement in which delivery of a reinforcer is contingent on a response being different from the previous response. For example, in a LAG-1 schedule, a subject receives reinforcement only when their current response differs from their immediately preceding response . This approach has been shown to effectively promote response variability in both verbal and nonverbal behaviors.
In cognitive neuroscience, LAG-1 represents a dynamic neural field model of learning, attention, and gaze that can be fitted to human learning and eye-movement data. This model comprises three control systems: one for visuospatial attention, one for saccadic timing and control, and one for category learning .

What are the core components of the LAG-1 neural field model?

The LAG-1 neural field model consists of three primary control systems that work in concert:

• Visuospatial attention system: This component manages how visual attention is distributed across stimuli and features.

• Saccadic timing and control system: This governs when and where eye movements occur during visual processing.

• Category learning system: This component handles how categorical information is acquired and associated with visual features .
  At its theoretical core, LAG-1 proposes that simple associative learning provides information about feature importance through the contrast of weights developed during learning. These learned weights influence an attentional priority map that integrates with bottom-up visual signals to drive saccade initiation. With only three free parameters (learning rate, trial impatience, and fixation impatience), the model can produce qualitatively accurate predictions for multiple aspects of human behavior .

How are Lag schedules implemented in verbal response variability training?

Implementing Lag schedules in verbal response variability training involves several methodological steps:

• Baseline assessment: The researcher first documents the participant's current response patterns to target questions or prompts, noting if responses are rote, invariable, or inappropriate.

• Schedule selection: The appropriate Lag level (e.g., Lag 1, Lag 2) is selected based on the participant's baseline abilities and the intervention goals.

• Structured implementation:

  • Questions are presented in a randomized order.

  • Responses are reinforced only when they meet the lag criterion (i.e., differ from the previous 1, 2, or more responses).

  • Error correction procedures are implemented when responses don't meet the criterion.

• Visual supports: Pictures of preferred activities or appropriate responses may be shown to prompt correct responding.

• Prompted trials: When errors occur, the therapist provides a model and then tests the participant's ability to generate the response with decreased support .
  The research by Lee, McComas, and Jawor (2002) demonstrated this approach using a Lag 1/DRA (differential reinforcement of alternative behavior) with three participants diagnosed with ASD. Their intervention used discrete-trial training and a token economy system to reinforce appropriate varied responses to social questions .

What error correction procedures are used with Lag schedules?

When implementing Lag schedules, specific error correction procedures help maintain the integrity of the intervention:

• Verbal model provision: When a participant fails to provide a response that meets the lag criterion, the therapist provides a verbal model of an appropriate alternative response.

• Visual stimulus pairing: The verbal model is often paired with a visual stimulus that represents the modeled response (e.g., showing a picture of the participant engaged in the activity being modeled).

• Prompted trial sequence:

  • The therapist restates the same question while showing the visual stimulus (without verbal model)

  • After a correct prompted response, the question is re-presented without visual support

  • If the participant responds correctly, the therapist proceeds with the intervention sequence

  • If incorrect, the error correction procedure is repeated
    In the case study with Belle (an 11-year-old with ASD), if she responded incorrectly to a social question, the therapist would implement error correction by holding up a picture of a preferred activity and providing a verbal model. After Belle repeated the model, a prompted trial followed where the therapist restated the question with visual support but without verbal prompting .

How does the LAG-1 model integrate bottom-up and top-down attentional processes?

The LAG-1 model offers a sophisticated integration of bottom-up and top-down attentional processes through its computational architecture:

• Information gain extraction: LAG-1 extracts a form of information gain from pairwise differences in associations between visual features and categories. This serves as a measure of feature diagnosticity that evolves during learning.

• Reentrant signal pathway: This information gain is provided as a reentrant signal that combines with:

  • Bottom-up visual information from stimulus processing

  • Top-down spatial priority signals from cognitive control systems

• Attentional priority map: These combined signals influence an attentional priority map that determines the distribution of attentional resources.

• Saccade initiation: The state of this priority map directly influences when and where saccades are initiated, creating a dynamic relationship between learning and gaze behavior .
  This integration allows the model to predict how attention shifts during learning as the learner discovers which features are most diagnostic for category decisions. The model demonstrates how simple associative learning produces sophisticated attentional dynamics that shape visual exploration during learning tasks.

What parameters can be adjusted in the LAG-1 model when fitting to human data?

The LAG-1 model employs three primary free parameters when fitting to human data:

• Learning rate: Controls how quickly associations between features and categories are formed and modified over time.

• Trial impatience: Determines the threshold at which a participant will make a category response within a trial.

• Fixation impatience: Governs how readily a participant will shift their gaze from one location to another during stimulus processing .
  With adjustment of just these three parameters, LAG-1 can produce qualitatively accurate fits for numerous behavioral measures including:

• Learning curves across trials

• Response timing metrics

• Eye movement patterns

• Fixation order preferences

• Changing patterns of visual attention during feedback phases
  This parsimony in parameterization suggests that the model captures fundamental dynamics in how learning shapes attention and gaze behavior in humans.

How effective are Lag schedules in promoting response variability in individuals with ASD?

Research demonstrates that Lag schedules are effective interventions for promoting response variability in individuals with ASD, though effectiveness varies across participants and implementation approaches:
Effectiveness Data from Research Studies:

• Appropriate error correction procedures are implemented

• Visual supports are provided

• The lag criterion is gradually increased as performance improves

• Multiple target stimuli (e.g., varied social questions) are used to prevent narrow stimulus control

What methodological considerations are important when designing Lag schedule interventions for clinical populations?

When designing Lag schedule interventions for clinical populations, several methodological considerations are critical:

• Baseline assessment: Thoroughly assess the participant's current response patterns, noting specific areas of invariability, stereotypy, or inappropriate responding.

• Appropriate Lag level selection:

  • Match the Lag level to the participant's current abilities

  • Consider starting with Lag 1 before progressing to higher levels

  • Increase the Lag requirement gradually based on performance data

• Stimulus variation:

  • Use multiple exemplars of similar questions/prompts

  • Present questions in randomized order to prevent order effects

  • Ensure the discriminative stimuli (S^D) vary sufficiently to prevent narrow stimulus control

• Reinforcement systems:

  • Identify effective reinforcers through preference assessments

  • Implement token economy systems for older participants

  • Provide immediate reinforcement when lag criteria are met

• Generalization planning:

  • Include generalization probes across settings

  • Plan for generalization across people

  • Test maintenance of acquired skills post-intervention

• Data collection system:

  • Track both correct responses and types of errors

  • Monitor novel versus prompted responses

  • Document spontaneous responses separate from trained responses
    The study with Belle employed a changing-criterion design with interspersed generalization probes and a maintenance probe, which allowed for systematic evaluation of intervention effects while also assessing whether skills generalized beyond the immediate training context .

How does the LAG-1 model predict changes in human eye movement patterns during category learning?

The LAG-1 model makes specific predictions about how eye movement patterns change during category learning based on its integrated architecture:

• Early learning phase predictions:

  • Initially high fixation counts across all features

  • Shorter fixation durations as participants explore widely

  • Less consistent fixation order patterns

  • Frequent returns to previously fixated locations

• Late learning phase predictions:

  • Decreased fixation counts, focused on diagnostic features

  • Longer fixations on informative regions

  • More consistent, optimized fixation order patterns

  • Decreased fixation counts during feedback phases as learning progresses
    The model predicts these changes by simulating how associative learning gradually shapes the attentional priority map. As associations between features and categories strengthen, the model extracts information about feature diagnosticity, which then biases attention and gaze toward more informative features .
    Critically, LAG-1 produces these predictions without requiring explicit instruction about which features are important. Instead, the model learns which features are diagnostic through experience and adjusts attentional allocation accordingly, mimicking the process observed in human learners .

What are the neurophysiological correlates of the LAG-1 model components?

The LAG-1 model components have specific neurophysiological correlates that align with known brain systems:

• Visuospatial attention system:

  • Posterior parietal cortex: Contains priority maps that integrate bottom-up and top-down signals

  • Lateral intraparietal area (LIP): Maintains spatial representations of attentional priority

  • Visual cortex: Shows enhanced processing for attended locations/features

  • Frontal eye fields: Involved in voluntary attentional control

• Saccadic timing and control system:

  • Superior colliculus: Critical for saccade initiation and targeting

  • Basal ganglia: Involved in action selection and timing

  • Cerebellum: Contributes to saccade accuracy and adaptation

  • Brainstem saccade generators: Execute eye movement commands

• Category learning system:

  • Medial temporal lobe: Supports declarative aspects of category learning

  • Striatum: Involved in procedural category learning

  • Prefrontal cortex: Maintains task rules and category representations

  • Anterior cingulate: Monitors performance and learning progress
    The neural field approach employed by LAG-1 is particularly suited to modeling these brain systems because it captures the continuous, dynamic interactions between populations of neurons that characterize real neural systems. The model's reentrant signals between learning and attention systems mirror the extensive recurrent connectivity observed between brain regions involved in these cognitive processes .

What methodological improvements could enhance Lag schedule interventions for promoting response variability?

Several methodological improvements could enhance the effectiveness of Lag schedule interventions:

• Technology integration:

  • Automated response tracking systems

  • Computer-based presentation of stimuli with standardized timing

  • Mobile applications for extending practice beyond clinical settings

• Advanced schedule designs:

  • Adaptive Lag requirements that adjust based on performance

  • Combining Lag schedules with other reinforcement schedules

  • Progressive time-delay procedures to fade prompts systematically

• Measurement refinements:

  • More precise definitions of response variability

  • Development of standardized variability metrics

  • Techniques for measuring semantic rather than just topographical variation

• Generalization enhancement:

  • Structured procedures for programming generalization

  • Matrix training approaches to promote recombination of response elements

  • Peer-mediated intervention components

• Multiple response domain training:

  • Simultaneous targeting of verbal and non-verbal variability

  • Integrating Lag schedules into natural conversation practice

  • Combining variability training with pragmatic language interventions
    Future research should address limitations in current approaches, such as the limited variety of social questions used in many studies and the need for more robust generalization and maintenance assessment .

How might the LAG-1 model be extended to understand atypical attention patterns in clinical populations?

The LAG-1 model offers promising opportunities for extension to understand atypical attention patterns in clinical populations:

• Parameter modulation for clinical simulation:

  • Adjusting learning rates to model slower acquisition in developmental disorders

  • Modifying fixation impatience parameters to simulate attention deficits

  • Altering trial impatience to capture impulsivity or perseveration

• Structural modifications:

  • Adding noise to attentional priority maps to model distractibility

  • Implementing biases in feature processing to capture atypical sensory preferences

  • Modifying the strength of connections between model components to represent atypical neural connectivity

• Clinical applications:

  • Predicting individual differences in learning trajectories

  • Identifying optimal intervention targets based on model simulations

  • Developing personalized training regimens by fitting the model to individual data

• Integrating with other methodologies:

  • Combining LAG-1 with neuroimaging data to validate model components

  • Using model predictions to guide eye-tracking assessments of attention

  • Developing computational biomarkers for attention-related disorders Such extensions could help bridge the gap between computational neuroscience and clinical practice by providing mechanistic explanations for observed attentional phenomena in conditions such as autism spectrum disorder, attention-deficit/hyperactivity disorder, and learning disabilities.