Language Models “Grok” to Copy (2025.naacl-short)

Copied to clipboard

Challenge: We examine the pre-training dynamics of language models, focusing on their ability to copy text from preceding context.
Approach: They propose that Transformer-based language models develop copying abilities similarly to grokking . they argue that the connection between groking and context copying can improve in-context performance.
Outcome: The proposed model development is similar to grokking, but the speed is independent of tokens trained.

Similar Papers

On the In-context Generation of Language Models (2024.emnlp-main)

Copied to clipboard

Challenge: Large language models (LLMs) have the ability of in-context generation (ICG) when given an in-text prompt, they can implicitly recognize the pattern of the examples and complete the prompt in the desired way.
Approach: They propose a plausible latent variable model to model the distribution of pretrained corpora and formalize ICG as a problem of next topic prediction.
Outcome: The proposed model can model the distribution of pretrained corpora and then formalize ICG as a problem of next topic prediction.
What Context Features Can Transformer Language Models Use? (2021.acl-long)

Copied to clipboard

Challenge: Recent studies show that transformer-based language models benefit from conditioning on contexts of hundreds to thousands of previous tokens.
Approach: They propose to use lexical and structural information to ablate usable information in transformer language models.
Outcome: The proposed model improves when conditioning on contexts of thousands of previous tokens.
How Much Do Language Models Copy From Their Training Data? Evaluating Linguistic Novelty in Text Generation Using RAVEN (2023.tacl-1)

Copied to clipboard

Challenge: Current language models generate high-quality text, but are they copying it or have they learned generalizable linguistic abstractions?
Approach: They propose a suite of analyses for assessing the novelty of generated text . they focus on sequential structure (n-grams) and syntactic structure (syntactical structure).
Outcome: The proposed model-generated text is as novel as the baseline human-generated model- generated text, but it is copied substantially, the authors show .
Identifying Semantic Induction Heads to Understand In-Context Learning (2024.findings-acl)

Copied to clipboard

Challenge: Large language models (LLMs) have demonstrated remarkable performance, but lack of transparency in their inference logic raises concerns about their trustworthiness.
Approach: They conduct a detailed analysis of the operations of attention heads to understand their in-context learning of LLMs.
Outcome: The proposed analysis of attention heads reveals that they increase the output logits of object tokens and recall objects . the proposed model is a novel approach to understand the in-context learning of large language models.
Grokking of Hierarchical Structure in Vanilla Transformers (2023.acl-short)

Copied to clipboard

Challenge: a recent study has shown that neural sequence models like transformers can generalize hierarchically when training for extended periods.
Approach: They show that transformers can learn to generalize hierarchically after long training periods . they call this phenomenon structural grokking, which exhibits inverted U-shaped scaling in model depth .
Outcome: The proposed model generalizes better than both very deep and very shallow models on multiple datasets.
Investigating Transferability in Pretrained Language Models (2020.findings-emnlp)

Copied to clipboard

Challenge: Recent work on deep NLP models has centered on probing, a method that involves training classifiers for different tasks on model representations.
Approach: They propose a method for determining the impact of each pretrained layer on transfer task performance by ablation.
Outcome: The proposed method shows that pretraining models improve performance on downstream tasks . the results highlight the limitations of methods that operate on frozen models or single data samples.
Rethinking the Role of Scale for In-Context Learning: An Interpretability-based Case Study at 66 Billion Scale (2023.acl-long)

Copied to clipboard

Challenge: 70% of attention heads and 20% of the feed forward networks can be removed with minimal decline in task performance.
Approach: They propose to investigate whether in-context learning is not uniform across all components of a large language model.
Outcome: The proposed model can remove 70% of attention heads and 20% of feed forward networks with minimal decline in task performance.
Crosscoding Through Time: Tracking Emergence & Consolidation Of Linguistic Representations Throughout LLM Pretraining (2026.acl-long)

Copied to clipboard

Challenge: Large language models learn non-trivial abstractions during pretraining, but it is not well understood when and how these specific linguistic abilities emerge.
Approach: They propose a method to track the evolution of linguistic features during pretraining by using sparse crosscoders to discover and align features across model checkpoints.
Outcome: The proposed approach can detect features emergence, maintenance, and discontinuation during training stages.
Between Circuits and Chomsky: Pre-pretraining on Formal Languages Imparts Linguistic Biases (2025.acl-long)

Copied to clipboard

Challenge: Pretraining language models on formal language can improve their acquisition of natural language, but it is computationally limited.
Approach: They hypothesize that effective transfer occurs when the formal language captures hierarchical dependencies present in natural language and remains within the computational limitations of the model architecture.
Outcome: The proposed model can achieve lower loss on natural language and better generalization with a 33% smaller token budget than training on a matched amount of natural language.
ICLEval: Evaluating In-Context Learning Ability of Large Language Models (2025.coling-main)

Copied to clipboard

Challenge: Existing evaluation frameworks focus on language abilities and knowledge, often overlooking the assessment of ICL ability.
Approach: They propose to evaluate the ICL ability of Large Language Models (LLMs) using the ICLEval benchmark.
Outcome: The proposed benchmark demonstrates that ICL ability is universally present in different LLMs and model size is not the sole determinant of ICL efficacy.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations