Physics of Language Models: Part 1, Learning Hierarchical Language Structures

• URL: http://arxiv.org/abs/2305.13673v4
• Date: Mon, 19 May 2025 11:12:27 GMT
• Title: Physics of Language Models: Part 1, Learning Hierarchical Language Structures
• Authors: Zeyuan Allen-Zhu, Yuanzhi Li,
• Abstract summary: Transformer-based language models are effective but complex, and understanding their inner workings and reasoning mechanisms is a significant challenge.<n>We introduce a family of synthetic CFGs that produce hierarchical rules, capable of generating lengthy sentences.<n>We demonstrate that generative models like GPT can accurately learn and reason over CFG-defined hierarchies and generate sentences based on it.
• Score: 51.68385617116854
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Transformer-based language models are effective but complex, and understanding their inner workings and reasoning mechanisms is a significant challenge. Previous research has primarily explored how these models handle simple tasks like name copying or selection, and we extend this by investigating how these models perform recursive language structure reasoning defined by context-free grammars (CFGs). We introduce a family of synthetic CFGs that produce hierarchical rules, capable of generating lengthy sentences (e.g., hundreds of tokens) that are locally ambiguous and require dynamic programming to parse. Despite this complexity, we demonstrate that generative models like GPT can accurately learn and reason over CFG-defined hierarchies and generate sentences based on it. We explore the model's internals, revealing that its hidden states precisely capture the structure of CFGs, and its attention patterns resemble the information passing in a dynamic programming algorithm. This paper also presents several corollaries, including showing why absolute positional embeddings is inferior to relative and rotary embeddings; uniform attention alone is surprisingly effective (motivating our follow-up work on Canon layers); encoder-only models (e.g., BERT, DeBERTa) struggle with deep structure reasoning on CFGs compared to autoregressive models (e.g., GPT); and injecting structural or syntactic noise into pretraining data markedly improves robustness to corrupted language prompts.

Related papers

• Transformers Pretrained on Procedural Data Contain Modular Structures for Algorithmic Reasoning [40.84344912259233]
  We identify several beneficial forms of procedural data, together with specific algorithmic reasoning skills that improve in small transformers.<n>Our core finding is that different procedural rules instil distinct but complementary inductive structures in the model.<n>Most interestingly, the structures induced by multiple rules can be composed to jointly impart multiple capabilities.
  arXiv  Detail & Related papers  (2025-05-28T12:50:09Z)
• From Text to Graph: Leveraging Graph Neural Networks for Enhanced Explainability in NLP [3.864700176441583]
  This study proposes a novel methodology to achieve explainability in natural language processing tasks. It automatically converts sentences into graphs and maintains semantics through nodes and relations. Experiments delivered promising results in determining the most critical components within the text structure for a given classification.
  arXiv  Detail & Related papers  (2025-04-02T18:55:58Z)
• Exploring the Role of Reasoning Structures for Constructing Proofs in Multi-Step Natural Language Reasoning with Large Language Models [30.09120709652445]
  This paper is centred around a focused study: whether the current state-of-the-art generalist LLMs can leverage the structures in a few examples to better construct the proof structures with textitin-context learning.
  arXiv  Detail & Related papers  (2024-10-11T00:45:50Z)
• Boosting the Capabilities of Compact Models in Low-Data Contexts with Large Language Models and Retrieval-Augmented Generation [2.9921619703037274]
  We propose a retrieval augmented generation (RAG) framework backed by a large language model (LLM) to correct the output of a smaller model for the linguistic task of morphological glossing. We leverage linguistic information to make up for the lack of data and trainable parameters, while allowing for inputs from written descriptive grammars interpreted and distilled through an LLM. We show that a compact, RAG-supported model is highly effective in data-scarce settings, achieving a new state-of-the-art for this task and our target languages.
  arXiv  Detail & Related papers  (2024-10-01T04:20:14Z)
• Hidden Holes: topological aspects of language models [1.1172147007388977]
  We study the evolution of topological structure in GPT based large language models across depth and time during training. We show that the latter exhibit more topological complexity, with a distinct pattern of changes common to all natural languages but absent from synthetically generated data.
  arXiv  Detail & Related papers  (2024-06-09T14:25:09Z)
• In-Context Language Learning: Architectures and Algorithms [73.93205821154605]
  We study ICL through the lens of a new family of model problems we term in context language learning (ICLL) We evaluate a diverse set of neural sequence models on regular ICLL tasks.
  arXiv  Detail & Related papers  (2024-01-23T18:59:21Z)
• GPT Struct Me: Probing GPT Models on Narrative Entity Extraction [2.049592435988883]
  We evaluate the capabilities of two state-of-the-art language models -- GPT-3 and GPT-3.5 -- in the extraction of narrative entities. This study is conducted on the Text2Story Lusa dataset, a collection of 119 Portuguese news articles.
  arXiv  Detail & Related papers  (2023-11-24T16:19:04Z)
• "You Are An Expert Linguistic Annotator": Limits of LLMs as Analyzers of Abstract Meaning Representation [60.863629647985526]
  We examine the successes and limitations of the GPT-3, ChatGPT, and GPT-4 models in analysis of sentence meaning structure. We find that models can reliably reproduce the basic format of AMR, and can often capture core event, argument, and modifier structure. Overall, our findings indicate that these models out-of-the-box can capture aspects of semantic structure, but there remain key limitations in their ability to support fully accurate semantic analyses or parses.
  arXiv  Detail & Related papers  (2023-10-26T21:47:59Z)
• Parrot Mind: Towards Explaining the Complex Task Reasoning of Pretrained Large Language Models with Template-Content Structure [66.33623392497599]
  We show that a structure called template-content structure (T-C structure) can reduce the possible space from exponential level to linear level. We demonstrate that models can achieve task composition, further reducing the space needed to learn from linear to logarithmic.
  arXiv  Detail & Related papers  (2023-10-09T06:57:45Z)
• OntoType: Ontology-Guided and Pre-Trained Language Model Assisted Fine-Grained Entity Typing [25.516304052884397]
  Fine-grained entity typing (FET) assigns entities in text with context-sensitive, fine-grained semantic types. OntoType follows a type ontological structure, from coarse to fine, ensembles multiple PLM prompting results to generate a set of type candidates. Our experiments on the Ontonotes, FIGER, and NYT datasets demonstrate that our method outperforms the state-of-the-art zero-shot fine-grained entity typing methods.
  arXiv  Detail & Related papers  (2023-05-21T00:32:37Z)
• Unifying Structure Reasoning and Language Model Pre-training for Complex Reasoning [26.811507121199323]
  This paper proposes a unified learning framework that combines explicit structure reasoning and language pre-training to endow PLMs with the structure reasoning skill. It first identifies several elementary structures within contexts to construct structured queries and performs step-by-step reasoning along the queries to identify the answer entity. Experimental results on four datasets demonstrate that the proposed model achieves significant improvements in complex reasoning tasks involving diverse structures.
  arXiv  Detail & Related papers  (2023-01-21T08:18:11Z)
• Prompting Language Models for Linguistic Structure [73.11488464916668]
  We present a structured prompting approach for linguistic structured prediction tasks. We evaluate this approach on part-of-speech tagging, named entity recognition, and sentence chunking. We find that while PLMs contain significant prior knowledge of task labels due to task leakage into the pretraining corpus, structured prompting can also retrieve linguistic structure with arbitrary labels.
  arXiv  Detail & Related papers  (2022-11-15T01:13:39Z)
• Autoregressive Structured Prediction with Language Models [73.11519625765301]
  We describe an approach to model structures as sequences of actions in an autoregressive manner with PLMs. Our approach achieves the new state-of-the-art on all the structured prediction tasks we looked at.
  arXiv  Detail & Related papers  (2022-10-26T13:27:26Z)
• Benchmarking Language Models for Code Syntax Understanding [79.11525961219591]
  Pre-trained language models have demonstrated impressive performance in both natural language processing and program understanding. In this work, we perform the first thorough benchmarking of the state-of-the-art pre-trained models for identifying the syntactic structures of programs. Our findings point out key limitations of existing pre-training methods for programming languages, and suggest the importance of modeling code syntactic structures.
  arXiv  Detail & Related papers  (2022-10-26T04:47:18Z)
• Model Criticism for Long-Form Text Generation [113.13900836015122]
  We apply a statistical tool, model criticism in latent space, to evaluate the high-level structure of generated text. We perform experiments on three representative aspects of high-level discourse -- coherence, coreference, and topicality. We find that transformer-based language models are able to capture topical structures but have a harder time maintaining structural coherence or modeling coreference.
  arXiv  Detail & Related papers  (2022-10-16T04:35:58Z)
• Language Model Cascades [72.18809575261498]
  Repeated interactions at test-time with a single model, or the composition of multiple models together, further expands capabilities. Cases with control flow and dynamic structure require techniques from probabilistic programming. We formalize several existing techniques from this perspective, including scratchpads / chain of thought, verifiers, STaR, selection-inference, and tool use.
  arXiv  Detail & Related papers  (2022-07-21T07:35:18Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.