Un-Attributability: Computing Novelty From Retrieval & Semantic Similarity

2025-10-31

AI Summary

This work addresses the challenge of quantifying semantic novelty in language model outputs, proposing *unattributability*—the inability to semantically retrieve any pretraining corpus sample as the source of an output—as a formal, interpretable metric. Methodologically, it introduces a two-stage retrieval pipeline: first, efficient coarse-grained indexing using GIST embeddings; second, fine-grained re-ranking via ColBERTv2, with attribution thresholds calibrated against human-written text. The paper provides the first formal definition and empirical evaluation of unattributability. Experiments on the SmolLM family reveal three key findings: (1) instruction tuning significantly improves output unattributability; (2) increased reliance on longer contexts enhances semantic novelty; and (3) domain-specific characteristics shape the distribution of unattributable outputs. By grounding novelty assessment in semantic retrieval fidelity rather than surface-level heuristics, this work establishes a scalable, principled framework for evaluating generative originality—offering both interpretability and practical applicability for safety, copyright, and alignment research.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Computational Creativity

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Understanding how language-model outputs relate to the pretraining corpus is central to studying model behavior. Most training data attribution (TDA) methods ask which training examples causally influence a given output, often using leave-one-out tests. We invert the question: which outputs cannot be attributed to any pretraining example? We introduce un-attributability as an operational measure of semantic novelty: an output is novel if the pretraining corpus contains no semantically similar context. We approximate this with a simple two-stage retrieval pipeline: index the corpus with lightweight GIST embeddings, retrieve the top-n candidates, then rerank with ColBERTv2. If the nearest corpus item is less attributable than a human-generated text reference, we consider the output of the model as novel. We evaluate on SmolLM and SmolLM2 and report three findings: (1) models draw on pretraining data across much longer spans than previously reported; (2) some domains systematically promote or suppress novelty; and (3) instruction tuning not only alters style but also increases novelty. Reframing novelty assessment around un-attributability enables efficient analysis at pretraining scale. We release ~20 TB of corpus chunks and index artifacts to support replication and large-scale extension of our analysis at https://huggingface.co/datasets/stai-tuebingen/faiss-smollm

Problem

• Measuring semantic novelty in language model outputs
• Identifying unattributable outputs from pretraining corpus
• Assessing novelty via retrieval and similarity methods

Innovation

• Un-attributability measures semantic novelty via retrieval
• Two-stage pipeline uses GIST embeddings and ColBERTv2 reranking
• Compares model outputs to human text for novelty assessment