Summary of Geometric Signatures Of Compositionality Across a Language Model’s Lifetime, by Jin Hwa Lee et al.
Geometric Signatures of Compositionality Across a Language Model’s Lifetime
by Jin Hwa Lee, Thomas Jiralerspong, Lei Yu, Yoshua Bengio, Emily Cheng
First submitted to arxiv on: 2 Oct 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Information Theory (cs.IT); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper investigates whether contemporary language models (LMs) reflect the intrinsic simplicity of language, enabled by compositionality. The authors take a geometric approach by relating the degree of dataset compositionality to the intrinsic dimension (ID) of representations under an LM. They find that the degree of compositionality is reflected in the ID, and that the relationship between compositionality and geometric complexity arises due to learned linguistic features during training. The study also reveals a striking contrast between nonlinear and linear dimensionality, showing they respectively encode semantic and superficial aspects of linguistic composition. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Language models can create many sentences using simple rules and a limited vocabulary. Scientists wondered if these models reflect the simplicity of language itself. Researchers studied how well language models capture this simplicity by analyzing their internal representations of text data. They found that the complexity of these representations is closely tied to the simplicity of the text, with learned linguistic features playing a key role. |
