Language models can learn implicit multi-hop reasoning, but only if they have lots of training data
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918321731928064 |
|---|---|
| author | Yao, Yuekun Du, Yupei Zhu, Dawei Hahn, Michael Koller, Alexander |
| author_facet | Yao, Yuekun Du, Yupei Zhu, Dawei Hahn, Michael Koller, Alexander |
| contents | Implicit reasoning is the ability of a language model to solve multi-hop reasoning tasks in a single forward pass, without chain of thought. We investigate this capability using GPT2-style language models trained from scratch on controlled $k$-hop reasoning datasets ($k = 2, 3, 4$). We show that while such models can indeed learn implicit $k$-hop reasoning, the required training data grows exponentially in $k$, and the required number of transformer layers grows linearly in $k$. We offer a theoretical explanation for why this depth growth is necessary. We further find that the data requirement can be mitigated, but not eliminated, through curriculum learning. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_17923 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Language models can learn implicit multi-hop reasoning, but only if they have lots of training data Yao, Yuekun Du, Yupei Zhu, Dawei Hahn, Michael Koller, Alexander Computation and Language Implicit reasoning is the ability of a language model to solve multi-hop reasoning tasks in a single forward pass, without chain of thought. We investigate this capability using GPT2-style language models trained from scratch on controlled $k$-hop reasoning datasets ($k = 2, 3, 4$). We show that while such models can indeed learn implicit $k$-hop reasoning, the required training data grows exponentially in $k$, and the required number of transformer layers grows linearly in $k$. We offer a theoretical explanation for why this depth growth is necessary. We further find that the data requirement can be mitigated, but not eliminated, through curriculum learning. |
| title | Language models can learn implicit multi-hop reasoning, but only if they have lots of training data |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2505.17923 |