Grounded Token Initialization for New Vocabulary in LMs for Generative Recommendation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chen, Daiwei, Fu, Zhoutong, Jiang, Chengming, Zhang, Haichao, Zhou, Ran, Wang, Tan, Yao, Chunnan, Li, Guoyao, Cai, Rui, Cao, Yihan, Jiang, Ruijie, Borisyuk, Fedor, Shen, Jianqiang, Wu, Jingwei, Vinayak, Ramya Korlakai
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912999478919168
author Chen, Daiwei
Fu, Zhoutong
Jiang, Chengming
Zhang, Haichao
Zhou, Ran
Wang, Tan
Yao, Chunnan
Li, Guoyao
Cai, Rui
Cao, Yihan
Jiang, Ruijie
Borisyuk, Fedor
Shen, Jianqiang
Wu, Jingwei
Vinayak, Ramya Korlakai
author_facet Chen, Daiwei
Fu, Zhoutong
Jiang, Chengming
Zhang, Haichao
Zhou, Ran
Wang, Tan
Yao, Chunnan
Li, Guoyao
Cai, Rui
Cao, Yihan
Jiang, Ruijie
Borisyuk, Fedor
Shen, Jianqiang
Wu, Jingwei
Vinayak, Ramya Korlakai
contents Language models (LMs) are increasingly extended with new learnable vocabulary tokens for domain-specific tasks, such as Semantic-ID tokens in generative recommendation. The standard practice initializes these new tokens as the mean of existing vocabulary embeddings, then relies on supervised fine-tuning to learn their representations. We present a systematic analysis of this strategy: through spectral and geometric diagnostics, we show that mean initialization collapses all new tokens into a degenerate subspace, erasing inter-token distinctions that subsequent fine-tuning struggles to fully recover. These findings suggest that \emph{token initialization} is a key bottleneck when extending LMs with new vocabularies. Motivated by this diagnosis, we propose the \emph{Grounded Token Initialization Hypothesis}: linguistically grounding novel tokens in the pretrained embedding space before fine-tuning better enables the model to leverage its general-purpose knowledge for novel-token domains. We operationalize this hypothesis as GTI (Grounded Token Initialization), a lightweight grounding stage that, prior to fine-tuning, maps new tokens to distinct, semantically meaningful locations in the pretrained embedding space using only paired linguistic supervision. Despite its simplicity, GTI outperforms both mean initialization and existing auxiliary-task adaptation methods in the majority of evaluation settings across multiple generative recommendation benchmarks, including industry-scale and public datasets. Further analyses show that grounded embeddings produce richer inter-token structure that persists through fine-tuning, corroborating the hypothesis that initialization quality is a key bottleneck in vocabulary extension.
format Preprint
id arxiv_https___arxiv_org_abs_2604_02324
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Grounded Token Initialization for New Vocabulary in LMs for Generative Recommendation
Chen, Daiwei
Fu, Zhoutong
Jiang, Chengming
Zhang, Haichao
Zhou, Ran
Wang, Tan
Yao, Chunnan
Li, Guoyao
Cai, Rui
Cao, Yihan
Jiang, Ruijie
Borisyuk, Fedor
Shen, Jianqiang
Wu, Jingwei
Vinayak, Ramya Korlakai
Computation and Language
Artificial Intelligence
Machine Learning
Language models (LMs) are increasingly extended with new learnable vocabulary tokens for domain-specific tasks, such as Semantic-ID tokens in generative recommendation. The standard practice initializes these new tokens as the mean of existing vocabulary embeddings, then relies on supervised fine-tuning to learn their representations. We present a systematic analysis of this strategy: through spectral and geometric diagnostics, we show that mean initialization collapses all new tokens into a degenerate subspace, erasing inter-token distinctions that subsequent fine-tuning struggles to fully recover. These findings suggest that \emph{token initialization} is a key bottleneck when extending LMs with new vocabularies. Motivated by this diagnosis, we propose the \emph{Grounded Token Initialization Hypothesis}: linguistically grounding novel tokens in the pretrained embedding space before fine-tuning better enables the model to leverage its general-purpose knowledge for novel-token domains. We operationalize this hypothesis as GTI (Grounded Token Initialization), a lightweight grounding stage that, prior to fine-tuning, maps new tokens to distinct, semantically meaningful locations in the pretrained embedding space using only paired linguistic supervision. Despite its simplicity, GTI outperforms both mean initialization and existing auxiliary-task adaptation methods in the majority of evaluation settings across multiple generative recommendation benchmarks, including industry-scale and public datasets. Further analyses show that grounded embeddings produce richer inter-token structure that persists through fine-tuning, corroborating the hypothesis that initialization quality is a key bottleneck in vocabulary extension.
title Grounded Token Initialization for New Vocabulary in LMs for Generative Recommendation
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2604.02324