My Account Log in

1 option

A Unified Concept Model for Advancing Multilingual Summarization and Semantic Reasoning in the Automotive Space Mercedes-Benz R&D Pvt., Limited

SAE Technical Papers (1906-current) Available online

View online
Format:
Book
Conference/Event
Author/Creator:
Singh, Samagra, author.
Contributor:
Awasthi PhD, Anshuman
Ravi, Utkarsh
Shenoy, Lakshmi
Vikram, Prateek
Conference Name:
Symposium on International Automotive Technology (2026) (2026-01-28 : Pune, India)
Language:
English
Physical Description:
1 online resource cm
Place of Publication:
Warrendale, PA SAE International 2026
Summary:
The automotive industry produces a vast amount of multilingual textual data ranging from technical manuals to diagnostic reports that demand efficient summarization and reliable semantic reasoning. At present, the traditional large language models (LLMs) operating at the token level struggle not only with cross-lingual understanding and domain-specific reasoning but also are prone to hallucinations, leading to inaccurate insights and responses [2, 5]. This paper introduces a Unified Concept Model (UCM) architecture for the automotive domain that processes language at the concept level using multilingual, modality-agnostic embeddings, enabling coherent cross-lingual summarization and reasoning. The UCM encodes entire sentences as semantic vectors by leveraging the SONAR embedding space, a multilingual, modality-agnostic sentence representation that supports over 200 languages. This approach to encoding facilitates a deeper understanding across language boundaries and complex technical and legal issues. An LCM-inspired concept transformer then performs reasoning over these embeddings, and a GPT-style decoder reconstructs fluent summaries or explanations in the desired language. Evaluated on diverse automotive datasets in over 20 languages, UCM outperformed token-level baselines, achieving ROUGE-L scores of 88% (+16% over LCM) and reducing hallucination rates to 4%. These results demonstrate UCM's potential for scalable, accurate, and domain-specific AI systems in the automotive sector while enabling cross-lingual semantic reasoning beyond the capabilities of conventional LLMs. Furthermore, the paper briefly contextualizes UCM within the broader landscape of emerging AI models beyond LLMs, such as Large Knowledge Models and Large Reasoning Models, and discusses the problems and future directions for advancing concept-driven AI systems.
Notes:
Vendor supplied data
Publisher Number:
2026-26-0676
Access Restriction:
Restricted for use by site license

The Penn Libraries is committed to describing library materials using current, accurate, and responsible language. If you discover outdated or inaccurate language, please fill out this feedback form to report it and suggest alternative language.

Find

Home Release notes

My Account

Shelf Request an item Bookmarks Fines and fees Settings

Guides

Using the Find catalog Using Articles+ Using your account