From Large Language Model Predicates to Logic Tensor Networks: Neurosymbolic Offer Validation in Regulated Procurement

Cedric Haufe; Frieder Stolzenburg

cs.AIApr 7, 2026

From Large Language Model Predicates to Logic Tensor Networks: Neurosymbolic Offer Validation in Regulated Procurement

Authors: Cedric Haufe, Frieder Stolzenburg

Organizations: Harz University of Applied Sciences, Wernigerode, Germany · Merseburg University of Applied Sciences, Merseburg, Germany · Senior Visiting Fellow, University of New South Wales (UNSW), School of Computer Science and Engineering, Sydney, Australia

Abstract

We present a neurosymbolic approach, i.e. combine symbolic and subsymbolic artificial intelligence, to validating offer documents in regulated public institutions. We employ a language model to extract information and then aggregate it with an LTN (Logic Tensor Network) to make an auditable decision. In regulated public institutions, decisions must be made in a manner that is both factually correct and legally verifiable. Our neurosymbolic approach allows existing domain-specific knowledge to be linked to the semantic text understanding of language models. The decisions resulting from our pipeline can be justified by predicate values, rule truth values, and corresponding text passages. Our experiments on a real corpus show that the proposed pipeline achieves performance comparable to existing models, but its key advantage lies in its interpretability, modular predicate extraction, and explicit support for XAI (Explainable AI).

Figures & tables

Figure 1: Anonymised example of a valid offer document from the corpus.

Figure 2: Overview of the pipeline: Incoming potential offers are segmented, evaluated by an LLM, and then used by an LTN to make the final decision.

Symbol	Channel
$T_{c}$	Title_clear
$N_{c}$	Number_clear
$N_{p}$	Number_present
$V_{c}$	Validity_clear
$V_{v}$	Validity_vague
$R_{c}$	Reservation_clear

Table 1: The 11 channels derived from the 8 predicates. These 11 channels are used in the LTN decision layer described.

\mathrm{PosFeature}(d)=T_{c}(d)\vee N_{p}(d)\vee V_{v}(d)\vee R_{c}(d)\vee P_{p}(d)\vee D_{p}(d)\vee S_{p}(d).

(1)

Table 2: The following rules are formulated from the 11 derived channels.

Pipeline	Fuzzy-Logic	F 1 (pos.)
LTN	Łukasiewicz	$0.899\pm 0.064$
CISC	Łukasiewicz	$0.782\pm 0.060$
MCSR-BestConf	Product	$0.874\pm 0.043$
MCSR-TopProb	Łukasiewicz	$0.849\pm 0.049$
BERT	–	$0.859\pm 0.097$
LLM (BM25+Semantic-CE) - Qwen2.5:14b-instruct	–	$0.843\pm 0.059$

Table 3: Post-hoc average test performance of the approaches considered over 25 test folds (5 stratified splits

\times

5-fold cross-validation,

N=200

, 35 % valid offers).

Fuzzy-Logic	Precision	Recall	F 1	Accuracy
LTN
Gödel	0.837	0.960	$0.892\pm 0.055$	0.916
Łukasiewicz	0.862	0.946	$\boldsymbol{0.899\pm 0.064}$	0.924
Product	0.863	0.943	$0.898\pm 0.058$	0.923
MCSR-BestConf
Gödel	0.768	0.908	$0.819\pm 0.064$	0.857

Table 4: Average test performance of the approaches considered over 25 test folds (5 stratified splits

\times

5-fold cross-validation,

N=200

, 35 % valid offers). The F 1 score of the positive class is given (mean

\pm

standard deviation).

Logic Tensor Networks

See all 9 papers

Neurosymbolic Language Reasoning as Satisfiability Modulo Theory
Feb 20, 2026Hyunseok Oh, Sam Stern, Youngki Lee +1Constraint-Guided Reasoning in Language Models Neuro-Symbolic Reasoning
Seoul National University · U. Mass. Amherst · Microsoft
Delta1 with LLM: symbolic and neural integration for credible and explainable reasoning
Mar 13, 2026Yang Xu, Jun Liu, Shuwei Chen +2Neuro-Symbolic Reasoning Explainable Artificial Intelligence
School of Mathematics, Southwest Jiaotong University, Chengdu 611756, China · National–Local Joint Engineering Laboratory of System Credibility Automatic Verification, Southwest Jiaotong University, Chengdu 611756, Sichuan, China · School of Computing, Ulster University, Belfast, Northern Ireland, UK
A Neurosymbolic Approach to Natural Language Formalization and Verification
Nov 12, 2025Chenyang An, Sam Bayless, Stefano Buliani +27Formal Verification LLM Safety
Amazon Web Services · University College London · Queen Mary University of London +1

Feb 20, 2026J/K move · Enter open · S save

Neurosymbolic Language Reasoning as Satisfiability Modulo Theory

Hyunseok Oh, Sam Stern, Youngki Lee, Matthai Philipose

Seoul National University · U. Mass. Amherst · Microsoft

Constraint-Guided Reasoning in Language Models Neuro-Symbolic Reasoning Symbolic Reasoning in Language Models Satisfiability Modulo Theories

Open paper AskPDF arXiv

Natural language understanding requires interleaving textual and logical reasoning, yet large language models often fail to perform such reasoning reliably. Existing neurosymbolic systems combine LLMs with solvers but remain limited to fully formalizable tasks such as math or program synthesis, leaving natural documents with only partial logical structure unaddressed. We introduce Logitext, a neurosymbolic language that represents documents as natural language text constraints (NLTCs), making partial logical structure explicit. We develop an algorithm that integrates LLM-based constraint evaluation with satisfiability modulo theory (SMT) solving, enabling joint textual-logical reasoning. Experiments on a new content moderation benchmark, together with LegalBench and Super-Natural Instructions, show that Logitext improves both accuracy and coverage. This work is the first that treats LLM-based reasoning as an SMT theory, extending neurosymbolic methods beyond fully formalizable domains.

Abstract

Logic Tensor Networks

Explore similar work

Neurosymbolic Language Reasoning as Satisfiability Modulo Theory

Delta1 with LLM: symbolic and neural integration for credible and explainable reasoning

A Neurosymbolic Approach to Natural Language Formalization and Verification