Generating Realistic, Protocol-Compliant Maritime Radio Dialogues using Self-Instruct and Low-Rank Adaptation

Gürsel Akdeniz; Emin Cagatay Nakilcioglu

cs.CLFeb 16, 2026

Generating Realistic, Protocol-Compliant Maritime Radio Dialogues using Self-Instruct and Low-Rank Adaptation

Authors: Gürsel Akdeniz, Emin Cagatay Nakilcioglu

Organizations: Fraunhofer Center for Maritime Logistics and Services (CML), Hamburg, Germany

Abstract

VHF radio miscommunication remains a major safety risk in maritime operations, with human factors accounting for over 58% of recorded incidents in Europe between 2014 and 2023. Despite decades of operational use, VHF radio communications are still prone to noise, interference, linguistic variability, and the absence of real-time transcription, making procedural errors both frequent and difficult to correct. Developing AI-assisted systems to support real-time communication and decision-making requires a considerable amount of high-quality maritime data, yet operational, regulatory, and privacy constraints render such datasets scarce. This study introduces a compliance aware Self-Instruct methodology for generating realistic maritime radio dialogues that conform to the IMO's SMCP. Our approach integrates a 26-filter verification pipeline directly into the iterative generation loop to enforce entity information accuracy, hallucination detection, SMCP-compliance, logical consistency, and linguistic diversity. We employ LORA for parameter-efficient fine-tuning, reducing computational overhead during training and enabling efficient deployment of the resulting models on resource-constrained maritime systems. To assess dataset quality, we introduce a novel evaluation framework combining automated and expert assessments: Format Accuracy, Information Accuracy, Uniqueness, and Logical Coherence. Experiments using publicly available vessel, coastal and AIS datasets demonstrate that the approach produces synthetically diverse, procedurally compliant, and operationally realistic dialogues. Although downstream applications such as automatic speech recognition and natural language processing are reserved for future work, the released code, datasets, and verification tools provide a reproducible foundation for artificial intelligence-assisted maritime safety and other safety-critical domains.

Figures & tables

Figure 1: The Realistic Maritime Distress Call Generation Pipeline

Filter	Coefficient
Parentheses	1
Bracket	1
Mayday	1
Incomplete	2
Vessel Name Not After Mayday	1
Duplicate Sentences	2

Table 3: Coefficients for Format Accuracy

Filter	Coefficient
Wrong Category	2
Unknown	1
Vessel Name	2
Vessel MMSI	2
Vessel Call Sign	2
Vessel Type	2

Table 4: Coefficients for Information Accuracy

Coherence Level	Description	Points
Excellent	Fully coherent, logically structured, all relevant info	5
Good	Mostly coherent, minor logical gaps	4
Average	Noticeable inconsistencies	3
Poor	Significant logical gaps	2
Very Poor	Incoherent and confusing, fails to convey a clear message	1

Table 5: Logical Coherence Levels and Points

Temperature	0.9
top_p	0.9
top_k	400
max_new_tokens	400

Table 6: LLM Hyperparameters During the Synthetic Instance Generation

r	256
Epochs	10
Learning rate	2e-4
Optimizer	AdamW
Scheduler	Linear Scheduler
Warm-up Steps	30
Batch size	2

Table 7: LoRA Configuration

Disaster Category	Format Accuracy	Information Accuracy	Unique- ness	Logical Coherence*	Valid Chatters
Armed Attack, Piracy	0.623	0.751	0.418	0.10	0
Collision	0.581	0.742	0.528	0.11	0
Disabled, Adrift	0.635	0.765	0.423	0.11	0
Fire, Explosion	0.630	0.733	0.472	0.10	0
Flooding	0.609	0.759	0.479	0.10	0
Grounding	0.612	0.733	0.427	0.10	0

Table 9: Performance of the Vanilla LLM Prompted with the Context-only Prompt. * Logical Coherence evaluations are conducted on 10 randomly sampled chatters generated by each adapter.

Disaster Category	Number of Rejected Chatters	Percentage of Successful Attempts
Armed Attack, Piracy	1243	28.68
Collision	1482	25.22
Disabled, Adrift	445	52.91
Fire, Explosion	1548	24.41
Flooding	857	36.84
Grounding	371	57.40

Table 10: Number of Rejected Chatters in the Instance Generation Pipeline per Distress Category. 500 valid chatters were generated per category.

Disaster Category	Format Accuracy	Information Accuracy	Unique- ness	Logical Coherence	Valid Chatters
Armed Attack, Piracy	0.981	0.925	0.533	0.84	89
Collision	0.998	0.912	0.548	0.9	92
Disabled, Adrift	0.998	0.928	0.612	0.80	93
Fire, Explosion	0.978	0.962	0.443	0.78	86
Flooding	0.988	0.918	0.405	0.80	89
Grounding	0.986	0.971	0.421	0.70	83

Table 11: Performance of the LoRA Adapters

Disaster Category	Number of Unique Chatters
Armed Attack, Piracy	89
Collision	94
Disabled, Adrift	94
Fire, Explosion	97
Flooding	97
Grounding	93

Table 12: Number of Unique Chatters Generated by each LoRA Adapter

Panel of Figure 2 — Figure 2: Rouge-L Similarities of chatters generated with the LoRA adapter "Fire, Explosion"

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

SMCP Category	Instruction
Fire, explosion	Generate a maritime radio chatter. A vessel makes a distress call and reports a fire.
Flooding	Generate a maritime radio chatter. A vessel makes a distress call and reports flooding.
Collision	Generate a maritime radio chatter. A vessel makes a distress call and reports collision.
Grounding	Generate a maritime radio chatter. A vessel makes a distress call and reports grounding.
List, Danger of Capsizing	Generate a maritime radio chatter. A vessel makes a distress call and reports list-danger of capsizing.
Sinking	Generate a maritime radio chatter. A vessel makes a distress call and reports sinking.

Appendix

Table 13: Instructions in the Seed Instances

Disaster Category	Keywords
Fire, explosion	fire, explosion
Flooding	flood, taking on water, taken on water, took on water
Collision	collide, collision
Grounding	grounding, grounded, aground
List, Danger of Capsizing	list, danger of capsiz
Sinking	sink

Appendix

Table 16: Keywords Searched in the Distress Call by the Filter "Wrong Category"

Synthetic Data Generation for Language Models

Synthetic data generation for language models creates artificial text, tasks, and interaction traces to train or evaluate models.

See all 293 papers

Mar 23, 2026

SeaAlert: Robust Severity Classification and LLM-Based Information Extraction for Noisy Maritime Distress Communications

Maritime distress communications transmitted over very high frequency (VHF) radio are safety-critical voice messages used to report emergencies at sea. Under the Global Maritime Distress and Safety System (GMDSS), such messages follow standardized procedures and are expected to convey essential details, including vessel identity, position, nature of the distress, and required assistance. In practice, however, automatic analysis remains difficult because distress messages are often brief, noisy, and produced under stress, may deviate from the prescribed format, and are further degraded by automatic speech recognition (ASR) errors caused by channel noise and speaker stress. This paper presents SeaAlert, a controlled experimental framework for evaluating robust analysis of maritime distress communications using transformer-based severity classification and LLM-based structured extraction. To address the scarcity of labeled real-world data, we develop a synthetic data generation pipeline in which an LLM produces diverse maritime messages, including challenging variants in which standard distress codewords are omitted or replaced with less explicit expressions. The generated utterances are synthesized into speech, degraded with simulated VHF noise, and transcribed by an ASR system to obtain controlled noise-degraded transcripts. The resulting evaluation shows that transformer-based classification degrades more gracefully than lexical baselines under ASR noise and codeword masking, while LLM-based extraction is more effective than Regex-based extraction for noisy structured fields.

Tomer Atia, Yehudit Aperstein, Alexander Apartsin

School of Computer Science, Faculty of Sciences, HIT-Holon Institute of Technology, Holon 58102, Israel · Intelligent Systems, Afeka Academic College of Engineering, Tel Aviv 69988, Israel

LLM Information Extraction Language Model Robustness

Ask PDF

Oct 19, 2026

Speech LLMs That Breathe: Modeling Non-Verbal Vocalizations in Dialogue

The user experience of speech interaction models has improved, benefiting from the modeling of paralinguistic attributes and advances in speech reconstruction paradigms. However, human communication follows extensive norms regarding non-verbal vocalizations (NVVs) spanning respiratory, affective, and discourse. The omission of these events imposes inherent limitations on the naturalness of current approaches. In this work, we cover 13 major types of NVVs and construct a natural dialogue dataset that conforms to human conversational norms through an automated synthesis pipeline. We enhance the ability of speech interaction models to generate contextually appropriate NVV events through multi-stage fine-tuning and NVV-aware GRPO (Group Relative Policy Optimization). Based on distinct modeling paradigms, we substantially improve the expressiveness of VocalNet and VITA-Qinyu in empathy interactions and role-play dialogues, and further discuss the architectural necessity for achieving high-quality natural speech generation.

Heyang Liu, Lixin Zhang, Jiayi Huang +6

Shanghai Jiao Tong University · Ant Group

Nonverbal Vocalization Generation Speech Language Models Spoken Dialogue Systems Audio Generation Expressive Speech Synthesis

Ask PDF

Apr 12, 2026

VeriSim: A Configurable Framework for Stress-Testing Medical AI Under Patient Communication Noise

Medical large language models are typically evaluated on idealized patient cases that do not reflect how real patients communicate. We introduce VeriSim, a patient simulation framework that injects controllable noise along six clinically grounded communication dimensions while substantially preserving each patient's medical record. Truth adherence is supported by a verifier that extracts atomic claims from each candidate utterance and judges them against a UMLS-grounded vector index built with BioLORD embeddings, using the retrieved atoms' structured clinical metadata (e.g., drug class, anatomical site, treats-condition relations) rather than surface-text similarity alone. Across seven open-weight LLMs, realistic noise reduces diagnostic accuracy by 15-25 percentage points and increases conversation length by 34-55%; the 7-8B models degrade 1.4x more than 70B+ models. A board-certified physician and a licensed nurse rate VeriSim's conversations highly on truth, realism, clinical utility, and noise fidelity (inter-annotator agreement >= 0.80 across all dimensions), and an LLM-as-judge closely tracks their ratings, enabling scalable evaluation. We release VeriSim as an open-source framework for rigorous stress-testing of clinical AI: https://github.com/mohitmarvania/VeriSim

Sina Mansouri, Mohit Marvania, Vibhavari Ashok Shihorkar +5

Department of Computer Science, George Mason University · Department of Health Administration and Policy, College of Public Health, George Mason University · Department of Science, George Mason University

Healthcare LLM Reliability Clinical Language Model Evaluation Language Model Robustness Medical LLMs

Ask PDF

Abstract

Synthetic Data Generation for Language Models

Explore similar work

SeaAlert: Robust Severity Classification and LLM-Based Information Extraction for Noisy Maritime Distress Communications

Speech LLMs That Breathe: Modeling Non-Verbal Vocalizations in Dialogue

VeriSim: A Configurable Framework for Stress-Testing Medical AI Under Patient Communication Noise