[Paper Review] Cryptolysis v.0.0.1 - A Framework for Automated Cryptanalysis of Classical Ciphers

Serguei A. Mokhov, Marc-André Laverdière+2 more|arXiv (Cornell University)|Jan 5, 2011

graph theory and CDMA systems1 references4 citations

TL;DR

Cryptolysis v0.0.1 is a Java-based framework for automated cryptanalysis of classical ciphers, employing statistical methods such as n-gram language models and dictionary-based parsing to detect word boundaries and break substitution, shift, and SPN ciphers. It achieves effective cryptanalysis through heuristic search techniques like genetic algorithms and simulated annealing, with successful decryption of classical ciphers using minimal user input.

ABSTRACT

Cryptolysis is a framework that includes a collection of automated attacks on the classical ciphers based on the article by Clark Dawson on Optimisation heuristics for the automated cryptanalysis of classical ciphers from the Journal of Combinatorial Mathematics and Combinatorial Computing, 1998.

Motivation & Objective

• To automate the cryptanalysis of classical ciphers such as substitution, shift, and SPN ciphers.
• To improve readability of decrypted text by automatically detecting word boundaries using statistical and dictionary-based methods.
• To implement efficient, reusable cryptanalytic modules using object-oriented design and modular architecture.
• To support multiple cryptanalytic strategies, including genetic algorithms, simulated annealing, and tabu search.
• To enable training of language models and statistical parsers from corpora for use in cryptanalysis.

Proposed method

• The framework uses a modular architecture with distinct components for ciphers, analyzers, and parsers, each implementing standardized interfaces.
• Word boundary detection is performed via two methods: dictionary-based lookup of longest matching words and probabilistic 3-gram character models.
• The 3-gram model computes probabilities of word breaks based on trigram frequencies of preceding and following characters, including word boundaries.
• Cryptanalysis attacks are implemented using heuristic optimization: genetic algorithm, simulated annealing, and tabu search, all guided by fitness functions based on language statistics.
• Training involves building language statistics, dictionary parsers, and n-gram models from input corpora, which are serialized for later reuse.
• The system uses MARF (Multilingual Automatic Recognition Framework) for NLP components and Java’s portability and memory management for cross-platform deployment.

Experimental results

Research questions

• RQ1 Can automated word boundary detection be effectively applied to decrypted classical cipher texts to improve human readability?
• RQ2 How effective are statistical n-gram models compared to dictionary-based parsing in detecting word boundaries in undeciphered ciphertext streams?
• RQ3 To what extent can heuristic search methods like genetic algorithms and simulated annealing successfully break classical ciphers such as substitution and shift ciphers?
• RQ4 Can a modular, extensible framework be designed to support multiple classical ciphers and cryptanalytic techniques in a single unified system?
• RQ5 What are the performance and accuracy trade-offs between dictionary-based and n-gram-based word boundary detection in cryptanalysis?

Key findings

• The framework successfully implements and integrates multiple cryptanalytic techniques, including statistical, genetic, and simulated annealing attacks, for classical ciphers.
• The 3-gram character n-gram model achieved efficient and compact word boundary detection with lower storage overhead than dictionary-based methods.
• Dictionary-based parsing struggled with compound words and proper nouns, often misplacing spaces in ambiguous cases such as 'therefore' vs 'there for eternity'.
• The system demonstrated effective decryption of classical ciphers using heuristic search, with fitness functions based on language statistics improving convergence.
• Training of language models and parsers was successfully automated, with serialized outputs enabling reuse across sessions.
• The modular design allowed seamless integration of new cipher types and analysis methods through standardized interfaces.