Conventional and associative memory approaches to automatic spelling correction

Vladimir S Cherkassky, Nikolaos Vassilas, Gregory L. Brodt, Harry Wechsler

Research output: Contribution to journalArticlepeer-review

4 Scopus citations


This paper reviews conventional and emerging associative-memory approaches to robust data retrieval when the input keyword and/or database itself may contain noise (errors). Spelling checking is used as a primary example to illustrate various approaches and to contrast the difference between conventional (algorithmic) techniques and research methods based on neural associative memories. Based on several recent ad hoc models for associative spelling checkers a generic model is proposed that incorporates powerful N-gram encoding for word representation and supervised-learning associative memories. Recent research on associative spelling checkers is summarized and some original results are presented. It is concluded that many neural network models do not provide a practically viable solution for robust data retrieval, due to saturation and scaling problems. However, a combination of conventional and neural approaches is shown to have excellent error correction rates and low computational costs; hence, it can be a good choice for robust data retrieval in large databases.

Original languageEnglish (US)
Pages (from-to)223-237
Number of pages15
JournalEngineering Applications of Artificial Intelligence
Issue number3
StatePublished - May 1992


  • Associative memory
  • neural networks
  • robust data retrieval
  • spelling checkers
  • string matching


Dive into the research topics of 'Conventional and associative memory approaches to automatic spelling correction'. Together they form a unique fingerprint.

Cite this