The repository of world knowledge is experiencing a substantial influx of textual data in natural language, surpassing the contribution of structured databases. The nature of unstructured text data, which is sparse and contains high feature dimensions, poses a non-trivial challenge for the anomaly detection task. Text anomalies refer to rare or unusual patterns of data hidden in a text dataset, making them difficult to identify. Various machine learning methods based on clustering and classification tasks have been suggested and documented in the existing literature to tackle this challenge, each with its own advantages and limitations. The deviation-based method, particularly the sequential exception technique, has shown astonishing performance in identifying anomalies in categorical datasets. However, this technique has not been tested on text data. In this study, we adapted the sequential exception technique to detect text anomalies by modifying the dissimilarity function of the technique. We evaluated the adapted technique on two text datasets: the ENRON email messages and the 20Newsgroup dataset. The experimental results illustrate the capability of the proposed method to successfully identify text anomalies, with an F-score of 78.1% for the ENRON dataset and 95% for the 20 Newsgroup dataset.