NegAIT: A new parser for medical text simplification using morphological, sentential and double negation.

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Text Readability Analysis

Background:

• Text features significantly impact readability and comprehension.
• Negation is a key feature, yet tools for its detection and studies on its effects are limited.

Purpose of the Study:

• Introduce a novel negation parser (NegAIT) for detecting morphological, sentential, and double negation.
• Investigate negation usage across diverse corpora to understand its relationship with text difficulty.
• Assess the utility of negation features in predicting text readability using machine learning classifiers.

Main Methods:

• Developed and evaluated NegAIT against a human-annotated gold standard (500 Wikipedia sentences).
• Performed corpus statistical analysis comparing negation patterns in six text corpora (patient blogs, Cochrane reviews, PubMed abstracts, clinical trials, English Wikipedia, Simple English Wikipedia).
• Trained five binary classifiers (Naïve Bayes, SVM, decision tree, logistic regression, linear regression) using negation features to distinguish between easy and difficult texts.

Main Results:

• NegAIT achieved high precision (95%) and recall (100%) for morphological negation.
• Difficult texts exhibited lower overall negation frequency but significantly higher morphological negation (p<0.01).
• Classifiers using negation features outperformed the majority baseline, with Naïve Bayes achieving 77% accuracy.

Conclusions:

• The NegAIT parser effectively detects various negation types.
• Negation patterns differ significantly between easy and difficult texts, with morphological negation being a key differentiator.
• Negation features are valuable predictors of text readability, highlighting their importance in NLP applications.