Explainable Prediction of Text Complexity: The Missing Preliminaries for Text Simplification

Abstract

Text simplification reduces the language complexity of professional content for accessibility purposes. End-to-end neural network models have been widely adopted to directly generate the simplified version of input text, usually functioning as a blackbox. We show that text simplification can be decomposed into a compact pipeline of tasks to ensure the transparency and explainability of the process. The first two steps in this pipeline are often neglected: 1) to predict whether a given piece of text needs to be simplified, and 2) if yes, to identify complex parts of the text. The two tasks can be solved separately using either lexical or deep learning methods, or solved jointly. Simply applying explainable complexity prediction as a preliminary step, the out-ofsample text simplification performance of the state-of-the-art, black-box simplification models can be improved by a large margin.

Controllable Sentence Simplification via Operation Classification

“…Several existing works have attempted to use a classifer to determine which rewrite operation should be performed on an input at the sentencelevel. Applying a sentence-level binary classifier as an initial step to predict whether simplification should be performed has been found to yield improved SARI results, reducing conservatism and spurious transformations Garbacea et al, 2021).…”

“…Accuracy on the silver test set (98%) is much higher than previous works: Scarton and Specia (2018) and achieve mean accuracies of 51% and 70% for a similar 4class task. Garbacea et al (2021), who only train a binary (simp, no-simp) classifier achieve 81% accuracy.…”

“…In response, attempts have been made to produce controllable simplification systems that can constrain either the shape (length, amount of paraphrasing, lexical and syntactic complexity) of the output or the type of transformation to be applied (e.g., copy, split, merge, rewrite, etc.) (Scarton and Specia, 2018;Dong et al, 2019;Garbacea et al, 2021;Maddela et al, 2021).…”

Unsupervised Sentence Simplification via Dependency Parsing

“…Previous studies criticize existing systems for being opaque, suboptimal and semantically compromising (Garbacea et al, 2021;Maddela et al, 2021;Stajner, 2021). and DMLMTL (Guo et al, 2018), taken from (Garbacea et al, 2021).…”

Clear, easy, plain, and simple as keywords for text simplification

“…These are semantically close notions that are sometimes employed interchangeably and may generate ambiguity, for example, Naderi et al (2019) use the phrase plain language to refer to language used for people with disabilities and differentiate it from easy language, which they use for language for people with generic reading difficulties, contrary to Baumert (2016) and Maaß (2020). Moreover, even the current conceptualization of text simplification does not satisfy some specialists (Garbacea et al, 2021).…”