Resources

Demos of LT applications

  • Syntactic Sentence Simplifier

    This demo implements the iterative approach to sentence simplification described in Evans (2020). It uses BERT models finetuned using the data presented in my thesis and additional annotated datasets to identify complex NPs and compound constituents (clauses, verb phrases, and noun phrases). This information is used to split complex sentences into sequences of simpler sentences.

  • Syntactic Complexity Sign tagger

    This finetuned BERT model is a more accurate version of the CRF model presented in Dornescu et al. (2013) to classify signs of syntactic complexity. In this setting, signs of syntactic complexity are a predefined set of conjunctions, relative clauses, and punctuation marks. The tagger classifies them in accordance with the scheme described by Evans and Orasan (2013). There are three broad groups of signs: coordinators, left boudnaries of subordinate clauses, and right boundaries of subordinate clauses. The classification scheme is fine-grained, with numerous sub-classes of each of these three broad groups. The classifier was trained on the annotated corpus listed below (see signs of syntactic complexity). Sign tagging is an essential part of the sentence analysis exploited in our approach to sentence rewriting (Evans et al., 2014).

Public Google Colab Notebook

Sign Tagger

Signs are lexical-punctuational markers of syntactic complexity. The sign tagger identifies the coordinating or bounding functions of items such as punctuation marks (,, ;, :), coordinators (and, but, or), complementisers (that), relative and wh- pronouns (what, when, where, which, while, who), and bigrams consisting of a punctuation mark followed by one of the lexical signs of syntactic complexity (, and, ; but, , which, , who, , when, etc.) This notebook allows you to finetune a BERT model to classify signs of syntactic complexity with respect to an annotation scheme based on that presented in Evans and Orasan (2013) and Chapter 2 of Evans (2020).

Annotated corpora

  • Signs of syntactic complexity

    This resource comprises three collections of text from the genres/domains of news, patient healthcare information, and literature. In each case, a subset of signs of syntactic complexity has been annotated with information about their syntactic linking and bounding functions. The subset includes conjunctions, complementisers, wh-words, and punctuation marks. From this page, you can access the annotation scheme, the annotation guidelines, and the corpus.

  • Front page

    General information about me.

  • Research projects

    Information about my previous and current research activities.

  • Publications

    Bibliographic information and electronic versions of my research papers and technical reports.