Abstract:

Summaries of technical and scientific papers are major means to address the problem of information overload in the modern world. However, the quality of summaries does not always live up to high standards and requirements to this type of technical documents, which makes training in summarization one of the key issues in knowledge dissemination and assimilation. Effective summarizing is one of the most difficult to teach and, as research shows, it is one of the problematic strategies to master.

In this paper we present a summarization model that can be used in teaching to produce high quality summaries of technical and scientific papers. The model includes summarization algorithms developed based on the detailed analysis of the instruction requirements to technical summaries and technology tools that automate certain steps in the summarizing procedure.

A correct summary should:
• have 4 information parts: theme, purpose, method and result
• contain only relevant information
• be grammatically well-formed in such a way that it could facilitate its further automatic processing, e.g., machine translation; this is a specific feature of our model
• meet size constraints

The model guides trainees through a step-by-step summarization process such as:
a) document analysis, that includes automatic keyword extraction with a special in-house developed tool, automatic morphosyntactic analysis with an on-line tagger, and interactive semantic labeling against the model knowledge base;
b) identification of the summary content based on the formally calculated weights of document text fragments and templates from the model knowledge base, and
c) generating the summary text according to specially constructed and strictly imposed rules to have the summary text meeting all requirements to this type of a document and, (which is a special novelty of our approach) having a grammatical structure that lifts problems in machine translation.

The model is realized on the material of the mathematical modeling domain in the Russian language, but it can be ported to other domains and languages.

Keywords:

Training, summarization model, technology tools, keyword extraction.