KH

Intro

According to (Suchanek and Weikum )... 以下　Therefore, in this paper のところまで

 * It is widely accepted that a combination of a predicate and arguments is the minimum unit of knowledge.
 * Assuming this nature, a domain knowledge has been manually represented as a set of numerous triples - two entities connected by a relation - known as ontologies.
 * In lifescience for instance, hunderds of ontologies have been constructed.
 * Moreover, further effors are made to expand such triples by automatically harvest triples from texts in new reports and web resources or by joining different sets of ontologies.
 * However, it should be noted that the domain knwoledge is always revised in a form of books written in natural language and books will be stayed as the primary and the most comprehensive source of domain knowledge as far as it is made by compilation of communications.
 * Neverthless, no attempts was made to completely transcribe a textbook into formal representation nor compare the content of two representations.
 * If transcription of knowledge - from natural language to formal representation - proceed without mutual references, we can not be objectively evaluate the progress of knowledge representation; we cannot tell what fractions of knowledge is transcribed in triples and what fraction is missing.
 * Or we can not evaluate the yield of knowledge extraction methods.
 * If a machine can breakdown a textual content of a book into smallest unit - a set of a predicate and arguments - before extraction as triples, knwoledge can be transcribed in one by one manner either by human or by machine.
 * Moreover, such small piece of knowledge will receive simple grammatical structure that serve as better material for automatic transcription by language parsers.

( Major loss は　coherent によるものと　知識の形が期待するentityの関係以外の場合　があることが結論）
 * therefore in this paper, we report our attempt at harvesting knowledge from text in two discrete steps.
 * First we breakdown the text into a list of simpler sentences by discrete set of operations and then harvested knowledge in a form of relations of entities by the operation widely used in knowledge harvesting.
 * Then we can tell the yield of extraction in quantitative manner and where is the major cause of failure in harvest.

Intro: ２ページ最初　more simply 以下

 * More simply, assuming that knowledge represented in texts are features of biomedical entities and they are  described as relations to other biomedical entities - this turned out untrue at least in anatomy textbook -, the knowledge harvesting task can be divided in two subtasks. Namely, for a given biomedical text, (i) look up the text for technical terms stored in terminology, and (ii) find predicate that relate between technical terms.
 * At first glance....

Materials: in earlier days, is less complicated ... 以下

 * ....in earlier days, the content is still valid but the book is in public domain.
 * Therefore, we share the raw materials or processed materials without license issues.
 * the Gray's textbook describes morphological features and mutual relations among human body parts, i.e. bones, muscles, nerves and etc. In particular, in this demonstration, we try to harvest knowledge from Forebrain section, which describes the most complicated structure among organs with 792 sentences.