RDF triples generated from an example sentence. 

Context in source publication

Context 1

... as subsets of natural languages, have recently received much attention with regard to ontology-based knowledge acquisition systems, for its ability to eliminate ambiguity of expressions in natural languages. Several studies were devoted to the use of CNL in ontology-related data processing such as ontology construction, query generation, and data annotation [3][4]. A CNL-based guided look-ahead editor might help users select proper words that meet his intended but vague notions without proper knowledge on the sentence structures. The statements controlled by predefined grammars, usually defined in CFG which is a computational notation of natural language structures, can be translated into ontology-referenced data and queries with precision [2][5]. Our previous work, OntoPath, assists editing in such an intelligent way that it recognizes the resource type of a description and offers users context-sensitive actions to perform on that description. A domain-specific ontology plays a role in collecting language constituents, such as nouns and verbs, to be translated into RDF triples [9][10]. A lightweight look-ahead editor helps users, specifically medical experts, with guidance on choosing next words, using the approved grammars and semantic relations of entities from the ontology. Because most medical sentences have general recommended structures to ensure precise knowledge expression, CNL and a look- ahead guiding system can assume an important role in such an application. However, our previous work and other systems have difficulties in enlarging the expression capacity, expanding the grammar, specifying patternized sentences, and adapting informal expressions such as Korean sentences with English words. These difficulties are attributed to the fact that the grammar definition system like CFG does not include semantic structures, but sequential structures of a sentence. These limitations need to be solved to deal with various sentences so that users can exploit more familiar expressions, and are enforced into using the patternized sentences. A newly developed editor, which we propose in this paper, permits grammar definitions through CFG-LD that includes both sequential and semantic views on sentence structures. Using this grammar definition system, we can define grammars and the semantic structures of sentences to be used in our editor. The Grammar definitions include the structural descriptions of grammatical states to mention sequences of POS (Part-Of-Speech) with CFG. Designations of lexical dependency between sentence elements are also included. Using defined grammars, the implemented CNL editor enables us to get structure data from writer’s narratives with 1) more sophisticated expressions, 2) patternized expressions, and 3) informal expressions consisting of multi language constituents. We begin this paper with the description of related works on a CNL. An explanation of the representation of narratives using RDF triples is provided in section 2. The CFG-LD and its definition rules are discussed in section 3. In sections 4 and 5, we explain the architecture and implementation of the developed editor. Finally, we provide conclusions of this work in section 6. CNL was restricted subsets of natural languages on grammars and dictionaries to eliminate ambiguity and complexity of pure natural languages. Originally, the main purpose of controlled languages was to improve readability for human readers, particularly non-native speakers. An example is AECMA Simplified English that was created as a manual description language for aircraft maintenance guideline. Another advantage of CNL is to improve text processing capability of computers with removed complexity. Many studies have been done to develop systems that transform written sentences into formal logical expressions. Some well-known examples are as follows: ACE (Attempto Controlled English) [6], CLCE (Common Logic Controlled English) [7], and PENG-D [2]. For an automatic translation from a discourse representation structure into a variant of first-order logic, ACE is defined as a controlled natural language in Attempto project. The ACE based sentences are translated into the Semantic Web querying language PQL [8]. Other example of a controlled natural language is CLCE that has been developed by Sowa [9]. CLCE, as a formal language with an English-like syntax, is supplied with more expression than ACE in the sense that it supports ontology for sets, sequences, and integers, and also allows in-line declarations of words linked to relational databases. It supports the automated translation of written narratives to conceptual graph or other logical expressions [8]. PENG-D also proposed a computer-processable CNL to be translated into formal logical sentences decidable with an OWL (Web Ontology Language) language. In this subsection, we overview a translation from a sentence to RDF triples with a gross description narratives example of pathologic examination. The description language supported by OntoPath is compatible with a restricted form of RDF and RDF Schema. The system is designed to annotate the semantic metadata in RDF with the vocabularies that are already constrained by a given ontology in RDF Schema. The ontology then plays a role in guiding the generation of medical narratives as RDF documents. The narratives are validated with the syntactic and semantic rules of RDF Schema and are transformed into RDF documents. An RDF triple statement consists of a specific resource, which is an individual primitive semantic element with a named property and value for that resource. The basic RDF model represents the named properties and property values. A property is a rule that provides the meaning of the expressions, which is specifying the way the thing should be constituted. A built ontology such as a schema, which is a vocabulary description language, provides mechanisms for describing groups of related resources and the relationships between these resources. Instead of defining a class in terms of the properties its instances may have, the ontology describes properties in terms of the resource classes to which they apply. This is the role of the domain and range mechanisms. This example sentence can be translated as shown in the figure 1 when it is typed in the form of the predefined grammar through the guidance of the editor [9][10]. The instances about a real patient, ‘a specimen received’ and ‘a cyst’ are conceptualized as the instances of classes Tissue and Cyst , respectively. The properties, contains and measures , are also specified with their object values in the sentence. This translation can be definitely performed on the example sentence written in a predefined grammar. However, if the user describes the sentence with another manner, it can not be successfully translated, because the translation system will assume restricted grammars and translation processes. To expand the grammars for enlarging expression capacity, we can add more grammars using CFG, but it is still not enough for this translation work, since the translation can be different from the composed structure and semantic dependency among the sentence elements. In this chapter, we introduce CFG-LD, which is a grammar definition system for describing grammars with lexical dependences. As we have shown in the previous chapter, the translation between a simple English sentence and a RDF triple is possible through quite simple translation rule on the grammar. However, it is hard to deal with those sentences with different structures, and an annexed expression such as idioms (e.g., “there is something”) or patternized phrases appearing in the sentences. Other grammatical expressions following a different sequence of POS such as ‘subject-object-verb’ are also hardly handled through the original approach. Sequential and semantic structures of those sentences should be declared to enlarge the translation capacities. Resolving the various structures of sentences can be possible through the previously developed CNL systems listed in the previous chapter. Their built-in sentence resolutions mainly relied on English are restricted in the informal expressions consisting of multi-language constituents, and it is also hard to gather well-defined CNL grammars written in every desired language. Therefore, in our CNL based editor, we employ slightly modified grammar expressions named CFG- LD. It notifies a lexical parser for both grammars and lexical dependencies, to let the parser or system know sequential and semantic structures of the grammars where the ontology provides language constituents and domain and range relations of them. CFG is a famous computational notation used to express natural language structure, and to make development of applications that parse natural language sentences easily. Chomsky proposed the notion of CFG as a model for describing natural languages with following four quantities: Terminals, Non-terminals, Productions, and Start symbol [11]. The grammars described below in CFG express simple grammars to parse an example sentence, ‘Nam is a student supervised by a professor named Kim’ with a set of lexicons enabling aware of terminals’ ...