Natural Language Generation Systems Based On Halliday‘s Systemic Linguistics

Systemic linguistics is an ―applicable linguistics‖ which can be used in many areas. Among those applied areas of systemic linguistics, Halliday pays special attention to natural language processing, especially natural language generation. Since 1960 s, Halliday‘s systemic linguistics, has been used in the study of natural language generation. In the 1970 s, systemic linguistics began to be widely used in natural language generation systems, such as Proteus, Penman, Slang, Genesys, Chinese syntactic realization system(Csrs) and so on. At present, it has become the most widely-used linguistic theory in natural language generation systems.In light of such context, this paper tries to undertake an interdisciplinary research on those representative natural language generation systems, such as Penman, Csrs and KPML, from the perspective of systemic linguistics. The main research question of this paper is:(1) to find out how the core concept of ―system‖ in systemic linguistics can be applied to Penman, Csrs and KPML;(2) what problems will arise in this process. Based on the questions above, in turn, we will try to rectify and develop the original systemic theory from a theoretical perspective.In this paper, the main research methods are reference method, comparative analysis method and formal method. Through collecting and analyzing large reference materials, this paper gives a detailed introduction of ―system‖ in the three main natural language generation systems. Through comparative analysis method, this paper shows similarities and differences of the ―system‖ in those three natural language generation systems. Formal method is a graphical representation method, the system(network) representation method.Through the above research, this paper finds out:(1) ―System‖ develops rather quickly. At first, Saussure put forward the concept of ―system of language‖, regarding language as a system of signs. Later, Halliday introduced this concept, paying more attention to the paradigmatic relation, based on which five basic forms are thus evolved.(2) When the concept of ―system‖ is applied to natural language generation systems, there appear some problems that cannot be discovered in other areas, such as the ―syntagmatic gap‖ phenomenon in Penman.(3) At last, based on the artificial neural network model of artificial intelligence, this paper puts forward an improvement method, the system network representation method with better representation ability and realization ability.This paper also has its own limitations: the studies on generation systems is rather limited, which remains further exploration in other natural language generation systems. Besides, this paper mainly focuses on text generation systems, however, in later study, it needs to pay more attention to multimodal text generation systems, which is a much more energetic direction.