Natural language generation

In the text generation (including natural language generation, English Natural Language Generation, NLG) is automatically created by the software of a computer text. The generation of texts as part of the field of computational linguistics a special form of artificial intelligence.

  • 3.1 Process and Applications in Visual Arts and Literature

Generation process

For the generation process, there are, depending on the method and point of view used different descriptive models and technical terms, without having to disagree in principle.

According to E. Reiter architecture for the generation of today is by default a text planner, a sentence planner, and a Oberflächenrealisierer. For the relations between text segments, use is made of the theory of rhetorical structures, RST, to make the discourse relations. A text is coherent if it can be represented by a tree of rhetorical relations and elementary text units (RST: Man Thompson): The following links are defined as relations between main and subordinate clauses: CAUSE, RESULT, ELABORATION, CONTRAST, SEQUENCE, LIST, CONCESSION and others.

According to M. Hess generation requires two components.

  • The strategic component of what should be said selection information, content selection, range planning. This component usually used search and planning strategies of artificial intelligence.
  • The tactical component, as it should be said: The planning of the linguistic form. This often tailored to the generation grammar aspect will be used.

Ulrich Müller Gaudenzdorfer developed together with the German scholars and computational linguists Raimund Drewek 1981-1999 a system for text generation which was called SARA ( set -random generator ).

Text generation from knowledge bases

" A prerequisite for any kind of generation that exists as text information to be generated as a formal, computerlinguistisch processable information, such as information from databases, or knowledge representations. "

The text generation from such knowledge bases are available in versions for different tasks.

  • Interface to expert systems
  • Production of technical documents in multiple languages ​​from a knowledge base
  • Automatic generation (of directions, weather reports and stock market reports)
  • Generation component of dialogue systems

Applications in everyday culture

Knowledge-based software for natural language text generation can dialog compared to a human user to a limited extent simulate intelligence (see above, artificial intelligence ). In a simple design is output directly to a text input by the user, using rules and a relatively simple knowledge base a question or an answer. The most famous historical example, especially in its execution as a psychotherapist, is the program ELIZA, a chatbot.

A part of the communication with advanced intelligent virtual agent based on this principle, the quality of the dialogue depends inter alia on the linkage of the agent with the knowledge bases. The people of a dialogue with different interfaces can be facilitated when an agent generates text, answer the questions productive:

  • (Also called " Online Moderator " ) while retrieving an information offer, among other things, as a presentation agent of a website
  • In a speech-enabled program for Selecting an agent ( often used for telephonic pre-sorting of customers)
  • For dialogs with characters in computer games

Text generation as a creative process

Text generation may be a component of creative processes in art and literature. For longer works offer fully generated text body, whether generated sinnhaltig or only provided by the post with meaning, no literary quality. However, there are some in the art of the last century and in the contemporary art significant artistic method digital poetry in relation to the text generation.

Methods and applications in visual arts and literature

  • Interventions in the generating software or the knowledge base (artistic and literary experiments ). Example (after Reinhard Döhl ): Max Bense and his Stuttgart group used 1959 Zuse Z22 to " synthesize by means of an input lexicon and a number of syntactic rules texts and spending ."
  • Rework or installation generated text by authors ( literature).
  • Dialogue with the audience ( for example, in art installations. Example: David Link, Poetry Machine

Text generation by phrases thresher

Phrases threshers or bullshit generators (English bullshit generator, also buzzword generator ) existed before the implementation in software and mechanical devices. Probably the first threshing machine was designed as a software phrases LoveLetters_1.0, 1952 programmed by Christopher Strachey at the University of Manchester for the Ferranti Mark I. Similar generators can be found in many more developed versions in the WWW.

Such programs operate on simple concepts that are applied complex designed for more sophisticated method of text generation: terms or phrases are taken from lists, strung together and grammatically correct adjusted ( grammatical realization ). A frequent practice is for the generation of Markov chains. The result is a syntactically correct text that can act sinnhaltig, but in fact nonsense (English bullshit ) is because phrases threshers access unused knowledge about the importance of particles. Thus, for example, jokingly satirize empty rhetoric of literature.

History

Apart from mechanical threshers phrases as precursors and, apart from the earliest attempts to generate texts by software, the first phase starts natural language generation with programs that schematically access to text generation to knowledge that is already stored in text form. So functioned from 1963 BASEBALL, an interface to the data of the American Baseball Baseballiga and SAD SAM, an interface for entering relationships that already answered questions. After several other works in this direction in 1966 ELIZA program by Joseph Weizenbaum appeared. In the second phase, the knowledge of facts and rules is coded: LUNAR, in 1972, is the interface to the database of the lunar samples collection of the Apollo 11 mission. PARRY, 1975, simulates a paranoid in conversation with a psychiatrist. ROBOT, 1977, the first commercial question -answering system. VIE -LANG, 1982, by Ernst Buchberger, a dialog system in German language, the sentences from a semantic network is generated. HAM -ANS, 1983, by Wolfgang Hoeppner, is a dialogue system in the German language, the simulated example, a hotel manager.

Chatterbot Ferranti Mark 1
152745
de