Jump to content

Semantic decomposition (natural language processing)

Add links
From Wikipedia, the free encyclopedia

This is an old revision of this page, as edited by Datenverlust (talk | contribs) at 09:31, 9 October 2018 (extended the introduction.). The present address (URL) is a permanent link to this revision, which may differ significantly from the current revision.

A semantic decomposition is an algorithm that breaks down the meaning of concepts into less complex concepts[1]. The result of a semantic decomposition is a representation of meaning. This representation of meaning can be used in an application e.g. for a Artificial intelligence.

The basic idea of a semantic decomposition is taken from the adult learning skills of humans, where words are explained using other words, which is based on the Meaning-text theory. The Meaning-text theory is used as a theoretical linguistic framework, to describe the meaning of concepts with other concepts.

Motivation

A artificial notion of meaning needs to be created for a strong AI to emerge [2]. Without a language, and with that the meaning of the words used in this language, an AI is unable to think. Without thought there is only reacting, no reasoning. AI today is able to syntactically capture language for many specific problems but never establishes meaning for the words of these languages or is able to abstract to concepts [3].

Creating an artificial representation of meaning requires the analysis of what meaning is. There are many terms associated with meaning, like semantics, pragmatics, knowledge, understanding or word sense [4]. All of those describe an aspect and bear a multitude of theories explaining what meaning is. These theories need to be analysed to develop an artificial notion of meaning best fitted to our current state of knowledge.

Abstract approach on how knowledge representation and reasoning allow a problem specific solution (answer) to a given problem (questions)

Representing meaning as a graph is one of the two ways AI, cognition and linguistic research think about meaning (connectionist view). Logicians and formal representation of meaning on the other side build upon the idea of symbolic representation where description logics describe languages and the meaning of symbols. This neat vs. scruffy discussion is going on for the last 40 years [5]. The research so far has identified semantic measures and with that Word Sense Disambiguation Word-sense disambiguation (WSD) - the differentiation of meaning of words - as a main problem of language understanding[6]. As an AI-complete problem WSD is a core problem of natural language understanding [7] [8]. AI approaches which use knowledge given reasoning creates a notion of meaning combining the state of the art knowledge of natural meaning with the symbolic and connectionist formalisation of meaning for AI. The abstract approach is shown in Figure . First we create a connectionist knowledge representation as a semantic network consisting of concepts and their re- lations which will serve as basis for the representation of meaning [9] [10] [11] [12].

This graph is build out of different knowledge sources like WordNet, Wiktionary and BabelNET. Here the graph is created by lexical decomposition which breaks each concept semantically down until a set so semantic primes are reached [1]. The primes are taken from the theory of the Natural Semantic Metalanguage [13], which we have analysed for their usefulness in formal languages[14]. Upon this graph marker passing [15] [16] [17] is used to create the dynamic part of meaning representing thoughts[18]. The marker passing algorithm, where symbolic information is passed along relations form one concept to another, uses node and edge interpretation to guide its markers. The node and edge interpretation models the symbolic influence of certain concepts.

To evaluate the resulting artificial representation of meaning multiple experiments can be constructed [19]. Three of them are:

  • First the whole representation is tested through two experiments: First in its application to create a semantic distance measure [20] and SensEval [21] and second to handle WSD. This is tested on data sets like the Stanford Rare Word Similarity dataset [22].
  • Second different parameters of the marker passing are evaluated through finding semantic distances between words and the context dependence needed for word send disambiguation.
  • Third different parameters of the semantic decomposition algorithm are tested in their effect.

In conclusion the following contributions can be drawn form from using NSM in NLU: A new meaning representation combining connectionism and symbolism is defined and automated in its creation. The known marker passing algorithms are extended and their use for reasoning is analyzed. This analysis will hopefully give insight into the notion of meaning and how it can be used for reasoning.

Future work this areas of research will use the created representation of meaning to build heuristics and evaluate them through capability matching and agent planning, chat bots or other applications of Natural language understanding.

See also:

References

  1. ^ a b Nick Riemer. The Routledge Handbook of Semantics. Routledge, August 2015
  2. ^ Loizos Michael. Jumping to Conclusions. In International Workshop on Defeasible and Ampliative Reasoning, Buenos Aires, 2015
  3. ^ John F Sowa. Knowledge Representation: Logical, Philosophical, and Computational Foundations. Thomson Learning, 2000
  4. ^ Sebastian Löbner Semanitk eine Einführung, 2003
  5. ^ Marvin L Minsky. Logical versus analogical or symbolic versus connectionist or neat versus scruffy. AI Magazine, 12(2):34, 1991
  6. ^ ] Edmonds Philip Agirre, Eneko. Word sense disambiguation: Algorithms and applications, volume 33. Springer, 2007
  7. ^ Nancy Ide and Jean Veronis. Intro- duction to the special issue on word sense disambiguation: the state of the art. Computational Linguistics, 24(1):2-40, 1998
  8. ^ Roman V Yampolskiy. AI-Complete, AI-Hard, or AI-Easy - Classification of Problems in AI. MAICS, pages 94-101, 2012
  9. ^ Katia Sycara, Matthias Klusch, Seth Widoff, and Jianguo Lu. Dynamic service matchmaking among agents in open information environments. SIG- MOD Record, 28(1):47-53, 1999
  10. ^ Phillipa Oaks, Arthur H M ter Hofstede, and David Edmond. Capabilities: Describing What Ser- vices Can Do. In Service-Oriented Computing - ICSOC 2003, pages 1-16 pringer Berlin Heidelberg, 2003
  11. ^ Johannes Fähndrich est First Search Planning of Service Composition Using Incrementally Redefined Context-Dependent Heuristics. In th German Con- ference Multiagent System Technologies, pages 404-407, Springer Berlin Heidelberg, 2013
  12. ^ Johannes Fähndrich, Sebastian Ahrndt, and Sahin Albayrak. Towards Self-Explaining Agents. PAAMS (), 221(Chapter 18):147-154, 2013
  13. ^ Cliff Goddard and Anna Wierzbicka. Semantic and Lexical Universals. Theory and Empirical Findings. John Benjamins Publishing, 1994
  14. ^ Johannes Fähndrich,Sebastian Ahrndt, and Sahin Albayrak. Formal Language Decom- position into Semantic Primes. ADCAIJ: Advances in Distributed Computing and Artificial Intelligence Journal, 3(8):56, 2014
  15. ^ James A Hendler. Integrating marker- passing and problem-solving: A spreading activation ap- proach to improved choice in planning. Hillsdale, N.J. : Lawrence Erlbaum Associates, 1988
  16. ^ Graeme Hirst. Semantic Interpretation and the Resolution of Ambiguity. Cambridge University Press, March 1992
  17. ^ Johannes Fähndrich, Sebastian Ahrndt, and Sahin Albayrak. Self-Explanation through Semantic Annotation and (automated) Ontology Creation: A Survey. In 10th International Symposium Advances in Artificial Intelligence and Applications, pages 1-15, ACM 2015
  18. ^ F Crestani. Application of Spreading Activation Techniques in Information Retrieval. Artificial Intelligence Review, 11(6):453-482, 1997
  19. ^ https://d-nb.info/1162540680/34 PP. 147
  20. ^ Johannes Fähndrich abine Weber, and Sebastian Ahrndt. Design and Use of a Seman- tic Similarity Measure for Interoperability Among Agents, In Multiagent System Technologies, pages 41-57 Springer International Publishing, 2016
  21. ^ P Edmonds. SENSEVAL: The evaluation of word sense disambiguation systems. ELRA, 2002
  22. ^ M T Luong, R Socher, and Christo- pher D Manning. Better word representations with recur- sive neural networks for morphology. CoNLL-2013, 2013
Retrieved from "https://en.wikipedia.org/w/index.php?title=Semantic_decomposition_(natural_language_processing)&oldid=863198875"