Download presentation
Presentation is loading. Please wait.
1
NLDB’04 – June 23 – 25, Salford, Manchester, United Kingdom 0/9 Semantic Tagging and Chunk- Parsing in Dynamic Modeling G. Fliedl, Ch. Kop, H.C. Mayr, A. Salbrechter, G. Weber, Ch. Winkler
2
NLDB’04 – June 23 – 25, Salford, Manchester, United Kingdom 1/9 natural language sentence parsing or tagging Die Auftragsab- teilung bearbeitet die Posten Text Mapping into UML Validation Interpretation to collect KCPM entries order customer article Motivation /1 About the Project /1
3
NLDB’04 – June 23 – 25, Salford, Manchester, United Kingdom 2/9 Motivation /2 notions: interaction diagram notions: state charts notions: activity- diagram notions: use case diagram KCPM entries thing-, operation-, cooperation-,connection- types, pre- and post conditions notions: DFD petri-nets EPK notions: UML-object- model, ER, HERM, … About the Project /2
4
NLDB’04 – June 23 – 25, Salford, Manchester, United Kingdom 3/9 Conceptual Predesign (KCPM) Modeling notions of KCPM Dynamic operation type, conditions, cooperation type, thing type (+ meta attributes) task must be executed (operation type) someone is responsible to execute the task (actor / thing type) tasks manipulates things (parameter / thing type) (e.g. thing types perform tasks, tasks update/read thing types, thing types are recipients) a set of tasks can only be executed under specific pre-conditions and the execution leads to post-conditions (cooperation type) things can be involved in conditions (involved thing types) (e.g. things have a specific state expressed by these conditions)
5
NLDB’04 – June 23 – 25, Salford, Manchester, United Kingdom 4/9 FEATURES POS-Tagger NP and PP Chunking Verb classification (12 Verbclasses and Subclasses) Word stemming Morphological analysis Tree representation in XML Integration in MS Word Tagging /1
6
NLDB’04 – June 23 – 25, Salford, Manchester, United Kingdom 5/9 Tagging /2
7
NLDB’04 – June 23 – 25, Salford, Manchester, United Kingdom 6/9 Mapping Idea (From NL to KCPM) (1)Fundamental relationships between the respective phrases have to found out based on the - verb category, - PAS and the semantic roles – Agent (AG), Theme (TH), Goal (GO)… - sentence mode (active voice, passive voice) - type of clause - identification of noun phrases - articles … (2)After that, these relationships are interpreted (mapped) to the KCPM dynamic model Mapping /1
8
NLDB’04 – June 23 – 25, Salford, Manchester, United Kingdom 7/9 Example 1 Mapping /2 comes in eV order TH Der Auftrag trifft ein (The order comes in) Ad (1) Fundamental relationships Ergative Verb 1 internal Argument Internal Argument can be found either - The first N3 tag before the verb (default/normal case) - Directly after the verb (N3 is not in the topic position) Ad (2) Interpretation - An ergative verb is a candidate for a condition - The internal Argument is the involved thing type of the condition
9
NLDB’04 – June 23 – 25, Salford, Manchester, United Kingdom 8/9 Example 2 Mapping /3 checks tvag2 article TH Die Auftragsabteilung prüft jeden Artikel des Auftrags (The order department checks each article of the order) Ad (1) Fundamental relationships Binary Agentive Verb 1 external Argument (AG) 1 internal Argument (TH) freedom of topicalisation semantic roles are hard to identify. cases of the nouns are used (articles, quantifiers) Ad (1) Fundamental relationships - An agent verb is a candidate for an operation - Arguments are the involved thing types of the operation order department order AG Poss.
10
NLDB’04 – June 23 – 25, Salford, Manchester, United Kingdom 9/9 Conclusion We try to find the balance between free sentences which we still analyzable and interpretable Extended POS-Tagging is a possible solution to that To find semantic Relationship then remains to the interpretation part Extended Tagging has to combined with deep parsing in some cases
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.