Task 4.9 (3.11)Natural Language and Speech Interfaces to Knowledge Repositories
Anastasia Karanastasi {[email protected]}
Fraunhofer Institut Objectives Task Activity Overview Proposal for JPA3 Task 4.9 (3.11)
Natural Language and Speech Interfaces to Knowledge Repositories
Principles, methodologies and software for the automation of theconstruction of natural language and speech interfaces to knowledge repositories
Disambiguation of natural language user queries using Domain Ontologies and User Profiles, result ranking
Explore the interplay of speech processing with natural languageprocessing in a particular domain.
Develop an application demonstrator of natural language and speech interfaces to knowledge repositories
Heuristic Evaluation of the prototype - Evaluation Study involving representatives of the target user group
Objectives Task Activities Task 4.9 (3.11)
is able to address uniformly a range of problems in sentence analysis each of which traditionally had required a separate computational mechanism. In particular a single architecture:
handles both syntactic and semantic ambiguities by using upper and domain ontologies
handles ambiguity at both a general and a domain specific environment
consults user profiles to personalize the disambiguation
Objectives Task Activities Task 4.9 (3.11)
utilizes the rich structures that OWL provides to describe ontologies
uses User Profiles to guide the semantic search in the domain ontology and to rank the results in a way a user meets his preferences
we use it to create a Natural Language Interface for the domain of soccer that is used in a question answering system
Objectives Task Activities Task 4.9 (3.11) Objectives Task Activities Task 4.9 (3.11)
A multimedia (video) application within the domain of soccer has
been selected to be used as a first demonstrator of the application.
The natural language processing component developed an
architectural framework, and progressed in the theoretical foundations of the approach.
A detailed design of the natural language prototype system based
on the above foundations exists. The aim of this prototype is toinvestigate the automation possible in the implementation of natural language interfaces to knowledge management systems.
An implementation of a particular application based on the prototype
has started and is progressing well. Objectives Task Activities Task 4.9 (3.11) Objectives Task Activities Task 4.9 (3.11)
Keywords list – limited list of English or German words Acoustic model – cross-word triphone acoustic model for
Language model – language model is presented as a grammar
which includes all used keywords and garbage model withoutlimitation on the order
The grammar is regular expression created on the base of
Garbage model – special language model for non-keywords
represented as a grammar with a free phonemes order
Output – output of keywords spotting is MPEG-7 document
Objectives Task Activities Task 4.9 (3.11)
the system of tree annotations from the of
English, which is a database of trees for about 50,000 English sentences
a Java implementation of a maximum-entropy part-of-
speech (POS) tagger. The POS tagging task is based on the Stanford Log-linear Tagger
a noun compound bracketing mechanism with domain
Objectives Task Activities Task 4.9 (3.11)
a grammatical relation annotation scheme
Map the syntactic descriptions found in the sentence into the
correct roles that the elements (described by the nominals) play in the situation at hand (described by the verb).
a thesaurus (WordNet), where the list of stem of nouns and
verbs are refined to provide the system with semantics, like senses and synonyms of the basic concepts within the utterance
Objectives Task Activities Task 4.9 (3.11) Objectives Task Activities Task 4.9 (3.11)
‘The player with shirt number 9 gave Milan the victory’
Objectives Task Activities Task 4.9 (3.11)
Biatov K. and Larson M, Speaker Clustering via Bayesian Information Criterion using Global Similarity Constraint. The 10th International Conference Speech and Computer SPECOM 2005, Patras, Greece, 2005.
Karanastasi, A., Christodoulakis, S. OntoNL: An Ontology-based Natural Language Interface Generator for Multimedia Repositories. In the Proc. of the 7th DELOS International Workshop on Audio-Visual Content and Information Visualization in Digital Libraries (AVIVDiLib’05)
Karanastasi, A., Kazasis, F., Christodoulakis, S., A Natural Language Model and a System for Managing TV-Anytime Information in Mobile Environments, in the ACM/Verlag Personal and Ubiquitous Computing Journal, Volume 9, 2004
E. Bertini, T. Catarci, S. Kimani and A. Dix, A Review of Standard Usability Principles in the Context of Mobile Computing, in Studies in Communication Sciences journal, 5(1), 2005. Objectives Task Activities Task 4.9 (3.11)
Extensions Formal models in general environments to exploit contextual
information on user preferences to reduce semantic and syntacticambiguities and avoid clarification dialogues
Generality of the approach as well as its performance in a different
Couple practical approaches of natural language with theoretical
approaches of reasoning to produce new models for the construction of natural language interfaces to knowledge repositories
Development of a robust front-end processing module to allow
Objectives Task Activities Task 4.9 (3.11)
Marinoff’s Therapy: A Critique of His Books on Philosophical Practice Shlomit C. Schuster Shlomit C. Schuster is a practicing philosophical counselor. She has written several scholarly articles on philosophical counseling and is the author of Philosophy Practice: An Alternative to Counseling and Psychotherapy (Praeger, 1999) and The Philosopher’s Autobiography: A Qualitativ
the system of tree annotations from the of
English, which is a database of trees for about 50,000 English sentences
a Java implementation of a maximum-entropy part-of-
speech (POS) tagger. The POS tagging task is based on the Stanford Log-linear Tagger
a noun compound bracketing mechanism with domain
Objectives
a grammatical relation annotation scheme
Map the syntactic descriptions found in the sentence into the
correct roles that the elements (described by the nominals) play in the situation at hand (described by the verb).
Objectives
‘The player with shirt number 9 gave Milan the victory’
Objectives
Biatov K. and Larson M, Speaker Clustering via Bayesian Information 
Extensions Formal models in general environments to exploit contextual
information on user preferences to reduce semantic and syntacticambiguities and avoid clarification dialogues
Generality of the approach as well as its performance in a different
Couple practical approaches of natural language with theoretical
approaches of reasoning to produce new models for the construction of natural language interfaces to knowledge repositories
Development of a robust front-end processing module to allow
Objectives