Full Description
This book presents all components and knowledge sources required for Transparent Information Retrieval. Depending on the respective topic and taking care of their interoperability, both deep and shallow technology is used. The processing starts from the analysis of the text data and collects its results in a multilingual conceptual network, this way enabling Transparent Information Retrieval where users communicate with the system in their native language while the documents could be in a different language, transparent to the users.
To do so, the author investigates all text analysis components required for multilingual indexing, starting from preparatory work like language and topic identification, continuing with sentence splitting and tokenization (including Chinese), and describing lexical analysis, also for multiword entries and Named Entities. Entries are then disambiguated both on syntactic (by a tagger) and semantic level (by multilingual word sense disambiguation). The analysis results are collected in a dynamic multilingual ConceptNet, which is an index structure extended by monolingual relations (like synonyms, or head-modifier links) as well as multilingual ones (translations). In addition to many European languages also Turkish, Arabic, Persian, and Chinese are treated.
The book concludes with a description of components needed to build the required resources, like crawlers, bilingual term extraction, and tools for defaulting linguistic annotations. For each component, readers will find a technology overview, a discussion of its main challenges in computational treatment, a description of the technical solution selected, and evaluation information.
Contents
Preface.- 1. System Design.- 2. TINA Analysis Strategy.- 3. Text Analysis Preprocessing.- 4. Text Segmentation.- 5. Lexical Analysis.- 6. Special Entries.- 7. Disambiguation.- 9. Transparent Information Retrieval (TIR) and the LtConceptNet.- 9. Resources.
