Contact:

Hermann Helbig

[email protected]

The NLI-Z39.50

The NLI-Z39.50 is a natural language interface (NLI) for access to library databases supporting the standardized Z39.50 protocol (ISO-23950). The NLI-Z39.50 has been funded by the DFG (Deutsche Forschungsgemeinschaft).

Online Demonstration

In the following online demonstration, several example queries are processed by the NLI-Z39.50.

NLI-Z39.50 online demonstration (requires internet access)

Description of the Graphical User Interface

Natural language interface (simple)

This interface allows the user - without knowing technical details - to submit a natural language to the NLI. All search options have been assigned default values. Moreover, suitable target systems are selected by the Broker. In addition, the query history is displayed and updated during the search process.

Fig. 2: Graphical User Interface of the NLI-Z39.50.

Natural language interface (expert mode)

In expert mode of the NLI-Z39.50, the user has control over a variety of parameters directly influencing the search results. Furthermore, target systems can be selected from a list of available systems for a parallel search.

Fig. 3: Graphical User Interface of the NLI-Z39.50 in expert mode.

Because the screenshots are from the original German version of the NLI-Z39.50 and have not been translated, the following sections describe and translate all elements of the graphical user interface of the NLI-Z39.50. The graphical user interface of the NLI-Z39.50 contains the following elements:

Natural language query (Natürlichsprachliche Anfrage)
This text field contains the user's natural lamnguage query. After selecting target systems and options for a search, the button Starten (Start search) will initiate the search for bibliographic references.

Query history (Anfrageverlauf)
The previously submitted queries of a user are displayed for a successive refinement of a query.

Options (Optionen)
The following options are available in the NLI-Z39.50:

1.	Number of results (Ergebnisse anzeigen) 10, 25, 50, 100, or 300 results can be displayed on a page.
2.	Filtering of results (Ergebnisse filtern) To increase precision, the set of retrieved documents can be filtered. This option has proved useful if databases do not fully support all required search features occuring in a query.
3.	Merging results (Ergebnisse zusammenfassen) The NLI supports the merging of results. For example, database records for different editions of a document may be contained in a single database or different database records for the same document (in bibliographic formats such as MARC21, UniMARC, MAB2) may exist in different databases. Selecting this option enables the NLI to use a de-duplication of results.
4.	Using compound splitting (Kompositazerlegung) Using constituents of a compound in a database query enables the NLI-Z39.50 to find results containing either the compound or all parts of a compound. This option increases the number of hits / results found.
5.	Including phonetically similar proper nouns (Phonetische Suche) This options enables the NLI-Z39.50 to include phonetically similar proper nouns in a database query.
6.	Expanding the query with semantically related terms (Anfrageerweiterung) The NLI-Z39.50 will automatically add semantically related concepts in a database query if this option is set. Using semantically related concepts (such as synonyms, hyponyms, etc.) will increase the numebr of results found.
7.	Ordering of results (Sortierung) Results can be ordered by Autor (author), Titel (title), Jahr (aufsteigend) (publication year ascending), and Jahr (absteigend) (publication year descending).
8.	Cross lingual search in English or German (Zielsprache) Concepts in the database query can be expressed in German (Deutsch), English (English), or both languages (Deutsch+Englisch).
9.	Selecting a display format (Anzeigeformat) Results can be displayed in a compact form (Kompakt), a short form (Kurz) or in a long form (Lang).

Target systems (Zielsysteme)
This set of checkboxes provides the user with a means to select target systems for a query, which can then be processed in parallel. Target systems include, for example, the Library of Congress, most major library catalogs in German-speaking countries, and the library of Die Deutsche Bibliothek in Frankfurt / Leipzig.

NLI Screenshots

For viewing the screenshots, no internet access is required (click on thumbnails to see a larger screenshot).

NLI-Z39.50 query page|
|
Trace of query processing|
< | | |
| | |
Result display |
,

System Architecture

The NLI consists of several modules communicating via messages. Each module fulfils a designated task. Figure 1 gives an overview on the system architecture.

Fig. 1: System architecture of the NLI-Z39.50.

User Client: Control of data flow and communication
WCFA: Natural language analysis with the WOCADI parser, employing the semantic computer lexicon HaGenLex. The natural language analysis returns a semantic network for the user query according to the MultiNet paradigm
TRANSFOR: Transformation of semantic networks into database queries
IR Client: Access to local and remote databases via ZOOM or SOAP
ANSWER: Processing of the results and answer generation
Broker/Critic: Selecting target systems and generating additional feedback for the user in problem situations

Query Examples

The following queries demonstrate, which kind of queries are understood by the NLI-Z39.50. Preferably, the input is a complete sentence, but database queries are generated even for incomplete queries (noun phrases or one-word-queries), and malformed input.

Welche Artikel von Peter Jackson zur künstlichen Intelligenz sind bei Springer in den letzten Jahren veröffentlicht worden?
(Which articles by Peter Jackson on artificial intelligence have been published by Springer during the last few years?)
Kennst du Bücher von Dietrich Dörner über strategisches Denken, die nach 1997 veröffentlicht wurden?
(Do you know books by Dietrich Dorner about strategic thinking which were published after 1997?)
Bei welchen Verlagen hat Nilsson über Expertensysteme oder KI veröffentlicht?
(Where did Nilsson publish on expert systems or AI?)
Ich suche englischsprachige Artikel und Bücher, die im letzten Jahr von Helmut Schnelle geschrieben wurden.
(I am looking for English articles and books written by Helmut Schnelle last year.)

Evaluations

Different retrieval and indexing strategies have been tested and evaluated for the NLI-Z39.50 in successful particpations in CLEF in the years 2003, 2004, and 2005.

References

University of Hagen, IICS, [email protected]