Natural language query
Natural language query interpretation
Pivot query
Pivot query formalization
Formal queries ordered list
formal
The pivot query
In the SWIP system, the query interpretation process consists of two main steps
which are illustrated in Figure 1. The rst step, the natural language query interpretation roughly consists in the identication of named entities, a dependency
analysis and the translation of the obtained dependency graph into a new query,
called pivot query and presented in this section. In the second step, pivot query
formalization, predened query patterns are mapped to the pivot query; we thus
obtain a list of potential interpretations of the user query, which are then ranked
according to their estimated relevance and proposed to the user in the form of
reformulated NL queries. The displayed relevancy mark and the reformulation
of the suggested query interpretation in NL are for us a way to overcome the
habitability problem introduced in [8]. The habitability problem states that end
users can be quite at a loss when facing too much freedom in query expression:
he/she can express queries out of the system capabilities (with respect to the
information need or to the query formulation) and, in worst cases, the system
can misunderstand the query and the user not realize it. The relevancy mark
Natural language query
Natural language
query interpretation
Pivot query
Pivot query formalization
Formal queries ordered list
Fig. 1. Overview of the SWIP interpretation process.
and the NL query reformulation give to the user feedback in an understandable
way and consequently increase the condence in the system.
Pivot query is a structure half-way between the NL query and the targeted
formal query, and aims at storing results from the rst interpretation step. It
explicitly represents extracted relations between keywords of the NL query sentence. We use this pivot language in order to facilitate the implementation of
multilingualism by means of a common intermediate format: a specic module
of translation of NL to pivot has to be written for each dierent language, but
the pivot query formalization step remains unchanged. For the moment, we have
adapted our system for English and French.
3.1 Denition and syntax of the pivot language
The detailed grammar of the pivot language is presented in [13]. A pivot query
is composed of keywords connected with relationships which are more or less
explicit. The optional ?" symbol before a keyword means that this keyword
is the focus of the query: we want to obtain specic results corresponding
to that keyword. A pivot query is composed of a conjunction of subqueries:
unary subqueries, like ?"actor" which asks for the list of actors in the KB;
binary subqueries which qualify a keyword with another keyword: the query
?"actor": "married" asks for the list of married actors; and ternary subqueries
which qualify, by means of a keyword, the relationship between two other keywords: the query ?"actor": "married to"= "Penelope Cruz" asks for the actor(s) that is/are/was/were married to Penelope Cruz