Automatic Generation and Reranking of SQL-Derived Answers to NL Questions - Trustworthy Eternal Systems via Evolving, Software Data and Knowledge - page 66

At the first iteration we use ROOT to add are to Σ , then we also exploit
it to add capital and include state to Π as soon as we check that there is an
occurrence state.capital in IS . At this point these three relations have been
deleted from SDC opt
q 1

q 1 used in the next iteration. Note that
since are is a short stem, it should be deleted from Σ .
At the second iteration (first recursion step) we don't have a ROOT relation
so we use nsubj to add add border to Σ and state to Π .Sincewith rcmod
we find an occurrence border.state name in IS , border is added also to Π .At
this point, seeking through the end of the list we discard dobj because even if
border

obtaining SDC opt

∈ Π we do not find state.border in IS , so these other three relations
are deleted from SDC opt

q 1 obtaining SDC opt q 1 for the last iteration.

q 1 composed by two mod relations, so
we add all stems to Π and delete their associated relations from the list.

In the third iteration we have SDC opt

3.3 Building the SELECT Clauses Set

Once we have identified the set Π of projection-oriented stems, we can use it to
search in metadata all the fields that could match with them. The generation
process for S is described by the following generative grammar.

S→ AGGR '(' FIELD ')' | FIELD

AGGR → max | min | sum | count | avg

FIELD → TAB.COL

∈ x∈Π π table name ( σ table name = x ( IS . Tables))

TAB

∈ x∈Π π column name ( σ column name = x ( IS . Columns))

COL

, we also associate a weight w i , calculated according
to the procedure described in Section 4.3 (we will discuss it later). For example,
considering the IS scheme in Figure 1, the SELECT clauses originated from Π
of Table 1 are shown in Fig. 3. Note that the superscript numbers indicate the
weight associated with each statement.

With each element of S

Fig. 3. A subset of SELECT clauses for q 1

3.4 Building the WHERE Clauses Set

Before generating WHERE clauses, the selection-oriented set of stems Σ should
be divided into two distinct sets: Σ L and Σ R .
The set Σ L contains stems that find their matching in IS and allow us to build
the set of left-hand side expressions FIELD w i , where FIELD is defined

above and computed with Σ L in place of Π ( w i is its associated weight).

W L →