Efficient SQL statement to encode text for NLP locutions - regex

Question

Background
A locution is a noun-phrase consisting of at least two words, such as:
black olive
hot pepper sauce
rose finn apple potato
The separate words black and olive are an adjective (black - JJ) and a noun (olive - NN). However, humans know that black olive is a noun (that differentiates it from, say, a green olive).
The problem here is how to most efficiently transform a list of normalized ingredient names (such as the list above) into a specific format for a natural language processor (NLP).
Example Data
The table can be created as follows:
CREATE TABLE ingredient_name (
id bigserial NOT NULL, -- Uniquely identifies the ingredient.
label character varying(30) NOT NULL
);
The following SQL statements show actual database records:
insert into ingredient_name (label) values ('alfalfa sprout');
insert into ingredient_name (label) values ('almond extract');
insert into ingredient_name (label) values ('concentrated apple juice');
insert into ingredient_name (label) values ('black-eyed pea');
insert into ingredient_name (label) values ('rose finn apple potato');
Data Format
The general format is:
lexeme1_lexeme2_<lexemeN> lexeme1_lexeme2_lexemeN NN
Given the list of words above, the NLP expects:
black_<olive> black_olive NN
hot_pepper_<sauce> hot_pepper_sauce NN
rose_finn_apple_<potato> rose_finn_apple_potato NN
The database has a table (recipe.ingredient_name) and a column (label). The labels are normalized (e.g., single space, lower case).
SQL Statement
The code that produces the expected results:
CREATE OR REPLACE VIEW ingredient_locutions_vw AS
SELECT
t.id,
-- Replace spaces with underscores
translate( t.prefix, ' ', '_' )
|| '<' || t.suffix || '>' || ' ' ||
translate( t.label, ' ', '_' )
|| ' NN' AS locution_nlp
FROM (
SELECT
id,
-- Ingredient name
label,
-- All words except the last word
left( label, abs( strpos( reverse( label ), ' ' ) - length( label ) ) + 1 ) AS prefix,
-- Just the last word
substr( label,
length( label ) - strpos( reverse( label ), ' ' ) + 2
) AS suffix
FROM
ingredient_name
WHERE
-- Limit set to ingredient names having at least one space
strpos( label, ' ' ) > 0
) AS t;
Question
What is a more efficient (or elegant) way to split the prefix (all words except the first) and suffix (just the last word) in the above code?
The system is PostgreSQL 9.1.
Thank you!

Answer 1

CREATE OR REPLACE VIEW ingredient_locutions_vw AS
SELECT
t.id,
format('%s_<%s> %s NN',
array_to_string(t.prefix, '_'),
t.suffix,
array_to_string(t.label, '_')
) AS locution_nlp
FROM (
SELECT
id,
-- Ingredient name
label,
-- All words except the last word
label[1:array_length(label, 1) - 1] AS prefix,
-- Just the last word
label[array_length(label, 1)] AS suffix
FROM (
select id, string_to_array(label, ' ') as label
from ingredient_name
) s
WHERE
-- Limit set to ingredient names having at least one space
array_length(label, 1) > 1
) AS t;
select * from ingredient_locutions_vw ;
id | locution_nlp
----+--------------------------------------------------------
1 | alfalfa_<sprout> alfalfa_sprout NN
2 | almond_<extract> almond_extract NN
3 | concentrated_apple_<juice> concentrated_apple_juice NN
4 | black-eyed_<pea> black-eyed_pea NN
5 | rose_finn_apple_<potato> rose_finn_apple_potato NN
(5 rows)