View
217
Download
2
Embed Size (px)
Citation preview
‘Input’ for an NLG system
weather(bremen, wet)weather(bremen, windy)
event(bremen, arts-festival)event(bremen, film-festival)event(munich, oktoberfest)
nearby(bremen, moors)quality(bremen,attractive)
quality(bremen,green)population(bremen, 750000)
industry(bremen, service)atmosphere(bremen, clean)
What can we express with ‘primitive actions’ and what needs
to be expanded with a plan?
How much do we want to fit inside one sentence?
(pa1 / property-ascription :speechact assertion :domain (b / city :name Bremen) :range (b3 / city
:property-ascription(g / color-quality :lex green)))
(pa4 / property-ascription :tense present :domain (c / climate :lex climate :owned-by (bremen / city)) :class-q nonclass :range (w1 / material-quality :lex wet))
(1) putting everything together
weather(bremen, wet)weather(bremen, windy)
event(bremen, arts-festival)event(bremen, film-festival)event(munich, oktoberfest)
nearby(bremen, moors)quality(bremen,attractive)
quality(bremen,green)population(bremen, 750000)
industry(bremen, service)polution(bremen, low)
(i / generalized-possession :domain (b / city :name bremen) :range (f1 / festival
:use-modification-q use:use-modification-id (a / area :lex art)))
These individual SPLs are in the file given on the website
You can try generating with this file and see what you get…
• Note that the first line of the file is a simple list of the names ‘facts’ to be expressed
– (pa4 pa5 i1 i3 gp0 pa2 pa0 pa1 pa3 gp3 gp1)
• You can try varying the order of these elements to see what happens to the ‘text’
• Alternatively, or in addition, you can try varying the individual pieces of SPL within the list: make sure that you keep the brackets balanced!
Here’s one result of such a generation…
"Bremen is a green CITY , and Bremen has NEARBY MOORs , and Bremen 's POPULATION is seven hundred and fifty thousand SOULs , and Bremen 's CLIMATE is wet , and Munich has many EVENTs , and Bremen is an ATTRACTIVE CITY , and Bremen has a film FESTIVAL , and Bremen has low POLLUTION , and Bremen 's CLIMATE is windy , and Bremen has industries in the SERVICE SECTOR ."
Task
• Set out as clearly as possible all the steps necessary to get from the input to the output
– making the output a better text than the last one!– state just what each stage of the process is to achieve:
what does it expect as input and what does it deliver as output
– what knowledge does it need to achieve this– use all of the material that we have touched upon so far
Hints and pointers…
Kinds of RST elaboration
• Elaboration-general-specific
• Elaboration-object-attribute
• Elaboration-set-member
• Elaboration-specific-general
• Elaboration-whole-part
• Elaboration-process-step
Moore & Paris (1993, p670)
Elaboration-object-attribute
Intention: (know-about ?hearer ?concept)
elaboration
The RST structure
(know-about ?hearer has(?object,?attribute))
?object has(?object,?attribute)
‘Conceptual’ aggregation
weather(bremen, wet)weather(bremen, windy)
event(bremen, arts-festival)event(bremen, film-festival)event(munich, oktoberfest)
nearby(bremen, moors)quality(bremen,attractive)
quality(bremen,green)population(bremen, 750000)
industry(bremen, service)atmosphere(bremen, clean)
(pa / property-ascription :domain (b / city :name Bremen) :range (b2 / city
:property-ascription((a / evaluative-quality :lex attractive) (g / color-quality :lex green)):relations((ac / accompaniment :participation-q participation :domain b2 :range (m / geographical-feature :lex moor
:number plural:property-ascription (n / spatial-quality :lex nearby))))))
“Bremen is an attractive and green city with nearby moors.”
quality (bremen, attractive)quality (bremen, green)nearby (bremen, moors)
Aggregation