CS5545: Realisation

1
CS5545 Realisation

• Background Reading Building Natural Language
  Generation Systems, chap 6

2
Realisation

• Third (last) NLG stage
• Generate actual text
• Take care of details of language
• Syntactic details
• Eg Agreement (the dog runs vs the dogs run)
• Morphological details
• Eg, plurals (dog/dogs vs box/boxes)
• Presentation details
• Eg, fit to 80 column width

3
Realisation

• Problem There are lots of finicky details of
  language which most people developing NLG systems
  dont want to worry about
• Solution Automate this using a realiser

4
Syntax

• Sentences must obey the rules of English grammar
• Specifies which order words should appear in,
  extra function words, word forms
• Many aspects of grammar are somewhat bizarre

5
Syntactic Details Verb Group

• Verb group is the main verb plus helping words
  (auxiliaries).
• Encodes information in fairly bizarre ways, eg
  tense
• John will watch TV (future add will)
• John watches TV (present - s form of verb for
  third-person singular subjects)
• John is watching TV (progressive form of BE
  verb, plus ing form of verb)
• John watched TV (past use ed form of verb)

6
Verb group

• Negation
• Usually add not after first word of verb group
• John will not watch TV
• Exception add do not before 1-word VG
• Inflections on do, use infinitive form of main
  verb
• John did not watch TV
• Exception to exception use first strategy if
  verb is form of BE
• John is not happy

7
Realiser

• Just tell realiser verb, tense, whether negated,
  and it will figure out the VG
• (watch, future) -gt will watch
• (watch, past, negated) -gt did not watch
• Etc
• Similarly automate other obscure encodings of
  information

8
Morphology

• Words have different forms
• Nouns have plural
• Dog, dogs
• Verbs have 3ps form, participles
• Watch, watches, watching, watched
• Adjectives have comparative, superlative
• Big, bigger, biggest

9
Formation of variants

• Example plural
• Usually add s (dogs)
• But add es if base noun ends in certain letters
  (boxes, guesses)
• Also change final y to i (tries)
• Many special cases
• children (vs childs), people (vs persons), etc

10
Realiser

• Calculates variants automatically
• (dog, plural) -gt dogs
• (box, plural) -gt boxes
• (child, plural) -gt children
• etc

11
Punctuation

• Rules for structures
• Sentences have first word capitalised, end in a
  full stop
• My dog ate the bone, the meat, etc..
• Lists have conjunction (eg, and) between last two
  elements, comma between others
• I saw Tom, Sue, Zoe, and Ciaran at the meeting.
• Etc
• Realiser can automatically insert appropriate
  punctuation for a structure

12
Punctuation

• Rules on combinations of punc
• Dont end full stop if sentence already ends in a
  full stop
• He lives in Washington D.C.
• He lives in Washington D.C..
• Brackets absorb some full stops
• John lives in Aberdeen (he used to live in
  Edinburgh).
• John lives in Aberdeen (he used to live in
  Edinburgh.).
• Again realiser can automate

13
Pouring

• Usually we insert spaces between tokens, but not
  always
• My dog
• Mydog
• I saw John, and said hello.
• I saw John , and said hello
• Automated by realiser

14
Pouring

• Often want to insert line breaks to make text fit
  into a page of given width
• Breaks should go between words if possible
• Breaks should go between words if poss
• ible
• If not possible, break between syllables and add
  a hyphen
• Realiser automates

15
Output formatting

• Many possible output formats
• Simple text
• HTML
• RTF (MS Word)
• Realiser can automatically add appropriate
  markups for this

16
Summary

• Realiser automates the finicky details of
  language
• So NLG developer doesnt have to worry about
  these
• One of the advantages of NLG

17
simplenlg classes

• Realiser realise objects
• SyntaxPhraseSpec
• SPhraseSpec represents a sentence as verb,
  subjects, complements
• NPPhraseSpec represents a noun phrase as
  determiner, modifiers, noun
• can be subject or complement in SPhraseSpec
• PPPhraseSpec represents prepositional phrase
• Can be modifier in SPhraseSpec, NPPhraseSpec

18
Realiser API

• class Realiser
• public void Realiser () // default realiser
  (text)
• public void Realiser (boolean HTML) //
  HTML output ?
• public String realiser(Object spec)
• Usage
• // create text spec ts
• Realiser r new Realiser()
• String output r.realise(ts)

19
SPhraseSpec example

• SPhraseSpec p new SPhraseSpec()
• p.addSubject("my dog")
• p.addSubject("your cat")
• p.addVerb("like")
• p.addComplement("balls")
• p.setTense(SPhraseSpec.Tense.FUTURE)
• p.setNegated(true)
• Results in
• My dog and your cat will not like balls.

20
NPPhraseSpec example

• NPPhraseSpec subject new NPPhraseSpec("dog")
• subject.addModifier("big")
• subject.addDeterminer("the")
• Results in
• the big dog

21
PPhraseSpec Example

• PPPhraseSpec pp new PPPhraseSpec("in", "the
  park"))
• p new SPhraseSpec("I", "be")
• p.addModifier(pp)
• Results in
• I am in the park

22
Lexicon

• Lexicon class allows you to directly do
  morphology
• Usually dont need to call explicitly

23
Lexicon example

• Lexicon lex new Lexicon()
• System.out.println(lex.getPlural("child"))
• System.out.println(lex.getPast("eat"))
• System.out.println(lex.getPastParticiple("eat"))
• System.out.println(lex.getSuperlative("happy"))
• This will print out
• children
• ate
• eaten
• happiest

24
Other feature Lexicaliser

• Lexicaliser (different from lexicon) allows you
  to specify templates which produce SPhraseSpec
  from Protégé instances.
• Wont discuss here, see doc

25
Simplenlg

• Being developed, currently working in
• Referring expressions
• Better documentation
• Let me know if you have any comments or
  suggestions!