1. Ehud Reiter, Computing Science, University of Aberdeen1 CS4025: Realisation and Entrepeneurship

2. Ehud Reiter, Computing Science, University of Aberdeen2 Realisation l Third (last) NLG stage l Generate actual text l 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. Ehud Reiter, Computing Science, University of Aberdeen3 Realisation l Problem: There are lots of finicky details of language which most people developing NLG systems don’t want to worry about l Solution: Automate this using a realiser l

4. Ehud Reiter, Computing Science, University of Aberdeen4 Syntax l Sentences must obey the rules of English grammar »Specifies which order words should appear in, extra function words, word forms l Many aspects of grammar are somewhat bizarre

5. Ehud Reiter, Computing Science, University of Aberdeen5 Syntactic Details: Verb Group l Verb group is the main verb plus helping words (auxiliaries). l 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. Ehud Reiter, Computing Science, University of Aberdeen6 Verb group l 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 does not watch TV vs John watch not TV »Exception to exception: use first strategy if verb is form of BE –John is not happy vs John does not be happy

7. Ehud Reiter, Computing Science, University of Aberdeen7 Realiser l Just tell realiser verb, tense, whether negated, and it will figure out the VG »(watch, future) -> will watch »(watch, past, negated) -> did not watch »Etc l Similarly automate other “obscure” encodings of information

8. Ehud Reiter, Computing Science, University of Aberdeen8 Morphology l Words have different forms l Nouns have plural »Dog, dogs l Verbs have 3ps form, participles »Watch, watches, watching, watched l Adjectives have comparative, superlative »Big, bigger, biggest

9. Ehud Reiter, Computing Science, University of Aberdeen9 Formation of variants l 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. Ehud Reiter, Computing Science, University of Aberdeen10 Realiser l Calculates variants automatically »(dog, plural) -> dogs »(box, plural) -> boxes »(child, plural) -> children »etc

11. Ehud Reiter, Computing Science, University of Aberdeen11 Punctuation l Rules for structures »Sentences have first word capitalised, end in a full stop –My dog ate the meat. »Lists have conjunction (eg, and) between last two elements, comma between others –I saw Tom, Sue, Zoe, and Ciaran at the meeting. »Etc l Realiser can automatically insert appropriate punctuation for a structure

12. Ehud Reiter, Computing Science, University of Aberdeen12 Punctuation l Rules on combinations of punc »Don’t 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.). l Again realiser can automate

13. Ehud Reiter, Computing Science, University of Aberdeen13 Pouring l Usually we insert spaces between tokens, but not always »My dog »I saw John, and said hello. »I saw John, and said hello l Automated by realiser

14. Ehud Reiter, Computing Science, University of Aberdeen14 Pouring l 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 l Realiser automates

15. Ehud Reiter, Computing Science, University of Aberdeen15 Output formatting l Many possible output formats »Simple text »HTML »RTF (MS Word) l Realiser can automatically add appropriate markups for this

16. Ehud Reiter, Computing Science, University of Aberdeen16 Realiser systems l simplenlg – relatively limited functionality, but well documented, fast, easy to use, tested l KPML – lots of functionality but poorly documented, buggy, slow l openccg – somewhere in between l Many more

17. Ehud Reiter, Computing Science, University of Aberdeen17 (Montreal) French simplenlg l Vaudry and Lapalme, 2013 l Lots of “silly” rules, like English. l Eg, negation »il ne parle pas –“he does not speak” »il ne parle plus –“he does not speak anymore” »personne ne parle –“nobody speaks”

18. Ehud Reiter, Computing Science, University of Aberdeen18 Morphophonology l English: only example is a/an »An apple vs A banana l French: much more prominent »le + homme → l’homme »la + honte → la honte »le + beau + homme → le bel homme »à + le → au

19. Ehud Reiter, Computing Science, University of Aberdeen19 Other languages l Simplenlg for German, Portuguese »Each language has its “quirks” l Most challenging/different is tribal languages, eg from New Guinea »Allman et al, 2012

20. Ehud Reiter, Computing Science, University of Aberdeen20 Summary l Realiser automates the finicky details of language »So NLG developer doesn’t have to worry about these »One of the advantages of NLG

21. Ehud Reiter, Computing Science, University of Aberdeen21 Company: 2009 l 2009: Data2text set up »Commercialising research in NLG and data-to-text, esp SumTime and Babytalk »Ehud Reiter, Yaji Sripada, 2 business people, 2 software developers »Crowded into Meston 215

22. Ehud Reiter, Computing Science, University of Aberdeen22 Company: 2013 l 2013: Arria/Data2text goes public »Arria is London-based outfit which bought Data2text –Sales, marketing, corporate, IT architects, project managers to complement techies »Listed on AIM stock market in Dec 2013 »About 40 employees

23. Ehud Reiter, Computing Science, University of Aberdeen23 Arria/Data2text l Web site »Vision (Dale video) »People »Clients (case studies) »Stock market (AIM) »Demos

24. Ehud Reiter, Computing Science, University of Aberdeen24 Responsibilities l Employee (“meet the payroll”) l Investor (“how is my money doing”) l Client (“fix this yesterday”) l End user (“how is my baby doing”)

25. Ehud Reiter, Computing Science, University of Aberdeen25 Success l NLG/data-to-text essential, but only a small part of overall story »“boring” IT »Support »Change management »Sales and marketing

26. Ehud Reiter, Computing Science, University of Aberdeen26 Questions