Applying Embodied Construction Grammar:

1
Applying Embodied Construction Grammar
a description of some Afrikaans morphological
constructions

• Gerhard B van Huyssteen
• Potchefstroom University for CHE
• South Africa
• Acknowledgement Sulené Pilon

ICLC 2003
2
Overview

• HLT and CL in South Africa
• Project Automatic Morphological Analysis of
  Afrikaans
• Requirements of a Formalism
• Two Afrikaans Constructions
• Plural Construction
• Nominalising Construction
• Concluding remarks

3
HLT in South Africa

• CL and NLP
• well-established research fields in USA, Europe,
  and other parts of the world
• unexplored territory in South Africa
• no catholic HLT projects for many years
• Since 2000
• awareness of importance of HLT
• governmental level advisory committee of DACST
  (2002)
• academic level new projects programmes

4
CL at the PUCHE

• Since 2001- prioritised CL as strategically
  important
• establish research focus area Language and
  Technology
• establish first complete graduate study programme
  in CL in South Africa
• set up dedicated HLT laboratory
• acquire text and speech corpora for
• Afrikaans
• South African English
• Setswana
• Two related Afrikaans projects
• Spelling Checker project (funded by University)
• Automatic Morphological Analysis of Afrikaans
  project (funded by NRF)

5
AMAA project

• Aim to develop efficient, reusable modules for
  the automatic morphological analysis of Afrikaans
• tokeniser hyphenator
• word segmenter POS tagger
• compound analyser stemmer
• Project team includes 4 linguists, 1
  computational linguist (from University of
  Tilburg, Netherlands), 2 computer scientists
• Problem communication between
• different disciplines
• different languages

6
In Search of a Formalism

• A formalism is a set of features used to
  precisely and rigorously interpret linguistic
  analysis (i.e. rules, principles, conditions,
  etc.) in logical or mathematical terms, in order
  to develop a calculus (cf. Crystal, 1997 156)
• Looking for
• a formal rule system (i.e. formal grammar or
  formalism)
• for declarative purposes
• not for more procedural purposes (like parsing
  and generation)
• to represent Afrikaans morphological structure
• not particularly interested in syntax, semantics,
  pragmatics

7
Requirements Formalisms

• Accessibility
• Transparent
• Supported by literature
• Efficiency
• Linguistically efficient
• Must be able to capture all linguistic phenomena
  accurately
• Computationally efficient
• To be implemented in a computer environment
• Flexibility
• Describe language structure with ease
• Represent the underlying linguistic theory
• Reusability
• apply in different environments and applications

8
Some specific requirements

• Must represent regexps
• developing a rule-based stemmer, using PERL
• Must rank the rules
• exceptions (i.e. low-level instantiations) are
  ranked higher than rules (i.e. schemas)
• longer rules are ranked higher than shorter
  rules
• DIM construction -tjie is removed before
  jie paaltjie hondjie
• Must be compatible with CG

9
Procedure

• Identify main morphological processes
• Inflection
• Derivation
• Compounding
• Identify constructions
• PLURAL construction
• PAST construction
• NOMINALISING construction
• REDUPLICATION construction
• Draw categorisation networks
• Translate into ECG
• Implement in stemmer

10
Afrikaans Plural Construction

• Inflectional process, realised by means of
  suffixation
• 2 prototypical constructions
• -e hond honde dogs bal balle balls
• -s venster vensters windows tafel tafels
  tables
• Elaborations of the general schema
• e 3 3e 3s
• s ma mas mothers
• Extensions of the general schema
• -a datum data

11
Categorisation Network
GB van Huyssteen (PUCHE)
ICLC 2003
12
PLURAL construction I
construction SUFFIXATION subclass of
AFFIXATION constructional constituents roo
t suffix constraints constituency
rootm/rootf ? suffixm/suffixf
form constraints rootf meets
suffixf suffixf .dependency ? dependent
rootf .dependency ? autonomous
dependent meaning constraints profile-det
? suffix
13
PLURAL construction II
construction PLURAL subclass of
SUFFIXATION constructional evokes INFLECTI
ON constituents root NOUN-SG LET NUM
ABBR suffix PLURAL-SUF constraints ro
otm.scope-of-pred ? BOUNDED-REGION suffixm.scop
e-of-pred ? UNBOUNDED-REGION form meaning c
onstraints scope-of-pred ? UNBOUNDED-REGION
14
PLURAL construction III
construction PLURAL-s subclass of
PLURAL constructional constituents root
NOUN-SG-CN suffix s constraints rootf
/(C)?V(C)Va-z/ suffixf
/s/ rootm.profile ? THING ranking
16 form constraints s /(C)?V(C)Va-z
/(C)?V(C)Va-zs/ meaning constraint
s profile ? THING
15
PLURAL construction IV
construction PLURAL-s subclass of
PLURAL-s constructional constituents root
NOUN-SG-PROPER NOUN-SG-CN LETT NUM ABBR
suffix s constraints rootf
/PROPN(V)/ /CN(iouá)/ /(a-zlm
nrsxz)/ /(1-9123456)/ /ABBR(V
)/ rootm.profile ? THING SAR suffixf
/s/ ranking 13 form constraints s
/PROPN(V)/PROPN(V)s/ s
/CN(V)/CN(V)s/ s /(/a-zlmnrsxz/)/
(a-zlmnrsxzs)/ s /(1-9123456)/
(1-9123456)s/ s /ABBR(V)/ABBR(V)s
/ meaning constraints profile ? THING
16
PLURAL construction V
construction PLURAL-specified subclass of
PLURAL constructional constituents root p
ad sambreel hemp seun bod
Aardklop (lspr)eeu man (m)?eeu
vrou voël kasteel bal oom suffix P
LURAL-SUF constraints ranking
1 form constraints s/pad/paaie/
s/sambreel/sambrele/ s/hemp/hemde/
s/seun/seuns/ s/bod/botte/ s/Aardklop/(Aardk
loppeAardklops) s/(lspr)eeu/(lspr)eeus/ s/ma
n/(mannemans) s/(m)?eeu/(m)?eeue/
s/vrou/(vrouevrouens) s/voël/(voëlsvoële) s/
kasteel/kastele/ s/bal/(balleballas) s/oom/oom
s/ meaning constraints profile ? THING
17
Categorisation Network
GB van Huyssteen (PUCHE)
ICLC 2003
18
NOMINALISING construction I
construction NOMINALISING subclass of
AFFIXATION constructional evokes DERIVATIO
N constituents root VERBADJADV affix
NOM-PREFIXNOM-SUFFIXNOM-CIRCUMFIX constrai
nts rootm.profile ? PROCESSSARCAR affixm.p
rofile ? THING form meaning constraints
profile ? THING
19
NOMINALISING construction II
construction NOMINALISING-ge()Cery subclass
of NOMINALISING-ge()ery constructional consti
tuents root VERB circumfix
ge()ery constraints rootf
/VERB(áéíóúC/ rootm.profile ?
PROCESS circumfixf /ge()Cery/ ranking
1 form constraints s/VERB(áéíóúC/ge
(VERB)(áéíóúCCery/ meaning constraints
20
NOMINALISING construction III
construction NOMINALISING--VCing subclass of
NOMINALISING-ing constructional constituents
root VERB suffix ing constraints
rootf /VERB(VVC)/ rootm.profile ?
PROCESS suffixf /-VCing/ ranking
10 form constraints s/VERB(VVC)/VERB(
VC)ing/ meaning constraints
21
NOMINALISING construction IV
construction NOMINALISING-er subclass of
NOMINALISING-SUF constructional constituents
root VERB suffix er constraints r
ootf /(VERB)/ rootm.profile ?
PROCESS suffixf /er/ ranking
12 form constraints s/(VERB)/(VERB)er
/ meaning constraints attr ? HUMAN
22
Summary of adaptations

• Our adaptations provided for our needs
• added regexps as form constraints
• added ranking as constructional constraints
• added attributes as meaning constraints
• added more CG concepts/constructs
• profile
• valence factors
• profile determinacy
• conceptual and phonological autonomy and
  dependency
• constituency
• correspondence?
• Make it therefore more accessible for us

23
Evaluation ECG as a Declarative Formalism

• Accessible?
• very little ECG material (specifically on
  morphology) available
• isolated do whatever we want to do
• Efficient
• Linguistically efficient?
• handled our data beautifully
• Computationally efficient?
• not our primary concern
• improved communication with computational
  linguist and computer scientists
• Flexibility
• represents essence of Cognitive Linguistics
  beautifully
• easy to add features/adaptations
• Reusable?
• not our primary concern
• Main Advantage
• compatibility with Cognitive Grammar

24
Conclusion

• Your conclusion
• What are we doing wrong?
• What are we missing?
• Are we abusing ECG?