Corpus Annotation The Index Thomisticus Treebank

1
Corpus AnnotationThe Index Thomisticus Treebank

• Marco Passarotti
• Università Cattolica del Sacro Cuore
• Milan

2
1.Corpus annotation
3
Annotation

• Annotation is the practice of
• adding to corpus data different kinds of
  meta-data
• Editorial (relationship corpus original
  source) additions/omissions, corrections,
  normalizations
• Descriptive (classificatory info) bibliographic
  info, author, year, language
• Administrative (documentary info) file and tags
  description, availability of the corpus, revision
  status
• Analytic (interpretation and analysis of corpus
  components) structural features (e.g. text
  subdivision), quotations, foreign words,
  named-entity recognition (names, addresses,
  dates, measures), linguistic information
  (corpus annotation)
• Leech (2004) Corpus annotation is the practice
  of adding interpretative linguistic information
  to a corpus

4
Layers of linguistic annotation

• Digital text tokenization
• (spoken corpora) Phonetic and prosodic (tone,
  stress)
• Lemmatization (lmz)
• morphological
• morpho-syntactic
• Syntactic analysis (treebanks)
• Topic-focus articulation anaphora resolution
• Pragmatic-rhetorical annotation (RST speech
  acts, illocutionary force)
• Semantic annotation word-sense disambiguation
• Multimodal annotation of speech word, gesture,
  facial expression, intonation, perceivable
  context

5
(No Transcript)
6
2.TreebanksCorpora annotated at syntactic level
7
PhSG vs. DG

• PhSG (Chomsky, Schützenberger)
• Words, PoS, Phrases, Start Symbols
• Set inclusion categorization (e.g. word, N, NP,
  S)
• DG (Tesnière, Melcuk, etc.)
• Only words (Chomskys terminals)
• Head-dependent relations between words
• Lexical nodes are linked through binary relations
  (dependencies) and are annotated with functional
  categories
• Sentence word order not marked
• Suitable for free-word-order languages (richly
  inflected) Dutch (Alpino), Czech (PDT), Latin
  (IT-TB LDT)

8
3.Latin treebanks
9
Latin features

• Richly inflected
• Much homonymy
• Free word order non-projective dependencies
• if we put the words in the linear order the
  edges cannot be drawn above the words without
  crossings (McDonald, Hajic, Pereira, Ribarov,
  2005)

that glory would know my old age (Prop., I.8.46)
10
IT-Treebank and LDT

• A collaborative project (official partnership)
  between Università Cattolica (Milan) and the
  Perseus Project (Tufts University Boston)
• Latin Dependency Treebank texts from Classical
  era (around 35,000 tokens annotated)
• Index Thomisticus Treebank (Busa) Thomas Aquinas
  opera omnia (around 35,000 tokens annotated)
• Dependency Grammar (via PDT) a single annotation
  manual (guidelines)

11
4.Samples from IT-TBhttp//itreebank.marginalia.
it
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
5.Research perspectives
20
Research perspectives

• IT-TB
• Manual syntactic annotation
• Train data-driven NLP tools (Parsers, PoS
  Taggers) shallow parsing for semi-automatic
  annotation
• Dynamic valency-frames lexicon
• Updating the IT with new texts and new critical
  editions
• Enhancing text and manuscript images (Navarra
  University)
• IT-TB LDT
• Joint workshops (TLT 2009?)
• Annotation back-off (comparison)
• Consistency checking in related annotation (e.g.
  n-gram variation by Dickinson and Meurers)