Transcript Document

School of Computing
something
FACULTY OF ENGINEERING
OTHER
Tokenization and Morphology
Eric Atwell, Language Research Group
(with thanks to Katja Markert, Marti Hearst,
and other contributors)
Reminder
The main areas of linguistics
Rationalism: language models based on expert introspection
Empiricism: models via machine-learning from a corpus
Corpus: text selected by language, genre, domain, …
Brown, LOB, BNC, Penn Treebank, MapTask, CCA, …
Corpus Annotation: text headers, PoS, parses, …
Corpus size is no. of words – depends on tokenisation
We can count word tokens, word types, type-token distribution
Lexeme/lemma is “root form”, v inflections (be v am/is/was…)
What’s a word?
How many words do you find in the following short text?
What is the biggest/smallest plausible answer to this question?
What problems do you encounter?
It’s a shame that our data-base is not up-to-date. It is a
shame that um, data base A costs $2300.50 and that
database B costs $5000. All databases cost far too much.
Time: 3 minutes
Counting words: tokenization
Tokenisation is a processing step where the input text is
automatically divided into units called tokens where each is either a word
or a number or a punctuation mark…
So, word count can ignore numbers, punctuation marks (?)
Word: Continuous alphanumeric characters delineated by whitespace.
Whitespace: space, tab, newline.
BUT dividing at spaces is too simple: It’s, data base
Another approach is to use regular expressions to specify which substrings
are valid words.
Regular expressions for tokenization
• wordr
= r'(\w+)‘
• hyphen = r'(\w+\-\s?\w+)‘
• Eg data-base, Allows for a space after the hyphen
• apostrophe = r'(\w+\'\w+)‘
• Eg isn’t
• numbers = r'((\$|#)?\d+(\.)?\d+%?)‘
• Needs to handle large numbers with commas
Some Tokenization Issues
Sentence Boundaries
• Punctuation, eg quotation marks around sentences?
• Periods – end of line or not?
Proper Names
• What to do about
• “New York-New Jersey train”?
• “California Governor Arnold Schwarzenegger”?
Contractions
• That’s Fred’s jacket’s pocket.
• I’m doing what you’re saying “Don’t do!”.
Jabberwocky Analysis
This is nonsense … or is it?
This is not English … but it’s much more like English than it is
like French or German or Chinese or …
Why do we pretty much understand the words?
Jabberwocky Analysis
Why do we pretty much understand the words?
We recognize combinations of morphemes.
• Chortled - Laugh in a breathy, gleeful way; (Definition from Oxford
American Dictionary) A combination of "chuckle" and "snort."
• Galumphing - Moving in a clumsy, ponderous, or noisy manner.
Perhaps a blend of "gallop" and "triumph." (Definition from Oxford
American Dictionary)
Activity:
• Make up a word whose meaning can be inferred from the morphemes
that you used.
Jabberwocky Analysis
Why do we pretty much understand the words?
• Surrounding English words strongly indicate the parts-of-speech of
the nonsense words.
• toves: probably can perform an action
(because they did gyre and gimble)
• wabe: is probably a place.
(they did … in the wabe)
http://assets.cambridge.org/052185/542X/excerpt/052185542X_excerpt.pdf
Jabberwocky Analysis
• Surrounding English words strongly indicate the parts-of-speech of
the nonsense words.
• It’s similar in the French Translation:
Example from http://www.departments.bucknell.edu/linguistics/lectures/05lect02.html
Morphology
Morphology:
• The study of the way words are built up from smaller meaning units.
Morphemes:
• The smallest meaningful unit in the grammar of a language.
Contrasts:
• Derivational vs. Inflectional
• Regular vs. Irregular
• Concatinative vs. Templatic (root-and-pattern)
A useful resource:
• Glossary of linguistic terms by Eugene Loos
• http://www.sil.org/linguistics/GlossaryOfLinguisticTerms/contents.htm
Examples (English)
“unladylike”
• 3 morphemes, 4 syllables
un-
‘not’
lady ‘(well behaved) female adult human’
-like ‘having the characteristics of’
• Can’t break any of these down further without distorting the
meaning of the units
“technique”
• 1 morpheme, 2 syllables
“dogs”
• 2 morphemes, 1 syllable
-s, a plural marker on nouns
Morpheme Definitions
Root
• The portion of the word that:
• is common to a set of derived or inflected forms, if any, when all affixes
are removed
• is not further analyzable into meaningful elements
• carries the principle portion of meaning of the words
Stem
• The root or roots of a word, together with any derivational affixes, to which
inflectional affixes are added.
Affix
• A bound morpheme that is joined before, after, or within a root or stem.
Clitic
• a morpheme that functions syntactically like a word, but does not appear
as an independent phonological word
• Arabic: al in Al-Qaeda (definite particle)
• English: ‘s in Hal’s (genitive particle)
Inflectional vs. Derivational
Word Classes
• Parts of speech: noun, verb, adjectives, etc.
• Word class dictates how a word combines with morphemes to form new
words
Inflection:
• Variation in the form of a word, typically by means of an affix, that expresses
a grammatical contrast.
• Doesn’t change the word class
• Usually produces a predictable, nonidiosyncratic change of meaning.
• run -> runs | running | ran
Derivation:
• The formation of a new word or inflectable stem from another word or stem.
• compute -> computer -> computerization
Inflectional Morphology
Adds:
• tense, number, person, mood, aspect
Word class doesn’t change
Word serves new grammatical role
Examples
• come is inflected for person and number:
The pizza guy comes at noon.
• las and rojas are inflected for agreement with manzanas in grammatical
gender by -a and in number by –s
las manzanas rojas (‘the red apples’)
Derivational Morphology
Word class changes: verb  noun, noun  adjective etc
Nominalization (formation of nouns from other parts of speech,
primarily verbs in English):
• computerization
• appointee
• killer
• fuzziness
Formation of adjectives (primarily from nouns)
• computational
• clueless
• Embraceable
Difficult cases:
• building  from which word-class and sense of “build”?
Concatinative Morphology
Morpheme+Morpheme+Morpheme+…
Stems: also called lemma, base form, root, lexeme
•
hope+ing  hoping
hop  hopping
Affixes
• Prefixes: Antidisestablishmentarianism
• Suffixes: Antidisestablishmentarianism
• Infixes: hingi (borrow) – humingi (borrower) in Tagalog
• Circumfixes: sagen (say) – gesagt (said) in German
Agglutinative Languages
• uygarlaştıramadıklarımızdanmışsınızcasına
• uygar+laş+tır+ama+dık+lar+ımız+dan+mış+sınız+casına
•
Behaving as if you are among those whom we could not cause to become civilized
Templatic Morphology
Roots and Patterns
• Example: Hebrew or Arabic or Amharic (spoken in Ethiopia)
• Root:
• Consists of 3 consonants CCC
• Carries basic meaning
• Template:
• Gives the ordering of consonants and vowels
• Specifies semantic information about the verb
• Active, passive, middle voice
• Example (Hebrew):
• lmd (to learn or study)
• CaCaC -> lamad (he studied)
• CiCeC -> limed (he taught)
• CuCaC -> lumad (he was taught)
Morphological Analysis Tools
Porter stemmer:
• A simple approach: just hack off the end of the word!
• Frequently used, especially for Information Retrieval, but results are
pretty ugly!
porter.demo()
Original *****************************
Pierre Vinken , 61 years old , will join the board as a nonexecutive
director Nov. 29 . Mr. Vinken is chairman of Elsevier N.V. , the Dutch
publishing group . Rudolph Agnew , 55 years old and former chairman of
Consolidated Gold Fields PLC , was named a nonexecutive director of
this British industrial conglomerate . A form of asbestos once used to
make Kent cigarette filters has caused a high percentage of cancer
deaths among a group of workers exposed to it more than 30 years ago ,
researchers reported .
Results *******************************
Pierr Vinken , 61 year old , will join the board as a nonexecut
director Nov. 29 . Mr. Vinken is chairman of Elsevi N.V. , the Dutch
publish group . Rudolph Agnew , 55 year old and former chairman of
Consolid Gold Field PLC , wa name a nonexecut director of thi British
industri conglomer . A form of asbesto onc use to make Kent cigarett
filter ha caus a high percentag of cancer death among a group of
worker expos to it more than 30 year ago , research report .
Morphological Analysis Tools
WordNet’s morphy()
• A slightly more sophisticated approach
• Use an understanding of inflectional morphology
• Uses a set of Rules of Detachment
• Use an Exception List for irregulars
• Handle collocations in a special way
• Do the transformation, compare the result to the WordNet
dictionary
• If the transformation produces a real word, then keep it, else use
the original word.
• For more details, see
• http://wordnet.princeton.edu/man/morphy.7WN.html
Some morphy() output
>>> wntools.morphy('dogs')
'dog'
>>> wntools.morphy('running', pos='verb')
'run'
>>> wntools.morphy('corpora')
'corpus'
>>>
Morphological Analysis Tools
Very sophisticated programs have been developed
Use a techniqued called Two-Level Phonology
• Has been applied to numerous languages
Best known: PCKimmo
• After Kimmo Koskenniemi, based in part on work by Lauri Kartunnen in 1983
• Uses:
• A rules file which specifies the alphabet and the phonological (or spelling) rules,
• A lexicon file which lists lexical items and encodes morphotactic constraints.
• http://www.sil.org/pckimmo/
Commercial versions are available
• inXight’s LinguistX version based on technology developed by Kaplan and others
from Xerox PARC (or at least used to be)
Morphological Analysis Tools
“cheat”: store all variants in a dictionary database, eg
CatVar:
• Categorial Variation Database
• “A database of clusters of uninflected words (lexemes) and their
categorial (i.e. part-of-speech) variants.”
• Example: the developing cluster:(develop(V), developer(N),
developed(AJ), developing(N), developing(AJ), development(N)).
http://clipdemos.umiacs.umd.edu/catvar
based on published dictionaries: LDOCE, CELEX, OALD++,
PROPOSEL ...
MorphoChallenge
One problem with rule-based systems (PCkimmo) or dictionarylookup systems: Porting to new languages
In principle, Unsupervised Machine Learning could learn from
any language data-set, by finding recurring patterns which
correspond to roots, prefixes, postfixes
MorphoChallenge is a contest to find the best UML
morphological analyser
http://www.cis.hut.fi/morphochallenge2005/
http://www.cis.hut.fi/morphochallenge2007/
http://www.cis.hut.fi/morphochallenge2008/
Atwell, Roberts: Combinatory Hybrid Elementary Analysis of Text
http://www.cis.hut.fi/morphochallenge2005/P07_Atwell.pdf
Arabic morphological analysis
Arabic is particularly challenging - different script, infixes,
vowels may be left out in written Arabic …
Leeds researcher Majdi Sawalha: online analysis tool
http://www.comp.leeds.ac.uk/sawalha/
Sawalha, Majdi; Atwell, Eric (2010). Fine-Grain Morphological
Analyzer and Part-of-Speech Tagger for Arabic Text. in:
Proceedings of the Language Resource and Evaluation
Conference LREC 2010, 17-23 May 2010, Valetta, Malta.
http://www.comp.leeds.ac.uk/sawalha/sawalha10lrecB.pdf
Reminder
Tokenization - by whitespace, regular expressions
Problems: It’s data-base New York …
Jabberwocky shows we can break words into morphemes
Morpheme types: root/stem, affix, clitic
Derivational vs. Inflectional
Regular vs. Irregular
Concatinative vs. Templatic (root-and-pattern)
Morphological analysers: Porter stemmer, Morphy, PC-Kimmo
Morphology by lookup: CatVar, CELEX, OALD++
Unsupervised Machine Learning: MorphoChallenge