Ch 3: ENGLISH MORPHOLOGY (USING FSAs)

2-level system for building forms of words (e.g. plurals): 

1. morphological rules for the concept
2. then spelling rules for the details

morphological rules - how to form the plural (or other cases) in English
<regular noun> --> <regular noun> + '-s' 
fish --> fish
goose --> geese
etc.

spelling rules - baby --> babies
(we're still adding -s, just the details of how the output looks
have changed)

spelling rules change only the _surface form_

--

Need for a lexicon (dictionary):
potato --> potatoes
tomato --> tomatoes
photo --> photos
mango --> mangoes, mangos

--

SOME TERMINOLOGY:

morpheme - minimal unit of meaning
e.g. 'b' doesn't have a meaning

morphemes can be categorized as stems or affixes

affixes = prefix, suffix, infix
(English doesn't have infixes)

--

BUILDING CORRECT MORPHOLOGICAL FORM

Input: form in lexicon (= database = dictionary)
- base form
- features (e.g +PL = plural)

Output: correctly spelled word

Processing: 2 levels

1. morphological rules - how to form the plural (or other cases) in English
e.g. <regular noun> --> <regular noun> + '-s' 

2. spelling rules
e.g baby +PL --> babies

--

MORPHOLOGICAL PARSING

= inverse of building morphological form

input: correctly spelled word

output: root form and feature list

e.g. geese --> goose +PL

parsing - given an item (word, sentence, etc.), find its structure

--

LANGUAGE STRUCTURE AFFECTS DIFFICULTY OF MORPHOLOGICAL PARSING

ENGLISH IS EASY:

English only has a few affixes (-s, -ed, -ing)

--

English HAS CONCATENATIVE MORPHOLOGY

some languages have non-concatenative morphology
e.g. Bahasa Indonesia
orang --> orang-orang

--

HIGHLY INFLECTED LANGUAGES REQUIRE MORE WORK:

Other languages have more _case markers_ on nouns.

e.g. affixes representing +FROM, +TO, etc.

You might consider 's to be a case marker in English, i.e. +POSS
(possessive), but there are no others.

Many languages have 4-5 cases; a few have 10-15.


Highly inflective languages can also have many affixes for verbs:

In Spanish, +FUTURE, +UNREAL, and +1sing ("I") are all combined to
index into a table of affixes (i.e. 'amare' = 'that I might love')

--

AGGLUTINATIVE LANGUAGES HAVE DIFFERENT ISSUES:

Compare the difference between a language like Spanish
(fig. 3.1) and an agglutinative language like Turkish or Finnish.

In Turkish, each of these components of meaning will have a separate
affix.

(Might be easier to build, but there are combination rules, such as
vowel harmony...)

What about parsing?


Here are some data about Turkish:

Corpus      Word       Word       Distinct  Distinct  Compression   
            Tokens     Types      Terms     Stems     (%)           
            (instances) (unique
                         words)

Turkish     376,187    49,479     41,370     6,363          84.6    

English     567,574    19,044     18,348    11,671          36.4    

--

ROOT-AND-PATTERN LANGUAGES HAVE THEIR OWN ISSUES:

Semitic languages (Hebrew, Aramaic, Arabic, etc.) have template-based
morphology = root and pattern system

In these languages, roots have only consonants.

There are 3 levels of derivation:

1. root --> stem ("word")

2. stem --> inflected form

3. inflected form --> final form
(spelling/pronunciation rules)

Derivational affixes are not important in these languages; the
patterns take their place.


e.g. Hebrew

KTB = 'write' (root)


Here are a number of verb stems derived from KTB:

KaTaB   = to write
KiTTeiB = to inscribe (+INTENSIVE)
NKTB    = to be written (+PASSIVE)
HKTiB   = to cause to write (+CAUSATIVE)
HuKTB   = to caue to be written (+CAUSATIVE +PASSIVE)
HTKTB   = to write itself (+REFLEXIVE)

Each of these forms is a verb with inflectional forms (I wrote, you
wrote, etc.)

These patterns are semi-productive. Verb patterns are more likely to be
productive than noun patterns.


E.g. some noun pattern for KTB:

KTiBa   = signature
KTuBa   = marriage license


(Spelling/pron. rules: KTiBa --> k'tiva)

--

WHAT IS THE ROLE OF DERIVATIONAL AFFIXES?

2 ways to build words - inflection and derivation

An example of derivational affixes in English:

  antidisestablishmentarianism

One of the longest English words that is not made up.


anti-        opposition to
dis-         the removal of
establish    (root) to create/maintain something
-ment        the activity of
-arian       people who believe in
-ism         the philosophy of


So we are talking about the philosophy of people who believe in
opposing the removal of something.

= people who are in favor of maintaining something.

(In this case, the 'something' is maintaining the Church of England as the
official state church in the 19th century.)

--

PRODUCTIVE SUFFIXES:

productive suffix - can apply to anything of the right category,
e.g. '-s'

in general, inflectional suffixes are productive, but derivational
suffixes aren't

derivational suffixes: -tion, -ee, -er, -ize, etc.

Derivational suffixes are semi-productive:

- can't always apply them

- don't have precise meanings like inflectional suffixes

--

STEMMING

Why do we want to do stemming?

- NLU (natural language understanding)
- machine translation
- information retrieval?


CAN WE DO STEMMING WITHOUT A LEXICON?

Why do we want to do stemming without a lexicon?

Porter (1980)

App. B of textbook (pp. 833-836)

Problems w/ Porter stemmer (textbook, p. 83)


COPING WITH ENGLISH SPELLING/PRONUNCIATION:

Soundex algorithm (textbook, p. 89):

words that sound similar --> same code number