Files
gf-core/lib/resource/doc/spraakdata2005.html
2005-02-28 15:25:52 +00:00

616 lines
12 KiB
HTML

<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN">
<html><head><title></title></head>
<body bgcolor="#ffffff" text="#000000">
<center>
<img src="gf-logo.gif">
<h1>GF Resources for Swedish</h1>
<p>
Språkdata Seminar, Gothenburg, 1 March 2005
</p><p>
Aarne Ranta
</p><p>
<tt>aarne@cs.chalmers.se</tt>
</p></center>
<!-- NEW -->
<h2>Plan</h2>
<a href="01-gf-resource.html">Introduction to resource grammars</a> (pp. 1-16)
<p>
Swedish morphology and lexicon in GF
<p>
Syntax case study: Swedish sentence structure
<p>
Danish and Norwegian through parametrization
<!-- NEW -->
<h2>Swedish morphology and lexicon</h2>
Lexicon: list of words with inflection and other morphological
information (gender of nouns etc).
<p>
Paradigms: set of functions for extending the lexicon.
<!-- NEW -->
<h3>Parts of speech</h3>
A <b>word class</b> is a record type, with
<b>parametric</b> and <b>inherent</b> features (<tt>param</tt>eters).
For example, nouns are the type
<pre>
N = {s : Number => Species => Case => Str ; g : Gender} ;
</pre>
where
<pre>
param
Species = Indef | Def ;
Number = Sg | Pl ;
Case = Nom | Gen ;
</pre>
<!-- NEW -->
<h3>Defining a lexical unit</h3>
Every lexical unit has a word class as its type.
The <b>type checker</b> of GF verifies that all and only the
relevant information of the unit is given. For instance,
an entry for the noun <i>bil</i> ("car") looks as follows.
<pre>
bil =
{s = table {
Sg => table {
Indef => table {Nom => "bil" ; Gen => "bils" } ;
Def => table {Nom => "bilen" ; Gen => "bilens" }
} ;
Pl => table {
Indef => table {Nom => "bilar" ; Gen => "bilars" } ;
Def => table {Nom => "bilarna" ; Gen => "bilarnas" }
}
} ;
g = Utr
}
</pre>
<!-- NEW -->
<h3>The Golden Rule of Functional Programming</h3>
Whenever you find yourself programming by "copy and paste", write
a <b>function</b> instead.
<p>
Thus do <i>not</i> write
<pre>
gran =
{s = table {
Sg => table {
Indef => table {Nom => "gran" ; Gen => "grans" } ;
Def => table {Nom => "granen" ; Gen => "granens" }
} ;
Pl => table {
Indef => table {Nom => "granar" ; Gen => "granars" } ;
Def => table {Nom => "granarna" ; Gen => "granarnas" }
}
} ;
g = Utr
}
</pre>
<!-- NEW -->
<h3>Inflectional paradigms as functions</h3>
Instead, write a <b>paradigm</b> that can be applied to any word
that is "inflected in the same way":
<pre>
decl2 : Str -> N = \bil ->
{s = table {
Sg => table {
Indef => table {Nom => bil + "" ; Gen => bil + "s" } ;
Def => table {Nom => bil + "en" ; Gen => bil + "ens" }
} ;
Pl => table {
Indef => table {Nom => bil + "ar" ; Gen => bil + "ars" } ;
Def => table {Nom => bil + "arna" ; Gen => bil + "arnas" }
}
} ;
g = Utr
}
</pre>
This function can be used over and over again:
<pre>
bil = decl2 "bil" ;
gran = decl2 "gran" ;
dag = decl2 "dag" ;
</pre>
<!-- NEW -->
<h3>High-level definition of paradigms</h3>
Recall: functions instead of copy-and-paste!
<p>
First define (for each word class) a <b>worst-case function</b>:
<pre>
mkN : (apa,apan,apor,aporna : Str) -> Noun =
{s = table {
Sg => table {
Indef => mkCase apa ;
Def => mkCase apan
} ;
Pl => table {
Indef => mkCase apor ;
Def => mkCase aporna
}
} ;
g = case last apan of {
"n" => Utr ;
_ => Neutr
}
</pre>
where we uniformly produce the genitive by
<pre>
mkCase : Str -> Case => Str = \f -> table {
Nom => f ;
Gen => f + case last f of {
"s" | "x" => [] ;
_ => "s"
}
} ;
</pre>
<!-- NEW -->
<h3>High-level definition of paradigms</h3>
Then define, for instance, the five declensions as follows:
<pre>
decl1 : Str -> N = \apa -> let ap = init apa in
mkN apa (apa + "n") (ap + "or") (ap + "orna") ;
decl2 : Str -> N = \bil ->
mkN bil (bil + "en") (bil + "ar") (bil + "arna") ;
decl3 : Str -> N = \fil ->
mkN fil (fil + "en") (fil + "er") (fil + "erna") ;
decl4 : Str -> N = \rike ->
mkN rike (rike + "t") (rike + "n") (rik + "ena") ;
decl5 : Str -> N = \lik ->
mkN lik (lik + "et") lik (lik + "en") ;
</pre>
<!-- NEW -->
<h3>What paradigms are there?</h3>
Swedish nouns traditionally have 5 declensions. But each of them has
slight variations. For instance, the "2nd declension" has the following:
<pre>
gosse - gossar -- 211
nyckel - nycklar -- 231
seger - segrar -- 232
öken - öknar -- 233
hummer - humrar -- 238
kam - kammar -- 241
mun - munnar -- 243
</pre>
and many more (S. Hellberg, <i>The Morphology of Present-Day Swedish</i>,
Almqvist & Wiksell, Stockholm, 1978). In addition, we have at least
<pre>
mås - mås -- genitive form without s
sax - sax
</pre>
<!-- NEW -->
<h3>High-level access to paradigms</h3>
The "naïve user" does not want to go through 500 noun paradigms and
pick the right one.
<p>
A much more efficient method is the one used in
dictionaries: give <i>two</i> (or more) forms instead of one.
Our "dictionary heuristic function" covers the following cases:
<pre>
flicka - flickor
kor - kor (koret)
ko - kor (kon)
ros - rosor (rosen)
bil - bilar
nyckel - nycklar
hummer - humrar
rike - riken
lik - lik (liket)
lärare - lärare (läraren)
</pre>
<!-- NEW -->
<h3>The definition of the dictionary heuristic</h3>
<pre>
reg2Noun : Str -> Str -> Subst = \bil,bilar ->
let
l = last bil ;
b = Predef.tk 2 bil ;
ar = Predef.dp 2 bilar
in
case ar of {
"or" => case l of {
"a" => decl1Noun bil ;
"r" => sLik bil ;
"o" => mkNoun bil (bil + "n") bilar (bilar + "na") ;
_ => mkNoun bil (bil + "en") bilar (bilar + "na")
} ;
"ar" => ifTok Subst (Predef.tk 2 bilar) bil
(decl2Noun bil)
(case l of {
"e" => decl2Noun bil ;
_ => mkNoun bil (bil + "n") bilar (bilar + "na")
}
) ;
"er" => decl3Noun bil ;
"en" => ifTok Subst bil bilar (sLik bil) (sRike bil) ; -- ben-ben
_ => ifTok Subst bil bilar (
case Predef.dp 3 bil of {
"are" => sKikare (init bil) ;
_ => decl5Noun bil
}
)
(decl5Noun bil) --- rest case with lots of garbage
} ;
</pre>
<!-- NEW -->
<h3>When in doubt...</h3>
Test in GF by generating the table
<pre>
> cc mk2N "öken" "öknar"
{s = table Number {
Sg => table {
Indef => table Case {
Nom => "öken" ;
Gen => "ökens"
} ;
Def => table Case {
Nom => "ökenn" ;
Gen => "ökenns"
}
...
}
</pre>
Use the worst-case function if the heuristic does not work:
<pre>
mkN "öken" "öknen" "öknar" "öknarna"
</pre>
<!-- NEW -->
<h3>The module <tt>ParadigmsSwe</tt></h3>
For main category - <tt>N</tt>, <tt>A</tt>, <tt>V</tt> - a worst-case
function and a couple of "regular" patterns.
<pre>
mkN : (apa,apan,apor,aporna : Str) -> N ;
mk2N : (nyckel,nycklar : Str) -> N ;
mkV : (supa,super,sup,söp,supit,supen : Str) -> V ;
regV : (tala : Str) -> V ;
mk2V : (leka,leker : Str) -> V ;
irregV : (dricka, drack, druckit : Str) -> V ;
</pre>
Construction functions for subcategorization.
<pre>
mkV2 : V -> Preposition -> V2 ;
dirV2 : V -> V2 ;
mkV3 : V -> Preposition -> Preposition -> V3 ;
</pre>
<!-- NEW -->
<h3>Morphology extraction</h3>
Idea: search for <b>characteristic forms</b> of paradigms in a corpus.
<pre>
paradigm decl1
= ap+"a"
{ap+"a" & ap+"or" };
</pre>
For instance, if you find <i>klocka</i> and <i>klockor</i>, add
<pre>
klocka_N = decl1 "klocka" ;
</pre>
to the lexicon.
<p>
The notation for extraction and its implementation are
developed by Markus Forsberg and Harald Hammarström.
<!-- NEW -->
<h3>False positives</h3>
Problem: false positives, e.g. <i>bra - bror</i>.
<p>
Solution: restrict stem with a regular expression
<pre>
paradigm decl1 [ap : char* vowel char*]
= ap+"a"
{ap+"a" & ap+"or" };
</pre>
In general, exclude stems shorter than 3 characters.
<p>
It is necessary to check the results manually.
<!-- NEW -->
<h3>Patterns for what?</h3>
"Irregular patterns" are possible, e.g.
<pre>
paradigm vEI [sm:OneOrMore, t:OneOrMore]
= sm+"i"+t+"a"
{sm+"i"+t+"a" & (sm+"e"+t | sm+"i"+t+"it")} ;
</pre>
For rare patterns, it is more productive to build the
corresponding part of lexicon manually.
<!-- NEW -->
<h3>Current Swedish resource lexicon</h3>
49,749 lemmas (1,000 manual, rest extracted),
605,453 word forms.
No subcategorization information.
<p>
Uses the
<a href="http://www.cs.chalmers.se/~markus/FM/">
Functional Morphology</a>
format, which can be translated to GF, XFST, LEXC,
MySQL,...
<p>
FM's "native" analysis engine is based on a trie
and includes compound analysis using algorithms
from G. Huet's
<a href="http://sanskrit.inria.fr/huet/ZEN/">Zen Toolkit</a>.
Analysis speed is 12,000 words per minute
with compound analysis, 50,000 without
(on an Intel M1.5 GHz laptop).
<!-- NEW -->
<h2>Syntax case study: Swedish noun phrases</h2>
Problem: describe agreement and inheritance of definiteness
when a determiner is added to a common noun, possibly modified by
an adjective:
<p>
<i>
en bil<br>
bilen<p>
en stor bil<br>
den stora bilen<p>
denna bil<br>
denna stora bil
</i>
</p>
<!-- NEW -->
<h3>Abstract syntax for noun phrases</h3>
The <b>abstract syntax</b> of a GF grammar defines what grammatical
structures there are, without telling how they are defined.
<p>
The relevant fragment consists of 4 <b>categories</b> and
3 <b>functions</b>
<pre>
cat
N ; -- simple (lexical) common noun, e.g. "bil"
CN ; -- possibly complex common noun, e.g. "stor bil"
Det ; -- determiner, e.g. "denna"
NP ; -- noun phrase, e.g. "bilen"
AP : -- adjectival phrase, e.g. "stor"
fun
UseN : N -> CN ;
UseA : A -> AP ;
DetCN : Det -> CN -> NP ;
ModA : A -> CN -> CN ;
</pre>
<!-- NEW -->
<h3>Types of complex nouns and noun phrases</h3>
Just like to words, we assign <b>linearization types</b> to
phrase categories. They are similar to the lexical types,
but often with some extra information.
<pre>
lincat
CN = {s : Number => SpeciesP => Case => Str ; g : Gender ; isComplex : Bool} ;
NP = {s : NPForm => Str ; g : Gender ; n : Number ; p : Person} ;
Det = {s : Gender => Str ; n : Number ; b : SpeciesP} ;
AP = {s : AdjFormPos => Case => Str} ;
</pre>
Here we use some new parameter types:
<pre>
param
SpeciesP = IndefP | DefP Species ;
NPForm = PNom | PAcc | PGen GenNum ;
GenNum = ASg Gender | APl ;
AdjFormPos = Strong GenNum | Weak ;
</pre>
<!-- NEW -->
<h3>Building noun phrases with a determiner</h3>
Mutual agreement:
<ul>
<li> the determiner gets the gender of the noun
<li> the noun gets the number and definiteness of the determiner
</ul>
<pre>
DetCN : Det -> CN -> NP = \en, man ->
{s = \\c => en.s ! man.g ++
man.s ! en.n ! en.b ! npCase c ;
g = genNoun man.g ;
n = en.n ;
p = P3
} ;
</pre>
<!-- NEW -->
<h3></h3>
<!-- NEW -->
<h3></h3>
<!-- NEW -->
<h2>Syntax case study: Swedish sentence structure</h2>
Data: freedom in word order in main clause
<p>
<i>
jag har inte sett dig idag<br>
dig jag har inte sett idag<br>
idag har jag inte sett dig<br>
inte har jag sett dig idag<br>
sett har jag inte dig idag (??)<br>
sett dig har jag inte idag<p>
</i>
Rigid order in questions...
<p>
<i>
har jag inte sett dig idag
</i>
<p>
... and in subordinate clauses
<p>
<i>
jag inte har sett dig idag
</i>
<p>
<!-- NEW -->
<h3>The topological model</h3>
<!-- NEW -->
<h3>The <tt>Sats</tt> data type</h3>
<!-- NEW -->
<h3>Building clauses from <tt>Sats</tt></h3>
<!-- NEW -->
<h3>Construction of <tt>Sats</tt></h3>
Notice: we want to treat <tt>Sats</tt> as an abstract data type.
<!-- NEW -->
<h3>Verb subcategorization patterns formalized</h3>
<!-- NEW -->
<h3>Adding adverbials</h3>
<!-- NEW -->
<h3>Coverage of verb patterns in Swedish Academy Grammar</h3>
<!-- NEW -->
<h3>Remaining problems</h3>
<!-- NEW -->
<h2>Danish and Norwegian through parametrization</h2>
</body>
</html>