msyds/gf-core
1
0
Fork 0
forked from GitHub/gf-core
Code Pull Requests Activity
Files
c5eea4b375a67de8e2d74229d178cdb52cbde4df
gf-core/doc/multimodal.html
aarne 59c4536d5e fixes in multimodal document, last section
2006-01-09 19:30:27 +00:00

864 lines
29 KiB
HTML
Raw Blame History

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML>
<HEAD>
<META NAME="generator" CONTENT="http://txt2tags.sf.net">
<TITLE>Demonstrative Expressions and Multimodal Grammars</TITLE>
</HEAD><BODY BGCOLOR="white" TEXT="black">
<P ALIGN="center"><CENTER><H1>Demonstrative Expressions and Multimodal Grammars</H1>
<FONT SIZE="4">
<I>Author: Aarne Ranta &lt;aarne (at) cs.chalmers.se&gt;</I><BR>
Last update: Mon Jan 9 20:29:45 2006
</FONT></CENTER>
<P></P>
<HR NOSHADE SIZE=1>
<P></P>
<UL>
<LI><A HREF="#toc1">Abstract</A>
<LI><A HREF="#toc2">Multimodal grammars</A>
<UL>
<LI><A HREF="#toc3">Representing demonstratives in semantics and grammar</A>
<LI><A HREF="#toc4">Asynchronous syntax in GF</A>
<LI><A HREF="#toc5">Example multimodal grammar: abstract syntax</A>
<LI><A HREF="#toc6">Digression: discontinuous constituents</A>
<LI><A HREF="#toc7">From grammars to dialogue systems</A>
</UL>
<LI><A HREF="#toc8">Adding multimodality to a unimodal grammar</A>
<UL>
<LI><A HREF="#toc9">The multimodal conversion</A>
<LI><A HREF="#toc10">An example of the conversion</A>
<LI><A HREF="#toc11">Multimodal conversion combinators</A>
</UL>
<LI><A HREF="#toc12">Multimodal resource grammars</A>
<UL>
<LI><A HREF="#toc13">Resource grammar API</A>
<LI><A HREF="#toc14">Multimodal API: functions for building demonstratives</A>
<LI><A HREF="#toc15">Multimodal API: functions for building sentences and phrases</A>
<LI><A HREF="#toc16">Language-independent implementation: examples</A>
<LI><A HREF="#toc17">Multimodal API: interface to unimodal expressions</A>
<LI><A HREF="#toc18">Instantiating multimodality to different languages</A>
<LI><A HREF="#toc19">Language-independent reimplementation of TramDemo</A>
<LI><A HREF="#toc20">The order problem</A>
<LI><A HREF="#toc21">A recipe for using the resource library</A>
</UL>
</UL>
<P></P>
<HR NOSHADE SIZE=1>
<P></P>
<A NAME="toc1"></A>
<H2>Abstract</H2>
<P>
This document shows a method to write grammars
in which spoken utterances are accompanied by
pointing gestures. A computer application of such
grammars are <B>multimodal dialogue systems</B>, in
which the pointing gestures are performed by
mouse clicks and movements.
</P>
<P>
After an introduction to the notions of
<B>demonstratives</B> and <B>integrated multimodality</B>,
we will show by a concrete example
how multimodal grammars can be written in GF
and how they can be used in dialogue systems.
The explanation is given in three stages:
</P>
<OL>
<LI>How to write a multimodal grammar by hand.
<LI>How to add multimodality to a unimodal grammar.
<LI>How to use a multimodal resource grammar.
</OL>
<A NAME="toc2"></A>
<H2>Multimodal grammars</H2>
<P>
<B>Demonstrative expressions</B> are an old idea. Such
expressions get their meaning from the context.
</P>
<BLOCKQUOTE>
<I>This train</I> is faster than <I>that airplane</I>.
</BLOCKQUOTE>
<P></P>
<BLOCKQUOTE>
I want to go from <I>this place</I> to <I>this place</I>.
</BLOCKQUOTE>
<P></P>
<P>
In particular, as in these examples, the meaning
can be obtained from accompanying pointing gestures.
</P>
<P>
Thus the meaning-bearing unit is neither the words nor the
gestures alone, but their combination. Demonstratives
thus provide an example of <B>integrated multimodality</B>,
as opposed to parallel multimodality. In parallel
multimodality, speech and other modes of communication
are just alternative ways to convey the same information.
</P>
<A NAME="toc3"></A>
<H3>Representing demonstratives in semantics and grammar</H3>
<P>
When formalizing the semantics of demonstratives, we can combine syntax with coordinates:
</P>
<BLOCKQUOTE>
I want to go from this place to this place
</BLOCKQUOTE>
<P></P>
<P>
is interpreted as something like
</P>
<PRE>
want(I, go, this(place,(123,45)), this(place,(98,10)))
</PRE>
<P>
Now, the same semantic value can be given in many ways, by performing
the clicks at different points of time in relation to the speech:
</P>
<BLOCKQUOTE>
I want to go from this place CLICK(123,45) to this place CLICK(98,10)
</BLOCKQUOTE>
<P></P>
<BLOCKQUOTE>
I want to go from this place to this place CLICK(123,45) CLICK(98,10)
</BLOCKQUOTE>
<P></P>
<BLOCKQUOTE>
CLICK(123,45) CLICK(98,10) I want to go from this place to this place
</BLOCKQUOTE>
<P></P>
<P>
How do we build the value compositionally in parsing?
Traditional parsing is sequential: its input is a string of tokens.
It works for demonstratives only if the pointing is adjacent to
the spoken expression. In the actual input, the demonstrative word
can be separated from the accompanying click by other words. The two
can also be simultaneous.
</P>
<A NAME="toc4"></A>
<H3>Asynchronous syntax in GF</H3>
<P>
What we need is a notion of <B>asynchronous parsing</B>, as opposed to
sequential parsing (where demonstrative words and clicks must be
adjacent).
</P>
<P>
We can implement asynchronous parsin in GF by exploiting the generality
of <B>linearization types</B>. A linearization type is the type of
the <B>concrete syntax objects</B> assigned to semantic values.
What a GF grammar defines is a relation
</P>
<PRE>
abstract syntax trees &lt;---&gt; concrete syntax objects
</PRE>
<P>
When modelling context-free grammar in GF,
the concrete syntax objects are just strings.
But they can be more structured objects as well - in general, they are
<B>records</B> of different kinds of objects. For example,
a demonstrative expression can be linearized into a record of two strings.
</P>
<PRE>
{s = "this place" ;
this place (coord 123 45) &lt;---&gt; p = "(123,45)"
}
</PRE>
<P>
The record
</P>
<PRE>
{s = "I want to go from this place to this place" ;
p = "(123,45) (98,10"
}
</PRE>
<P>
represents any combination of the sentence and the clicks, as long
as the clicks appear in this order.
</P>
<A NAME="toc5"></A>
<H3>Example multimodal grammar: abstract syntax</H3>
<P>
A simple example of a multimodal GF grammar is the one called
the Tram Demo grammar. It was written by Björn Bringert within
the TALK project as a part of a dialogue system that
deals with queries about tram timetables. The system interprets
a speech input in combination with mouse clicks on a digital map.
</P>
<P>
The abstract syntax of (a minimal fragment of) the Tram Demo
grammar is
</P>
<PRE>
cat
Input, Dep, Dest, Click ;
fun
GoFromTo : Dep -&gt; Dest -&gt; Input ; -- "I want to go from x to y"
DepHere : Click -&gt; Dep ; -- "from here" with click
DestHere : Click -&gt; Dest ; -- "to here" with click
CCoord : Int -&gt; Int -&gt; Click ; -- click coordinates
</PRE>
<P>
An English concrete syntax of the grammar is
</P>
<PRE>
lincat
Input, Dep, Dest = {s : Str ; p : Str} ;
Click = {p : Str} ;
lin
GoFromTo x y = {s = ["I want to go"] ++ x.s ++ y.s ; p = x.p ++ y.p} ;
DepHere c = {s = ["from here"] ; p = c.p} ;
DestHere c = {s = ["to here"] ; p = c.p} ;
CCoord x y = {p = "(" ++ x.s ++ "," ++ y.s ++ ")"} ;
</PRE>
<P>
When the grammar is used in the actual system, standard parsing methods
are used for interpreting the integrated speech and click input.
Parsing appears on two levels: the speech input parsing
performed by the Nuance speech recognition program (without the clicks),
and the semantics-yielding parser sending input to the dialogue manager.
The latter parser just attaches the clicks to the speech input. The order
of the clicks is preserved, and the parser can hence associate each of
the clicks with proper demonstratives. Here is the grammar used in the
two parsing phases.
</P>
<PRE>
cat
Query, -- whole content
Speech ; -- speech only
fun
QueryInput : Input -&gt; Query ; -- the whole content shown
SpeechInput : Input -&gt; Speech ; -- only the speech shown
lincat
Query, Speech = {s : Str} ;
lin
QueryInput i = {s = i.s ++ ";" ++ i.p} ;
SpeechInput i = {s = i.s} ;
</PRE>
<P></P>
<A NAME="toc6"></A>
<H3>Digression: discontinuous constituents</H3>
<P>
The GF representation of integrated multimodality is
similar to the representation of <B>discontinous constituents</B>.
For instance, assume <I>has arrived</I> is a verb phrase in English,
which can be used both in declarative sentences and questions,
</P>
<BLOCKQUOTE>
she <I>has arrived</I>
</BLOCKQUOTE>
<P></P>
<BLOCKQUOTE>
<I>has</I> she <I>arrived</I>
</BLOCKQUOTE>
<P></P>
<P>
In the question, the two words are separated from each other. If
<I>has arrived</I> is a constituent of the question, it is thus discontinuous.
To represent such constituents in GF, records can be used:
we split verb phrases (<CODE>VP</CODE>) into a finite and infinitive part.
</P>
<PRE>
lincat VP = {fin, inf : Str} ;
lin Indic np vp = {s = np.s ++ vp.fin ++ vp.inf} ;
lin Quest np vp = {s = vp.fin ++ np.s ++ vp.inf} ;
</PRE>
<P></P>
<A NAME="toc7"></A>
<H3>From grammars to dialogue systems</H3>
<P>
The general recipe for using GF when building dialogue systems
is to write a grammar with the following components:
</P>
<UL>
<LI>The abstract syntax defines the semantics (the "ontology")
of the domain of the system.
<LI>The concrete syntaxes define alternative modes of input and output.
</UL>
<P>
The engineering advantages of this approach have to do partly with
the declarativity of the description, partly with the tools provided
by GF to derive different components of the system:
</P>
<UL>
<LI>The type checker guarantees that all the input and output
modes match with the ontology.
<LI>The grammar compiler generates parsers for each input grammar
and generators for each output grammar.
<LI>Translators between GF's abstract syntax and other ontology
description languages enable communication with different
kinds of dialogue managers and cover e.g. Prolog terms and XML objects.
<LI>Translators from GF's concrete syntax to speech recognition formats
make it possible to generate e.g. Nuance grammars and ATK language
models.
</UL>
<P>
An example of this process is Björn Bringert's TramDemo.
More recently, grammars have been integrated to the GoDiS dialogue
manager by Prolog representations of abstract syntax.
</P>
<A NAME="toc8"></A>
<H2>Adding multimodality to a unimodal grammar</H2>
<P>
This section gives a recipe for making any unimodal grammar
multimodal, by adding pointing gestures to chosen expressions. The recipe
guarantees that the resulting grammar remains semantically well-formed,
i.e. type correct.
</P>
<A NAME="toc9"></A>
<H3>The multimodal conversion</H3>
<P>
The <B>multimodal conversion</B> of a grammar consists of seven
steps, of which the first is always the same, the second
involves a decision, and the rest are derivative:
</P>
<OL>
<LI>Add the category <CODE>`Point`</CODE> with a standard linearization type.
<PRE>
cat Point ;
lincat Point = {point : Str} ;
</PRE>
<LI>(Decision) Decide which constructors are demonstrative, i.e. take
a pointing gesture as an argument. Add a <CODE>Point`</CODE> as their last argument.
The new type signatures for such constructors <I>d</I> have the form
<PRE>
fun d : ... -&gt; Point -&gt; D
</PRE>
<LI>(Derivative) Add a <CODE>point</CODE> field to the linearization type <I>L</I> of any
demonstrative category <I>D</I>, i.e. a category that has at least one demonstrative
constructor:
<PRE>
lincat D = L ** {point : Str} ;
</PRE>
<LI>(Derivative) If some other category <I>C</I> has a constructor <I>d</I> that takes
demonstratives as arguments, make it demonstrative by adding a <I>point</I> field
to its linearization type.
<LI>(Derivative) Store the <CODE>point</CODE> field in the linearization <I>t</I> of any
constructor <I>d</I> that has been made demonstrative:
<PRE>
lin d x1 ... xn p = t x1 ... xn ** {point = p.point} ;
</PRE>
<LI>(Derivative) For each constructor <I>f</I> that takes demonstratives <I>D_1,...,D_n</I>
as arguments, collect the <I>point</I> fields of the arguments in the <I>point</I>
field of the value:
<PRE>
lin f x_1 ... x_m =
t x_1 ... x_m ** {point = x_d1.point ++ ... ++ x_dn.point} ;
</PRE>
Make sure that the pointings <CODE>x_d1.point ... x_dn.point</CODE> are concatenated
in the same order as the arguments appear in the <I>linearization</I> <I>t</I>,
which is not necessarily the same as the abstract argument order.
<LI>(Derivative) To preserve type correctness, add an empty
<CODE>point</CODE> field to the linearization <I>t</I> of any
constructor <I>c</I> of a demonstrative category:
<PRE>
lin c x1 ... xn = t x1 ... xn ** {point = []} ;
</PRE>
</OL>
<A NAME="toc10"></A>
<H3>An example of the conversion</H3>
<P>
Start with a Tram Demo grammar with no demonstratives, but just
tram stop names and the indexical <I>here</I> (interpreted as e.g. the user's
standing place).
</P>
<PRE>
cat
Input, Dep, Dest, Name ;
fun
GoFromTo : Dep -&gt; Dest -&gt; Input ;
DepHere : Dep ;
DestHere : Dest ;
DepName : Name -&gt; Dep ;
DestName : Name -&gt; Dest ;
Almedal : Name ;
</PRE>
<P>
A unimodal English concrete syntax of the grammar is
</P>
<PRE>
lincat
Input, Dep, Dest, Name = {s : Str} ;
lin
GoFromTo x y = {s = ["I want to go"] ++ x.s ++ y.s} ;
DepHere = {s = ["from here"]} ;
DestHere = {s = ["to here"]} ;
DepName n = {s = ["from"] ++ n.s} ;
DestName n = {s = ["to"] ++ n.s} ;
Almedal = {s = "Almedal"} ;
</PRE>
<P>
Let us follow the steps of the recipe.
</P>
<OL>
<LI>We add the category <CODE>Point</CODE> and its linearization type.
<LI>We decide that <CODE>DepHere</CODE> and <CODE>DestHere</CODE> involve a pointing gesture.
<LI>We add <CODE>point</CODE> to the linearization types of <CODE>Dep</CODE> and <CODE>Dest</CODE>.
<LI>Therefore, also add <CODE>point</CODE> to <CODE>Input</CODE>. (But <CODE>Name</CODE> remains unimodal.)
<LI>Add <CODE>p.point</CODE> to the linearizations of <CODE>DepHere</CODE> and <CODE>DestHere</CODE>.
<LI>Concatenate the points of the arguments of <CODE>GoFromTo</CODE>.
<LI>Add an empty <CODE>point</CODE> to <CODE>DepName</CODE> and <CODE>DestName</CODE>.
</OL>
<P>
In the resulting grammar, one category is added and
two functions are changed in the abstract syntax (annotated by the step numbers):
</P>
<PRE>
cat
Point ; -- 1
fun
DepHere : Point -&gt; Dep ; -- 2
DestHere : Point -&gt; Dest ; -- 2
</PRE>
<P>
The concrete syntax in its entirety looks as follows
</P>
<PRE>
lincat
Dep, Dest = {s : Str ; point : Str} ; -- 3
Input = {s : Str ; point : Str} ; -- 4
Name = {s : Str} ;
Point = {point : Str} ; -- 1
lin
GoFromTo x y = {s = ["I want to go"] ++ x.s ++ y.s ; -- 6
point = x.point ++ y.point
} ;
DepHere p = {s = ["from here"] ; -- 5
point = p.point
} ;
DestHere p = {s = ["to here"] : -- 5
point = p.point
} ;
DepName n = {s = ["from"] ++ n.s ; -- 7
point = []
} ;
DestName n = {s = ["to"] ++ n.s ; -- 7
point = []
} ;
Almedal = {s = "Almedal"} ;
</PRE>
<P>
What we need in addition, to use the grammar in applications, are
</P>
<OL>
<LI>Constructors for <CODE>Point</CODE>, e.g. coordinate pairs.
<LI>Top-level categories, like <CODE>Query</CODE> and <CODE>Speech</CODE> in the original.
</OL>
<P>
But their proper place is probably in another grammar module, so that
the core Tram Demo grammar can be used in different systems e.g.
encoding clicks in different ways.
</P>
<A NAME="toc11"></A>
<H3>Multimodal conversion combinators</H3>
<P>
GF is a functional programming language, and we exploit this
by providing a set of combinators that makes the multimodal conversion easier
and clearer. We start with the type of sequences of pointing gestures.
</P>
<PRE>
Point : Type = {point : Str} ;
</PRE>
<P>
To make a record type multimodal is to extend it with <CODE>Point</CODE>.
The record extension operator <CODE>**</CODE> is needed here.
</P>
<PRE>
Dem : Type -&gt; Type = \t -&gt; t ** Point ;
</PRE>
<P>
To construct, use, and concatenate pointings:
</P>
<PRE>
mkPoint : Str -&gt; Point = \s -&gt; {point = s} ;
noPoint : Point = mkPoint [] ;
point : Point -&gt; Str = \p -&gt; p.point ;
concatPoint : (x,y : Point) -&gt; Point = \x,y -&gt;
mkPoint (point x ++ point y) ;
</PRE>
<P>
Finally, to add pointing to a record, with the limiting case of no demonstrative needed.
</P>
<PRE>
mkDem : (t : Type) -&gt; t -&gt; Point -&gt; Dem t = \_,x,s -&gt; x ** s ;
nonDem : (t : Type) -&gt; t -&gt; Dem t = \t,x -&gt; mkDem t x noPoint ;
</PRE>
<P>
Let us rewrite the Tram Demo grammar by using these combinators:
</P>
<PRE>
oper
SS : Type = {s : Str} ;
lincat
Input, Dep, Dest = Dem SS ;
Name = SS ;
lin
GoFromTo x y = {s = ["I want to go"] ++ x.s ++ y.s} **
concatPoint x y ;
DepHere = mkDem SS {s = ["from here"]} ;
DestHere = mkDem SS {s = ["to here"]} ;
DepName n = nonDem SS {s = ["from"] ++ n.s} ;
DestName n = nonDem SS {s = ["to"] ++ n.s} ;
Almedal = {s = "Almedal"} ;
</PRE>
<P>
The type synonym <CODE>SS</CODE> is introduced to make the combinator applications
concise. Notice the use of partial application in <CODE>DepHere</CODE> and
<CODE>DestHere</CODE>; an equivalent way to write is
</P>
<PRE>
DepHere p = mkDem SS {s = ["from here"]} p ;
</PRE>
<P></P>
<A NAME="toc12"></A>
<H2>Multimodal resource grammars</H2>
<P>
The main advantage of using GF when building dialogue systems is
that various components of the system
can be automatically generated from GF grammars.
Writing these grammars, however, can still be a considerable
task. A case in point are multilingual systems:
how to localize e.g. a system built in a car to
the languages of all those customers to whom the
car is sold? This problem has been the main focus of
GF for some years, and the solution on which most work has been
done is the development of <B>resource grammar libraries</B>.
These libraries work in the same way as program libraries
in software engineering, enabling a division of labour
between linguists and domain experts.
</P>
<P>
One of the goals in the resource grammars of different
languages has been to provide a <B>language-independent API</B>,
which makes the same resource grammar functions available for
different languages. For instance, the categories
<CODE>S</CODE>, <CODE>NP</CODE>, and <CODE>VP</CODE> are available in all of the
10 languages currently supported, and so is the function
</P>
<PRE>
PredVP : NP -&gt; VP -&gt; S
</PRE>
<P>
which corresponds to the rule <CODE>S -&gt; NP VP</CODE> in phrase
structure grammar. However, there are several levels of abstraction
between the function <CODE>PredVP</CODE> and the phrase structure rule,
because the rule is implemented in so different ways in different
languages. In particular, discontinuous constituents are needed in
various degrees to make the rule work in different languages.
</P>
<P>
Now, dealing with discontinuous constituents is one of the demanding
aspects of multilingual grammar writing that the resource grammar
API is designed to hide. But the proposed treatment of integrated
multimodality is heavily dependent on similar things. What can we
do to make multimodal grammars easier to write (for different languages)?
There are two orthogonal answers:
</P>
<OL>
<LI>Use resource grammars to write a unimodal dialogue grammar and
then apply the multimodal
conversion to manually chosen parts.
<LI>Use <B>multimodal resource grammars</B> to derive multimodal
dialogue system grammars directly.
</OL>
<P>
The multimodal resource grammar library has been obtained from
the unimodal one by applying the multimodal conversion manually.
In addition, the API has been simplified
by leaving out structures needed in written technical documents
(the original application area of GF) but not in spoken dialogue.
</P>
<P>
In the following subsections, we will show a part of the
multimodal resource grammar API, limited to a fragment that
is needed to get the main ideas and to reimplement the
Tram Demo grammar. The reimplementation shows one more advantage
of the resource grammar approach: dialogue systems can be
automatically instantiated to different languages.
</P>
<A NAME="toc13"></A>
<H3>Resource grammar API</H3>
<P>
The resource grammar API has three main kinds of entries:
</P>
<OL>
<LI>Language-independent linguistic structures (``linguistic ontology''), e.g.
<PRE>
PredVP : NP -&gt; VP -&gt; S ; -- "Mary helps him"
</PRE>
<LI>Language-specific syntax extensions, e.g. Swedish and German fronting
topicalization
<PRE>
TopicObj : NP -&gt; VP -&gt; S ; -- "honom hjälper Mary"
</PRE>
<LI>Language-specific lexical constructors, e.g. Germanic <I>Ablaut</I> patterns
<PRE>
irregV : (sing,sang,sung : Str) -&gt; V ;
</PRE>
</OL>
<P>
The first two kinds of entries are <CODE>cat</CODE> and <CODE>fun</CODE> definitions
in an abstract syntax. The multimodal, restricted API has
e.g. the following categories. Their names are obtained from
the corresponding unimodal categories by prefixing <CODE>M</CODE>.
</P>
<PRE>
MS ; -- multimodal sentence or question
MQS ; -- multimodal wh question
MImp ; -- multimodal imperative
MVP ; -- multimodal verb phrase
MNP ; -- multimodal (demonstrative) noun phrase
MAdv ; -- multimodal (demonstrative) adverbial
Point ; -- pointing gesture
</PRE>
<P></P>
<A NAME="toc14"></A>
<H3>Multimodal API: functions for building demonstratives</H3>
<P>
Demonstrative pronouns can be used both as noun phrases and
as determiners.
</P>
<PRE>
this_MNP : Point -&gt; MNP ; -- this
thisDet_MNP : CN -&gt; Point -&gt; MNP ; -- this car
</PRE>
<P>
There are also demonstrative adverbs, and prepositions give
a productive way to build more adverbs.
</P>
<PRE>
here_MAdv : Point -&gt; MAdv ; -- here
here7from_MAdv : Point -&gt; MAdv ; -- from here
MPrepNP : Prep -&gt; MNP -&gt; MAdv ; -- in this car
</PRE>
<P></P>
<A NAME="toc15"></A>
<H3>Multimodal API: functions for building sentences and phrases</H3>
<P>
A handful of predication rules construct sentences, questions, and imperatives.
</P>
<PRE>
MPredVP : MNP -&gt; MVP -&gt; MS ; -- this plane flies here
MQPredVP : MNP -&gt; MVP -&gt; MQS ; -- does this plane fly here
MQuestVP : IP -&gt; MVP -&gt; MQS ; -- who flies here
MImpVP : MVP -&gt; MImp ; -- fly here!
</PRE>
<P>
Verb phrases are constructed from verbs (inherited as such from
the unimodal API) by providing their complements.
</P>
<PRE>
MUseV : V -&gt; MVP ; -- flies
MComplV2 : V2 -&gt; MNP -&gt; MVP ; -- takes this
MComplVV : VV -&gt; MVP -&gt; MVP ; -- wants to take this
</PRE>
<P>
A multimodal adverb can be attached to a verb phrase.
</P>
<PRE>
MAdvVP : MVP -&gt; MAdv -&gt; MVP ; -- flies here
</PRE>
<P></P>
<A NAME="toc16"></A>
<H3>Language-independent implementation: examples</H3>
<P>
The implementation makes heavy use of the multimodal conversion
combinators. It adds a <CODE>point</CODE> field to whatever the implementation of the unimodal
category is in any language. Thus, for example
</P>
<PRE>
lincat
MVP = Dem VP ;
MNP = Dem NP ;
MAdv = Dem Adv ;
lin
this_MNP = mkDem NP this_NP ;
-- i.e. this_MNP p = this_NP ** {point = p.point} ;
MComplV2 verb obj = mkDem VP (ComplV2 verb obj) obj ;
MAdvVP vp adv = mkDem VP (AdvVP vp adv) (concatPoint vp adv) ;
</PRE>
<P></P>
<A NAME="toc17"></A>
<H3>Multimodal API: interface to unimodal expressions</H3>
<P>
Using nondemonstrative expressions as demonstratives:
</P>
<PRE>
DemNP : NP -&gt; MNP ;
DemAdv : Adv -&gt; MAdv ;
</PRE>
<P>
Building top-level phrases:
</P>
<PRE>
PhrMS : Pol -&gt; MS -&gt; Phr ;
PhrMS : Pol -&gt; MS -&gt; Phr ;
PhrMQS : Pol -&gt; MQS -&gt; Phr ;
PhrMImp : Pol -&gt; MImp -&gt; Phr ;
</PRE>
<P></P>
<A NAME="toc18"></A>
<H3>Instantiating multimodality to different languages</H3>
<P>
The implementation above has only used the resource grammar API,
not the concrete implementations. The library <CODE>Demonstrative</CODE>
is a <B>parametrized module</B>, also called a <B>functor</B>, which
has the following structure
</P>
<PRE>
incomplete concrete DemonstrativeI of Demonstrative =
Cat, TenseX ** open Test, Structural in {
-- lincat and lin rules
}
</PRE>
<P>
It can be <B>instantiated</B> to different languages as follows.
</P>
<PRE>
concrete DemonstrativeEng of Demonstrative =
CatEng, TenseX ** DemonstrativeI with
(Test = TestEng),
(Structural = StructuralEng) ;
concrete DemonstrativeSwe of Demonstrative =
CatSwe, TenseX ** DemonstrativeI with
(Test = TestSwe),
(Structural = StructuralSwe) ;
</PRE>
<P></P>
<A NAME="toc19"></A>
<H3>Language-independent reimplementation of TramDemo</H3>
<P>
Again using the functor idea, we reimplement <CODE>TramDemo</CODE>
as follows:
</P>
<PRE>
incomplete concrete TramI of Tram = open Multimodal in {
lincat
Query = Phr ; Input = MS ;
Dep, Dest = MAdv ; Click = Point ;
lin
QInput = PhrMS PPos ;
GoFromTo x y =
MPredVP (DemNP (UsePron i_Pron))
(MAdvVP (MAdvVP (MComplVV want_VV (MUseV go_V)) x) y) ;
DepHere = here7from_MAdv ;
DestHere = here7to_MAdv ;
DepName s = MPrepNP from_Prep (DemNP (UsePN (SymbPN (MkSymb s)))) ;
DestName s = MPrepNP to_Prep (DemNP (UsePN (SymbPN (MkSymb s)))) ;
</PRE>
<P>
Then we can instantiate this to all languages for which
the <CODE>Multimodal</CODE> API has been implemented:
</P>
<PRE>
concrete TramEng of Tram = TramI with
(Multimodal = MultimodalEng) ;
concrete TramSwe of Tram = TramI with
(Multimodal = MultimodalSwe) ;
concrete TramFre of Tram = TramI with
(Multimodal = MultimodalFre) ;
</PRE>
<P></P>
<A NAME="toc20"></A>
<H3>The order problem</H3>
<P>
It was pointed out in the section on the multimodal conversion that
the concrete word order may be different from the abstract one,
and vary between different languages. For instance, Swedish
topicalization
</P>
<BLOCKQUOTE>
Det här tåget vill den här kunden inte ta.
</BLOCKQUOTE>
<P></P>
<P>
(``this train, this customer doesn't want to take'') may well have
an abstract syntax of a form in which the customer appears
before the train.
</P>
<P>
This is a problem for the implementor of the resource grammar.
It means that some parts of the resource must be written manually
and not as a functor.
However, the <I>user</I> of the resource can safely
ignore the word order problem, if it is correctly dealt with in
the resource.
</P>
<A NAME="toc21"></A>
<H3>A recipe for using the resource library</H3>
<P>
When starting to develop resource grammars, we believed they
would be all that
an application grammarian needs to write a concrete syntax.
However, experience has shown that it can be tough to start
grammar development in this way: selecting functions from
a resource API requires more abstract thinking than just
writing strings, and its take longer to reach testable
results. The most light-weight format is
maybe to start with context-free grammars (which notation is
also supported by GF). Context-free grammars that
give acceptable even though over-generating
results for languages like English are quick to produce.
</P>
<P>
The experience has led to the following
steps for grammar development. While giving the work
a quick start, this recipe
increases abstraction at a later level, when it is time to
to localize the grammar to different languages.
If context-free notation is used, steps 1 and 2 can
be merged.
</P>
<OL>
<LI>Encode domain ontology in and abstract syntax, <CODE>Domain</CODE>.
<LI>Write a rough concrete syntax in English, <CODE>DomainRough</CODE>.
This can be oversimplified and overgenerating.
<LI>Reimplement by using the resource library, and build a functor <CODE>DomainI</CODE>.
This can helped by <B>example-based grammar writing</B>, where
the examples are generated from <CODE>DomainRough</CODE>.
<LI>Instantiate the functor <CODE>DomainI</CODE> to different languages,
and test the results by generating linearizations.
<LI>If some rule doesn't satisfy in some language, use the resource in
a different way for that case (<B>compile-time transfer</B>).
</OL>
<!-- html code generated by txt2tags 2.3 (http://txt2tags.sf.net) -->
<!-- cmdline: txt2tags -\-toc multimodal.txt -->
</BODY></HTML>
View Git Blame Copy Permalink