Online presentations

Using S5 inside Drupal.

CNLs for multilingual queries in MOLTO

Olga Caprotti, University of Gothenburg Milen Chechev, Ontotext Ramona Enache, University of Gothenburg Meritxell Gonzalez, Universitat Politècnica de Catalunya Aarne Ranta, University of Gothenburg Jordi Saludes, Universitat Politècnica de Catalunya

MOLTO - Multilingual Online Translations is an EU project targeting the production of tools for high quality translations of online materials. The demo includes the ongoing work on using the grammar formalism GF with semantic web technologies for building an information retrieval system accessed via multilingual queries. The central result is a library for query primitives available in 5 languages that was furthermore used to develop queries for the PROTON ontology and for the biomedical patents domain. Further CNLs under consideration in MOLTO come from the domains of mathematics, business processes, and cultural heritage.

Demonstrations

CNLs in MOLTO

MOLTO deals with CNLs in two ways:

The CNL of patents in MOLTO

GF-Java

Java interface to GF

Java front-end to a GF service, modeled after the GF Webservice / GF Cloud Service

Status

Backends

GF Webservice / GF Cloud Service

Documentation:

Starting (requires GF version 2012-11-14 or newer):

    $ GF_RESTRICTED=yes gf --server=41297 --document-root `pwd`/document-root

(Might require the full path to the document-root directory.)

Note that the document root must contain a writeable subdirectory tmp for the storage commands to work.

See usage examples in tools/test_gfws.sh.

JPGF

JPGF could be another possible backend, but currently calls to it have not been implemented.

Building

Testing against the locally running GF server.

    $ mvn test

Building the jar-file.

    $ mvn package -DskipTests

Building the website

    $ mvn site

Setting up an Eclipse project

The Eclipse (or any other IDE) project files are not included in this repository. To generate the required Eclipse files, run:

    $ mvn -Declipse.workspace=/home/yourname/workspace/ eclipse:configure-workspace
    $ mvn eclipse:eclipse

Download

Download the jar if building from source should fail:

Lexicon extraction flagship

PDF generated from the Landslide: http://www.molto-project.eu/sites/default/files/Lexicon_extraction_in_MO...

Raw markdown code below.

Lexicon extraction in MOLTO

Krasimir Angelov, Lauri Carlson, Ramona Enache, Inari Listenmaa, Aarne Ranta, Shafqat Virk

Overview

  1. Introduction
  2. Types of lexicons
  3. Lexicon sources
  4. Showcases
  5. Future work

Introduction

Lexicon

Use cases

Types of lexicons

Monolingual

Multilingual

Monolingual lexicons

DictEng.gf:

    a_priori_Adv = mkAdv "a priori";
    aardvark_N = mkN "aardvark" ;
    ab_initio_Adv = mkAdv "ab initio";
    aback_Adv = mkAdv "aback";
    abactinal_A = mkA "abactinal" ;
    abandon_V2 = mkV2 (mkV "abandon");

DictGer.gf:

    a_priori_Adv = mkAdv "a priori" ;
    aachener_N = reg2N  "Aachener" "Aachener" masculine ;
    aal_N = reg2N  "Aal" "Aale" masculine ;
    aalfang_N = reg2N  "Aalfang" "Aalfänge" masculine ;
    aasvogel_N = reg2N  "Aasvogel" "Aasvögel" masculine ;
    abaenderbar_A = regA "abänderbar" ;

Multilingual lexicons

DictEngGer.gf:

    abandon_V2 = dirV2 (irregV "verlassen" "verlasst" 
                               "verließ" "verließe" "verlassen" );
    abase_V2 = dirV2 (irregV "erniedrigen" "erniedrigt"
                             "erniedrigte" "erniedrigte" "erniedrigt");
    abasement_N = mkN "Erniedrigung";

Uni-sense lexicons

Multi-sense lexicons

LinkedDictGer:
    brother_08111676_N = reg2N "Bruder" "Brüder" masculine ;
    brother_08112052_N = reg2N "Bruder" "Brüder" masculine ;
    brother_08112265_N = reg2N "Kamerad" "Kameraden" masculine ;

Multi-sense lexicons

Option 1: All synonyms in abstract syntax

Multi-sense lexicons

Option 2: One synonym per word sense

Source formats

Annotated data

Source formats

Unannotated data

Combining different sources

Showcases

  1. TermFactory: term extraction and conversion to GF
  2. WordNet: High-coverage lexicons for robust parsing
  3. Phrase tables: domain-specific lexicons for hybrid systems

TermFactory

    term1:fi-koira-N_-_ont-Dog
          syn:frame "N" ;
          gf:lin "mkN str" ;
          term:hasDesignation exp1:fi-koira-N ;
          term:hasReferent ont0:Dog .

    term1:fi-aviomies-N_-_ont-Husband
          syn:frame "jonkun N" ;
          gf:lin "mkN2 (mkN str) (casePrep genitive)" ;        
          term:hasDesignation exp1:fi-aviomies-N ;
          term:hasReferent ont0:Husband .

    term1:fi-kieltää-V_-_sem-Forbid
          syn:frame "V jotakuta olemasta" ;
          gf:lin "mkV2Vf (mkV str) (casePrep partitive) infElat" ;
          term:hasDesignation exp1:fi-kieltää-V ;
          term:hasReferent sem0:Forbid .

TermFactory

syn:frame and gf:lin. The former is a user-friendly way to annotate valency, the latter is a GF constructor. They have a mapping -->

    [] gf:mapping
        [ syn:frame "N" ; gf:lin "mkN str" ] ,
        [ syn:frame "jonkun N" ; 
          gf:lin "mkN2 (mkN str) (casePrep genitive)" ],
        [ syn:frame "V jotakuta olemasta" ; 
          gf:lin "mkV2Vf (mkV str) (casePrep partitive) infElat" ] .

    ont_Dog = mkN "koira" ;
    ont_Husband = mkN2 (mkN "aviomies") (casePrep genitive) ;
    sem_Forbid = mkV2Vf (mkV "kieltää") (casePrep partitive) infElat ;

High-coverage lexicons for robust parsing


Work by Shafqat Mumtaz Virk and K. V. S. Prasad

Parsing and linearization with WSD

Parsing and linearization with WSD

Explanation of the graph:

Robust parsing

Work by Krasimir Angelov, Aarne Ranta

How much do grammars leak? and Probabilistic Robust Parsing with Parallel Multiple Context-Free Grammars, submitted in COLING 2012 (not accepted)

-->

Phrase tables

Work by Ramona Enache

Ontology-based lexicon management in a multilingual translation system – a survey of use cases - PhD thesis, Chalmers/UGOT, forthcoming: Shafqat Virk, ___ - Conference papers

-->

Future work

Thank you!

AttachmentSize
Lexicon_extraction_in_MOLTO.pdf158.68 KB

Mathematics flagship (preliminar)

Mathematics flagship

Mathematical Grammar Library

Contributors: Jordi Saludes, Ares Ribó, Sebastian Xambó, Olga Caprotti, Aarne Ranta, Krasimir Angelov, Ramona Enach, Adam Slaski, Thomas Hallgren, Shafqat Mumtaz Virk

.notes: Contributors for Fre, Rus?

Cloud services for mathematics

Multilingual semantic wiki

Features

Example

Wiki translations

Multi-modal input/output of mathematics

.notes: This is also shown in the Query technologies flagship

Interfacing Sage

Sage interface workflow

Sage workflow

Word problem solver dialog system

Workflow

workflow

Word problem schemata

The current prototype allows to state word problems of the following form:

Setting the problem model

We considered two levels of discourse:

The fact that John has seven fruit is represented by the GF tree:

fromProp (E1owns john (gen Fruit n7))

and in Prolog by:

own(john, 7 * fruit)

while in core, it converts to the Prolog assertion:

p(X, fruit, own(john, X)) - 7 * unit(fruit)).

The latter is more suited to reasoning with it.

Normalization

Conjunctions are disaggregated: namely

is converted into:

Questions make the unknown explicit in the core expressions:

is represented in plain as:

find(own(mary,apple))

and in core as:

find(X, apple, own(mary,X))

Example

% abs:fromProp (E1owns john (gen Fruit n7))
% Eng:John has seven fruit .
p(X, fruit, own(john, X)) - 7 * unit(fruit).

% abs:fromProp (E1owns john (aplus (ConsAmount (gen Apple n2) 
%              (BaseAmount (some Orange) (gen Banana n3)))))
% Eng:John has two apples , some oranges and three bananas .
p(Y, apple, own(john, Y)) - 2 * unit(apple).
p(Z, banana, own(john, Z)) - 3 * unit(banana).
p(V, orange, own(john, V)) - some(orange).

% abs:fromQuestion (Q1owns john Orange)
% Eng:how many oranges does John have ?
find(W, orange, own(john, W)).

Solving a problem

To process a given word problem, the student must construct a set of of statements in core language that model that given word problem.

The authoring interpreter saves a word problem in a Prolog file consisting of:

Modeling a problem

When the dialog interpreter is started on a word problem file, the system uses the GF abstract lines to display the statement of the problem in the selected language.

Next, the student must go through a sequence of steps to have the problem correctly modeled by:

  1. assigning variables to given data and to unknowns

  2. discovering relations among the given data

  3. stating equations involving these unknowns

Steps 1 and 2

Steps 3 and 4

AttachmentSize
sage-wf.pdf28.17 KB
wproblems-wf.pdf54.2 KB
wiki-translations.png34.37 KB

Multilingual Semantic Wiki - Review 2012

Multilingual Semantic Wiki

Existing background technologies

Main tasks of WP11

AceWiki

Bidirectional ACE-to-OWL mapping

Every country that does not border a sea is a landlocked-country.

SubClassOf(
   ObjectIntersectionOf(
      :country
      ObjectComplementOf(
         ObjectSomeValuesFrom(
            :border
            :sea
         )
      )
   )
   :landlocked-country
)

Which country is a landlocked-country?

ObjectIntersectionOf(
    :country
    :landlocked-country
)

Article (screenshot)

Look-ahead editor (screenshot)

Reasoning (screenshot)

AceWiki in MOLTO

Generalization of AceWiki

Multiple languages

(Stable AceWiki: single language: OWL-compatible subset of ACE)

Multiple grammars

(Stable AceWiki: single grammar: OWL-compatible subset of ACE)

Multiple reasoners

Syntactic vs semantic reasoning

(Stable AceWiki: single reasoner: OWL-reasoner with multiple implementations)

Current integration with GF

is based on the GF Webservice and the GF online editor.

AceWiki and the online GF editor (screenshot)

Immediate future work

ACE in GF

Dissemination

Olga's first test presentation

Table of Contents

This should be viewable nicely using S5 engine based purely on web technologies. It allows us to create presentations for our meetings directly as wiki pages, so they are archived and we need not duplicate efforts. The look will be consistent for all the users. Since we can also use other types of nodes (supposedly), we do the work only once by editing tasks, deliverables etc and then present it in a slideshow fashion. At least this should work in theory: in practice I just spent 2 hours debugging the Drupal S5 module.

BTW, Eric Meyer's S5 package is available on its own locally and works fine.

Presentations carve text up into sections based on HTML headings. The level of the heading is discarded, and the heading text itself is held apart as a title for the text.

The TOC is made by child pages.

First slide

Here again we try to add a new page to the presentation

Second slide: embed a Prezi

Say I have a Prezi presentation, can I embed it?

The MOLTO Phrasebook on Prezi

Yes, apparently you need to select "FULL HTML" as the Input format for the Body of the node.

Opening Kickoff MOLTO EEU

MOLTO Enlarged EU Kickoff Meeting on Prezi

Sample presentation in md format

Landslide

Landslide

Generates a slideshow using the slides that power the html5-slides presentation.

A python with the jinja2, markdown, and pygments modules is required.

Here we show just how sample slides from the landslide distribution, written in the markdown format, can also be pushed to S5 to generate a slide presentation. MOLTO partners can check their slides by adding a new wiki page underneath here.

Sample slide src attached.

Markdown Formatting Instructions

Rendering Instructions

Slide #2

.fx: foo bar

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Aenean magna tellus, fermentum nec venenatis nec, dapibus id metus. Phasellus nulla massa, consequat nec tempor et, elementum viverra est. Duis sed nisl in eros adipiscing tempor.

Section #1

Integer in dignissim ipsum. Integer pretium nulla at elit facilisis eu feugiat velit consectetur.

Section #2

Donec risus tortor, dictum sollicitudin ornare eu, egestas at purus. Cras consequat lacus vitae lectus faucibus et molestie nisl gravida. Donec tempor, tortor in varius vestibulum, mi odio laoreet magna, in hendrerit nibh neque eu eros.

Unicode Characters Work

© é

Middle Title Slide

Slide #3

Hello Gentlemen

Are you prepared to see beyond the veil of reason? - DeceasedCrab

Slide #4

First code block:

!python
while True:
    print "Everything's gonna be allright"

Second code block:

!php
<?php exec('python render.py --help'); ?>

Third code block:

!xml
<?xml version="1.0" encoding="UTF-8"?>

  <img src="madonna.jpg" alt='Foligno Madonna, by Raphael' />
  This is Raphael's "Foligno" Madonna, painted in
    1511–1512.

Slide #5

Another code block:

!java
if (markdown &amp;&amp; isEasy()) {
    return true; 
}

Slide #6

An image:

logo

Slide #7

Landslide can generate QR codes:

.qr: 450|http://github.com/adamzap/landslide

This is a slide with no heading. It works too.

Slide #9

This slide has presenter notes.

Press p to open a new window for the presenter with its notes.

Presenter Notes

Hello from presenter notes

AttachmentSize
slides.md_.txt2.76 KB