going beyond google...

F.Chessa & G.Brelstaff 1

Going beyond Google Translate?

Francesca ChessaDLS, University of Sassari, Italy

Gavin Brelstaff (gjb@ crs4.it)CRS4 09010 Pula (CA) – Sardinia, Italy

CHItaly2011, AlgheroSeptember 2011

Ciclo 2012 di seminari interni CRS429/02/2012

F.Chessa & G.Brelstaff 2

today

http://bioinformatics.oxfordjournals.org/content/25/2/258.full

F.Chessa & G.Brelstaff 3

F.Chessa & G.Brelstaff 4

http://videolectures.net/w3cworkshop2011_brelstaff_interactive/

May 2011

F.Chessa & G.Brelstaff 5

poetry 2011

F.Chessa & G.Brelstaff 6

Healthy LivingFiction Poetry

poetry corner

F.Chessa & G.Brelstaff 7

Healthy LivingFiction Poetry

poetry corner

F.Chessa & G.Brelstaff 8

Healthy LivingFiction Poetry

poetry corner

F.Chessa & G.Brelstaff 9

Healthy LivingFiction Poetry

poetry corner

F.Chessa & G.Brelstaff 10

Healthy LivingFiction Poetry

poetry corner (UK, 2011)

F.Chessa & G.Brelstaff 11

The Waste Land iPad app earns back its costs in six weeks on the App StoreThe Guardian 8 Aug 2011

F.Chessa & G.Brelstaff 12

Alghero app - html

F.Chessa & G.Brelstaff 13

Echo Chamber –natural resonances

F.Chessa & G.Brelstaff 14

F.Chessa & G.Brelstaff 15

He was the cat that walked by himselfand all places were alike to him.

Kipling

Internet economy - www

Genius locithecreative spiritsof place– not just geolocation.

Minority languageaseed-bedfor poetic expression,- not just ICT communication.

Sardinia

Whenever we lose a language the“genetic basis” for such expression diminishes, globally

F.Chessa & G.Brelstaff 16

This talk:

� Alghero

� Poetry

� HCI

F.Chessa & G.Brelstaff 17

Screenshot

F.Chessa & G.Brelstaff 18

Resulting alignment

F.Chessa & G.Brelstaff 19

Una llavor de mela

és caiguda

a un fos

entre la terra negra.

Language Barrier

writerreader

Echo Chamber (ear,tongue,thought,eye)

F.Chessa & G.Brelstaff 20

Una llavor de

mela

és caiguda

a un fos

entre la terra

negra.

Parallel text alignment ↔ to communicate semantics• standards-based markup (TEI,XML), html delivery• going beyond GoogleTranslate →

An almond seed

into furrow

fell

amongst dark earth.

An almond seed

into furrow

fell

amongst dark earth.

Una llavor de mela

és caiguda

a un fos

entre la terra negra.

fellés caiguda

Translator

writerreader

F.Chessa & G.Brelstaff 21

Poetryis anextreme challenge

GoogleTranslate:is pitiful when readaloud as poetry

Echo Chamber(ear,tongue,thought,eye)

F.Chessa & G.Brelstaff 22

We are not doing statistical machine translation (SMT) here

A human translates each poem and marks up the equivalances at three different levels: word, phrase, idea.

Spatio-visual cues are thus obtained.

HCI and SMT

word

phrase

idea

F.Chessa & G.Brelstaff 23

Spatio-visual cues are thus available.

Q: Can such cues provide an intermediate representationthat can be usefully manipulated by SMT?

Q: Might an algorithm compute emergent semantic context from them?

HCI and SMT ?

F.Chessa & G.Brelstaff 24

és caiguda

fellOut

In

Algorithm Common spatio-visual representation

Emergentsemanticcontext ?

Translator

point & click

Semanticcontext

view

HCI and SMT ?

F.Chessa & G.Brelstaff 25

HCI: design factors

• non-verbal interface – non-linguistic thinking

• faciltate smooth switching of gaze between parallel texts

via the secondary highlighting

• text-selection by point & click at words

• see through cursor-sprite when on text

Translator

point & click

Semanticcontext

view

F.Chessa & G.Brelstaff 26

interactive alignment: steps

• atomic segmentation {word,phrase,idea}

• restoring state• select by click

• merge operation• alignment between parallel texts

• split operation• review

jQuery implemention

F.Chessa & G.Brelstaff 27

Demo (a desktop browser: IE8-9,FF3-10,Opera11,Chrome,Safari)

F.Chessa & G.Brelstaff 28

Demo:selection by click

F.Chessa & G.Brelstaff 29

Demo:selection & alignment

F.Chessa & G.Brelstaff 30

Presentation Content Structure Semantics

eXist XMLdb

XMLCSSXHTML not RDF

Unicode

http put

REST/ajax

DOMJavascript

jQuery

I n t e r a c t

TEI-p5

XMLSchema

XSL

XQL

notw3cRange

now go mobile...

F.Chessa & G.Brelstaff 31

Text Encoding Initiative – the TEI way

F.Chessa & G.Brelstaff 32

Conclusion

• Alghero html app for aligning parallel texts

• genric multilingual web application• text highlights = spatio-visual cues

• intermediate representation for semantic context?

• poetry: extreme challenge for SMT

• poetry: a new market?

F.Chessa & G.Brelstaff 33

Follow up:

Francesca ChessaDLS, University of Sassari, Italy

Gavin Brelstaff (gjb@ crs4.it )CRS4 09010 Pula (CA) – Sardinia, Italy