Computational Morphology and the META-NET Strategic Research Agenda for Multilingual Europe 2020
-
Upload
georg-rehm -
Category
Technology
-
view
153 -
download
1
description
Transcript of Computational Morphology and the META-NET Strategic Research Agenda for Multilingual Europe 2020
Co-funded by the 7th Framework Programme and the ICT Policy Support Programme of the European Commission through the contracts T4ME, CESAR, METANET4U, META-NORD (grant agreements no. 249119, 271022, 270893, 270899).
The State of Computational Morphology for Europe’s Languages and the META-NET Strategic Research Agenda
Georg Rehm
Network Manager META-NET DFKI, Berlin, Germany
3rd Int. Workshop on Systems and Frameworks for Computational Morphology (SFCM 2013)
Berlin, Germany – September 06, 2013
Outline
q Introduction
q Language White Paper Series: Europe’s Languages in the Digital Age
q The State of Computational Morphology for Europe’s Languages
q The META-NET Strategic Research Agenda for Multilingual Europe
q Conclusions
http://www.meta-net.eu 2
Multilingual Europe
3 http://www.meta-net.eu
q Where were we back in 2010?
q Challenge: Providing each language community with the most advanced technologies for communication and information so that maintaining their mother tongue does not turn into a disadvantage.
q While research has made considerable progress in recent years, the pace of progress is not fast enough to meet the challenge within the next 10-20 years.
q All stakeholders – researchers, LT user and provider industries, language communities, funding programmes, policy makers – should team up in a strategic alliance for a major dedicated push.
Objectives
META-NET is a network of excellence dedicated to fostering the tech-nological foundations of the European multilingual information society.
http://www.meta-net.eu 4
Four EU-Funded Projects
q Initial project: T4ME (FP7; 13 partners, 10 countries)
q Three ICT-PSP consortia since Feb. 2011: CESAR, METANET4U, META-NORD
q All four projects ended on January 31, 2013.
q All EU member states and several non-member states covered.
q META-NET in Sept. 2013: 60 members in 34 countries.
http://www.meta-net.eu 5
http://www.meta-net.eu/members
Language White Paper Series Europe’s Languages in the Digital Age
http://www.meta-net.eu 6
Language White Paper Series
http://www.meta-net.eu 7
q “Europe’s Languages in the Digital Age”. q Reports on the state of our languages in
the digital age and the level of support through language technology.
q Series covers 30 languages. q Key communication instruments to
address decision makers and journalists. q Inform about societal and technological
problems and challenges as well as economic opportunities.
q >2 years in the making. q >200 national experts as contributors. q >8.000 copies printed and distributed to
politicians and journalists.
Language White Paper Series
http://www.meta-net.eu 8
q Structure: § Part 1: Executive Summary § Part 2: Languages at Risk — A Challenge for Language Technology § Part 3: The [X] Language in the European Information Society § Part 4: LT support for [X] § Part 5: About META-NET; References, etc.
q Language White Paper Series (published at Springer): § Ca. 8.000 printed copies distributed by META-NET. § Printed copies can be purchased through the usual channels. § Ebooks available via SpringerLink (fee) and META-NET website (free). § http://www.meta-net.eu/whitepapers
30 Languages Covered
q Basque q Bulgarian* q Catalan q Czech* q Danish* q Dutch* q English* q Estonian* q Finnish* q French*
q Galician q German* q Greek* q Hungarian* q Icelandic q Irish* q Italian* q Latvian* q Lithuanian* q Maltese*
q Norwegian q Polish* q Portuguese* q Romanian* q Serbian q Slovak* q Slovene* q Spanish* q Swedish* q Croatian
http://www.meta-net.eu 9
* = Official EU language Next up: Welsh
Cross-Lingual Comparison
q In four application areas, each language is assigned to one of five clusters, ranging from excellent LT support to weak/no support:
1. Machine Translation 2. Speech Processing
3. Text Analytics
4. Language Resources q Results finalised at a
meeting in Berlin with representatives of all 30 languages (October 21/22, 2011).
http://www.meta-net.eu 10
MT
http://www.meta-net.eu 11
English
good
French, Spanish
moderate fragmentary
Catalan, Dutch, German, Hungarian, Italian, Polish, Romanian
weak or no support
Basque, Bulgarian, Croatian, Czech, Danish, Estonian, Finnish, Galician,
Greek, Icelandic, Irish, Latvian, Lithu-anian, Maltese, Norwegian, Portuguese,
Serbian, Slovak, Slovene, Swedish
excellent
Czech, Dutch, Finnish, French, German,
Italian, Portuguese, Spanish
moderate fragmentary
Basque, Bulgarian, Catalan, Danish, Estonian, Galician, Greek,
Hungarian, Irish, Norwegian, Polish, Serbian, Slovak, Slovene, Swedish
weak or no support
Croatian, Icelandic, Latvian, Lithuanian, Maltese, Romanian
excellent
English
good
Spee
ch
English
good
Dutch, French, German, Italian,
Spanish
moderate fragmentary
Basque, Bulgarian, Catalan, Czech, Danish, Finnish, Galician, Greek, Hungarian, Norwegian, Polish, Portuguese, Romanian, Slovak,
Slovene, Swedish
weak or no support
Croatian, Estonian, Icelandic, Irish, Latvian, Lithuanian, Maltese, Serbian
excellent
English
good
Czech, Dutch, French, German, Hungarian,
Italian, Polish, Spanish, Swedish
moderate fragmentary
Basque, Bulgarian, Catalan, Croatian, Danish, Estonian, Finnish, Galician,
Greek, Norwegian, Portuguese, Romanian, Serbian, Slovak, Slovene
Icelandic, Irish, Latvian, Lithuanian, Maltese
weak/no support excellent
Res
ourc
es
Text
Ana
lysi
s
Europe’s Languages and LT
http://www.meta-net.eu 12
Dutch French German Italian
Spanish
Catalan Czech
Finnish Hungarian
Polish Portuguese
Swedish
Basque Bulgarian
Danish Galician
Greek Norwegian Romanian
Slovak Slovene
Croatian Estonian Icelandic
Irish Latvian
Lithuanian Maltese Serbian
English
good support through Language Technology
weak or no support
Not enough R&I on European languages
LT research on European languages, except for English, is too weak and too slow
Many languages are badly covered
0
50
100
150
200
250
300
350
400
450
English
Ch
inese
Germ
an, Stand
ard
Fren
ch
Spanish
Japane
se
Arabic
Dutch
Portugue
se
Czech
Danish
Swed
ish
Hind
i Ko
rean
Turkish
Ita
lian
Russian
Finn
ish
Hebrew
Hu
ngarian
Sloven
e Urdu
Romanian
Zulu
Bulgarian
Catalan-‐Va
lencian-‐Ba
lear
Greek
Thai
Welsh
Estonian
Basque
Ge
rman, Swiss
InukStut
Indo
nesia
n Ineseñ
o LaSn
Marathi
Malay
Pushto
Serbian
Syria
c Tamil
UgariS
c Ukrainian
Uspanteko
Vietnamese
Languages treated in the 2010 editions of Journal of Computational Linguistics and Conferences of ACL, EMNLP and COLING. Many European languages with no reference at all: Slovak, Maltese, Lithuanian, Irish, Albanian, Croatian, Galician etc.
Key Observations
http://www.meta-net.eu 14
q When it comes to Language Technology support, there are massive differences between Europe’s languages and technology areas.
q LT support for English is ahead of any other language.
q Even support for English is far from being perfect.
q The gap between English and the other languages keeps widening!
q Several languages – Icelandic, Latvian, Lithuanian, Maltese – receive this weakest score in all four areas!
q At least 21 European languages in danger of digital extinction!(Languages put into the “weak or no support” category at least once.)
White Paper Box Sets (100 copies)
http://www.meta-net.eu 15
White Paper Website
http://www.meta-net.eu 16
Press Campaign
q Headline of press release: At Least 21 European Languages in Danger of Digital Extinction.
q Sent out to journalists, politicians and other stakeholder groups on the European Day of Languages (Sept. 26, 2012).
q Overwhelmed by the huge interest in the topic and our key findings!
q 600+ mentions in the press. q 50+ broadcast interviews with META-NET representatives (ca. 30 radio
interviews, ca. 25 television reports).
q News came in from 40+ countries in 35+ different languages. q Whole of Europe covered.
q Two Parliamentary Questions in the European Parliament on the “digital extinction of languages” topic.
http://www.meta-net.eu 17
Coverage by Country
http://www.meta-net.eu 18
Spain, 15.90%
Bulgaria, 10.80%
International, 7.90%
Latvia, 5.30%
Netherlands, 4.80%
Greece, 4.60% Romania, 4.40%
Serbia, 4.40%
Italy, 4.20%
Germany, 3.50%
Russia, 3.50%
Estonia, 2.90%
France, 2.60%
Slovenia, 2.40%
Iceland, 2.20% Malta, 2% USA, 1.50%
Denmark, 1.30%
Latin America, 1.30%
Lithuania, 1.30%
Ireland, 1.30% UK, 1.10%
Belgium, 0.90%
Finland, 0.70% Sweden, 0.70%
Poland, 0.70%
Norway, 0.40%
Mexico, 0.40%
Brazil, 0.40%
Slovakia, 0.40%
Basque Country, 0.40% Portugal, 0.40% Austria, 0.20%
New Zealand, 0.20%
Hungary, 0.20%
Bosnia and Herzegovina, 0.20%
Costa Rica, 0.20%
Cyprus, 0.20%
Canada, 0.20%
Australia, 0.20%
Spain Bulgaria International Latvia Netherlands Greece Romania Serbia Italy Germany Russia Estonia France Slovenia Iceland Malta USA Denmark Latin America Lithuania Ireland UK Belgium Finland Sweden Poland Norway Mexico Brazil Slovakia Basque Country Portugal Austria New Zealand Hungary Bosnia and Herzegovina Costa Rica Cyprus Canada Australia
Response: Examples
q Austria: Der Standard. q Denmark: Politiken, Berlingske Tidende. q Finland: Tiede. q Germany: Heise Newsticker, Süddeutsche Zeitung. q Greece: in.gr, Πρώτο Θέµα, Prosilipsis. q Hungary: Origo. q Iceland: Fréttablaðið, Morgunblaðið. q Italy: Wired. q Norway: Computerworld. q Slovenia: Delo, Dnevnik, Demokracija. q Serbia: Politika. q Spain: El Mundo. q UK: Huffington Post. q USA: Mashable, NBC News, Reddit.
http://www.meta-net.eu 19
Press Campaign: Highlights
http://www.meta-net.eu 20
Af Flemming Steen Pedersen// [email protected]
Langt flere kræftpatienter i hovedstadsområ-det skal behandles hurtigt og uden forsinkel-ser.
Det skal være slut med, at undersøgelse og behandling trækker i langdrag og overskrider de tidsfrister, som fagfolk har fastsat for at give patienterne de optimale chancer for at over-leve den frygtede sygdom.
Det er målet, når politikere i Region Hoved-staden nu lægger op til at udmønte en pulje på 32 mio. kr. til at øge personalet og udvide behandlingskapaciteten på kræftområdet på en række af regionens hospitaler.
Pengene kommer, efter at regionen er blevet kritiseret for, at alt for mange kræft-patienter er for lang tid om at komme igen-nem systemet. F.eks. er det ifølge den seneste opgørelse kun godt halvdelen af kvinder med brystkræft, som bliver behandlet inden for det fastsatte mål på 18 dage i de såkaldte kræft-pakker.
»Pengene betyder, at der kommer bedre forhold for kræftpatienter. Det er vigtigt, at folk får mulighed for at blive behandlet hur-tigt, så de ikke skal gå rundt og være bekym-rede,« siger formand for kvalitetsudvalget i Region Hovedstaden, Kirsten Lee (R).
Flere får kræft – og flere overleverKonkret er hensigten at udvide den onkologi-ske kapacitet – det vil sige stråle- og kemobe-handlingen – på såvel Rigshospitalet, Herlev Hospital, Hillerød Hospital og Bornholms Hospital.
Desuden sættes der penge af til at øge antal-let af operationer og udvide ambulatorieka-paciteten på det urologiske område på Herlev,
Bispebjerg og Frederiksberg. Foruden pro-blemer med lange ventetider for brystkræft-patienter er der således også patienter med prostatakræft, som venter for længe. På dags-ordenen er også at sikre hurtigere behandling til en tredje gruppe af patienter med hoved-halskræft, hvor et stort antal patienter ligele-des må vente længere end tidsgrænsen på 16 dage.
Udover at tilføre flere penge overvejes det også at indføre såkaldte servicemål for, hvor stor en andel af patienterne der skal i behandling inden for de fastsatte tidsgrænser i kræftpakkerne. Lignende servicemål findes i forvejen i Region Midtjylland og Region Syddanmark og betragtes som et middel til at presse hospitalerne og signalere, at bestemte områder har særlig høj politisk bevågenhed.
I de to regioner er målet, at henholdvis 90 og 95 pct. af patienterne skal igennem syste-met inden for forløbstiderne, og Kirsten Lee forventer, at et eventuelt servicemål i Region Hovedstaden kommer til at ligge på et tilsva-rende niveau.
I Kræftens Bekæmpelse hilser direktør Leif Vestergaard Pedersen det velkomment, at Region Hovedstaden nu bruger 32 mio. kr. til at udvide kapaciteten .
»Det har vist sig, at der er et forbedringspo-tentiale på dette område, og derfor er det godt, at man prioriterer det. Flere og flere får kræft, og flere og flere overlever. Det betyder, at kapa-citeten gradvist skal øges hele tiden. Service-mål er et godt initiativ, og et mål på 90-95 pct. er nok det realistiske, selv om udgangspunk-tet bør være 100 procent,« siger Leif Vesterga-ard Pedersen og tilføjer:
»Men så er det også vigtigt at holde fast i det mål og ikke stille sig tilfreds med, at 80 eller 85 pct. kommer igennem til tiden.« B
Kræft syge skal have hurtigerebehandling
Oprustning. Region Hovedstaden bruger 32 mio. kr. på at øge behandlingskapaciteten.
Af Jens Ejsing// [email protected]
Det danske sprog har det svært i den digitale verden.
Det konstaterer danske sprogforskere- og eksperter i forbindelse med den nye inter-nationale undersøgelse META-NET, der ser nærmere på, hvordan en lang række mindre, europæiske sprog som dansk klarer sig i den digitale verden.
Forskerne fra bl.a. Københavns Universitet og Dansk Sprognævn når frem til, at dansk i fremtiden kan få det endnu sværere i den digitale verden, fordi Google Translate, GPSer, applikationer til smartphones og andre sprog-teknologiske programmer ikke i tilstrækkelig grad formår at behandle de mange nuancer i det danske sprog.
Professor i sprogteknologi på Københavns Universitet, Bolette Sandford Pedersen, mener, at der er brug for en slags digital dansk sprogbank fyldt med data, så bl.a. oversættel-ser bliver så præcise og gode som muligt. Med
hjælp fra sprogbanken kan forskere ifølge professoren hjælpe virksomheder med at for-bedre programmer, der skal håndtere sproglig viden om bl.a. maskinoversættelse, tale-genkendelse og informationssøgning.
Dermed vil der blive længere mellem fejlag-tige oversættelser, som når »hæld olie på pan-den« med Google Translate bliver til »pour oil on the forehead« på engelsk. Oversættelser, der er i værste fald er så upræcise, at danskere ender med at fravælge deres eget sprog i den digitale verden.
Sproghjælp til virksomhederHun anerkender dog, at »teknologien til auto-matiske oversættelser på mange måder er fantastisk«.
»Den er bare ikke god nok, når det gælder dansk,« siger hun:
»Det er som om, at vi i et vist omfang lægger det i hænderne på Google eller andre virk-somheder at afgøre, om dansk skal behandles godt nok eller ej. Men det danske marked er ikke stort for dem. Spørgsmålet er derfor,
Dårlig sprogteknologi truer dansk på nettetOrd. Forskere arbejder på at forbedre danske oversættelser på internettet.
om vi ikke i højere grad selv skal gøre noget for at sikre, at det fornødne datamateriale er til rådighed, så vi får gode oversættelser og anden god sprogteknologi. Det kunne f.eks. være ved, at vi gjorde en indsats for at få opret-tet en sprogbank med en masse beriget mate-riale om dansk.«
»Hvis vi hele tiden oplever, at oversættel-ser er behæftede med fejl, tør vi ikke stole på dem,« siger hun og understreger, at »fejlagtige oversættelser kan føre til store misforståelser«.
Ifølge Dansk Sprognævns direktør, Sabine Kirchmeier-Andersen, kan dårlig sprogtekno-logi have konsekvenser for mange danskere, der ikke er så gode til engelsk.
»Hvis vi har ambitioner om at bruge det danske sprog i fremtidens teknologiske univers, skal der gøres en indsats nu for at fastholde ekspertise og udbygge den viden, vi har,« mener hun:
»Ellers risikerer vi, at kun folk, der taler fly-dende engelsk, vil få glæde af de nye generatio-ner af web-, tele- og robotteknologi, der er på vej.« B
INFOGRAFIK: HENRIK KIÆR / TEKST: FLEMMING STEEN PEDERSEN KILDE: REGION HOVEDSTADEN
De såkaldte kræftpakker, der blev indført i 2008 og 2009 for at sikre de danske kræftpatienter langt hurtigere undersøgelser og behandling, beskriver et standardudrednings- og -behand-lingsforløb. Det vil sige, hvilke undersøgelser og behandlinger der skal udføres, og hvor lang tid der højst må gå med de enkelte aktiviteter. Opgørelser fra Region Hovedstaden viser imidlertid, at en stor del af patienterne ikke behandles inden for de fastsatte tidsgrænser, og at der især er problemer inden for tre kræftsygdomme: brystkræft, hoved- og halskræft og prostatakræft.
Kræftbehandling trækker ud
PROSTATAKRÆFTServicemål: 35-39 dage
24
76
HOVED- OG HALSKRÆFTServicemål: 16 dage
40
60
BRYSTKRÆFTServicemål: 18 dage
4753
Procentdel inden for servicemål
Procentdel uden for servicemål
Sådan læses grafikken:
Positiv udviklingNegativ udvikling
H Der er omkring 80 sprog i EU. For 21 af dem – også dansk – gælder det, at der er store sprogteknologiske mangler, når det gælder bl.a. maskinoversættelse, talegenken-delse og informationssøgning.
H Ifølge en EU-undersøgelse køber et stigende antal europæiske internetbrugere varer eller tjenester på nettet, hvor det sprog, der bliver anvendt, ikke er deres eget. Det gælder over halvdelen af brugerne.
H Over hver tredje anvender et fremmed-sprog til at skrive mail eller indlæg på nettet.
fakta HSprog i Europa
REDIGERET AF JOANNA VALLENTIN. LAYOUT: JACOB FRIIS/ NATIONALT /06. BERLINGSKE / 1.SEKTION / LØRDAG 22.09.2012
Press Campaign: Highlights
http://www.meta-net.eu 21
38
Στην ψηφιακή εποχή δεν… µιλούν ελληνικά, όπως και αρκετές άλλες ευρωπαϊκές
γλώσσες, σύµφωνα µε πανευρωπαϊ-κή έκθεση µε την υπογραφή 200 και πλέον ειδικών. Η συγκεκριµένη µελέ-τη δηµοσιεύτηκε από το επιστηµονικό δίκτυο ΜΕΤΑ-ΝΕΤ µε αφορµή τη χτε-σινή Ευρωπαϊκή Ηµέρα Γλωσσών.
Για τις ανάγκες της έρευνάς τους, γλωσσολόγοι από 34 χώρες της Γη-ραιάς Ηπείρου βαθµολόγησαν τις διαθέσιµες γλωσσικές υπηρεσίες και δηµιούργησαν ένα «Λευκό Βι-βλίο» για κάθε ευρωπαϊκή γλώσσα. Στη µελέτη τους, οι ειδικοί αναζήτη-σαν µεταξύ άλλων τέσσερα βασικά ηλεκτρονικά εργαλεία, δηλαδή την ύπαρξη αυτόµατης µετάφρασης, τη δυνατότητα φωνητικής αλληλε-πίδρασης και ψηφιακής ανάλυσης κειµένου, ενώ ταυτόχρονα διερευνή-θηκε και η διαθεσιµότητα γλωσσικών πόρων ή πηγών.
Σε πρώτη φάση εξέτασαν τις ιστο-σελίδες που επιτρέπουν στους χρή-στες να κάνουν µεταφράσεις online, όπως, για παράδειγµα, η υπηρεσία του κολοσσού πληροφορικής Google Translate. Την ίδια ώρα, εξετάστηκε και η «επικοινωνία» των ελληνόφω-νων χρηστών µε τις…συσκευές τους, όπως για παράδειγµα η δυνατότητα
να «µιλήσει» κάποιος στο GPS στη µητρική του γλώσσα. Οι ερευνητές κατέληξαν στο συµπέρασµα ότι υπάρχουν τέτοιες συσκευές, αλλά δεν είναι τόσο διαδεδοµένες όσο οι αγγλόφωνες. Το «χρυσό» µετάλλιο κατακτά,
όπως είναι άλλωστε και λογικό, η αγγλική γλώσσα. Οι αγγλόφωνοι χρή-στες έχουν την καλύτερη δυνατή τε-χνολογική υποστήριξη, κάτι το οποίο ευνοεί την περαιτέρω εξάπλωση της γλώσσας. Από «τεχνολογικό απο-κλεισµό» κινδυνεύουν περισσότερο η ισλανδική, η λετονική, η λιθουανική και η µαλτέζικη γλώσσα, ενώ σε λίγο καλύτερη µοίρα βρίσκονται η ελλη-νική, η βουλγαρική, η ουγγρική και η πολωνική, που όπως αναφέρει η έρευνα έχουν «αποσπασµατική» τε-χνολογική υποστήριξη.
«Μέτρια» χαρακτηρίζεται η υπο-στήριξη χρηστών σε ολλανδική, γαλ-λική, γερµανική, ιταλική και ισπανική γλώσσα. Οι επικεφαλής της επιστη-µονικής οµάδας, Χανς Ουζκοράιτ και Γκεόργκ Ρεµ, αναφέρουν χαρακτηρι-στικά: «Υπάρχουν δραµατικές διαφο-ρές στην υποστήριξη της γλωσσικής
τεχνολογίας ανάµεσα στις διάφορες ευρωπαϊκές γλώσσες. Το χάσµα µετα-ξύ “µικρών” και “µεγάλων” γλωσσών ολοένα και διευρύνεται. Πρέπει να εξασφαλίσουµε τον εφοδιασµό των µικρότερων και λιγότερο πλούσιων σε ψηφιακούς πόρους γλωσσών µε τις απαραίτητες βασικές τεχνολογί-ες. ∆ιαφορετικά, οι γλώσσες αυτές είναι καταδικασµένες σε ψηφιακή εξαφάνιση».
Μάλιστα, οι ειδικοί τονίζουν ότι χω-ρίς αποφασιστική δράση οι γλώσσες αυτές δύσκολα θα… επιβιώσουν στον ψηφιακό κόσµου του 21ου αιώνα. Η κ. Μαρία Γαβριηλίδου, µέλος της επι-στηµονικής οµάδας από το Ινστιτούτο
Επεξεργασίας του Λόγου Ερευνητικό Κέντρο Αθηνά, λέει στον «Ε.Τ.»: «Η έρευνα αυτή δεν λέει ότι δεν θα ζήσει η ελληνική γλώσσα ή ότι κινδυνεύει µε εξαφάνιση». Η ειδικός εξηγεί ότι όσο υπάρχουν άνθρωποι που µιλά-νε, γράφουν και επικοινωνούν µε µια γλώσσα, τότε αυτή θα συνεχίσει να υπάρχει. Είναι σηµαντικό, όµως, να έχουν όλοι οι χρήστες τη δυνατότητα να «µιλήσουν» στις µηχανές, όπως τα GPS τους, στα ελληνικά και να έχουν στη διάθεσή τους γλωσσικά εργαλεία ηλεκτρονικών υπολογιστών.
Μεταξύ αυτών των «εργαλείων» είναι οι διορθωτές ορθογραφικών και συντακτικών λαθών, που χρησιµοποι-ούνται καθηµερινά από εκατοντάδες Ελληνες χρήστες και βασίζονται στη γλωσσική τεχνολογία. Παρ’ όλα αυτά, τονίζει ότι η ψη-
φιακή εξάπλωση µιας γλώσσας είναι σηµαντική «∆εν είναι στα χέρια του µέσου χρήστη. Οι εκάστοτε κυβερ-νήσεις, η Ευρωπαϊκή Ενωση και ο ιδιωτικός τοµέας πρέπει να χρηµα-τοδοτήσουν την ανάπτυξη αυτής της τεχνολογίας για όλες τις γλώσσες», αναφέρει και συνεχίζει: «Οι χρήστες, όµως, πρέπει να απαιτούν να υπάρ-χουν και στη γλώσσα τους τα µέσα αυτά και να µην ικανοποιούνται µε τα αγγλικά».
Πέµπτη 27 Σεπτεµβρίου 2012 ΕΛΕΥΘΕΡΟΣ ΤΥΠΟΣ
LifeΠΟΛΛΕΣ ΕΥΡΩΠΑΪΚΕΣ ΓΛΩΣΣΕΣ ΘΕΩΡΟΥΝΤΑΙ ΤΕΧΝΟΛΟΓΙΚΑ… ΞΕΠΕΡΑΣΜΕΝΕΣ
Με ψηφιακή εξαφάνιση κινδυνεύουν τα ελληνικά
ΕΛΕΝΗ ΒΕΡΓΟΥ[email protected]
Η γλώσσα της αποξένωσης…
XX GREEKLISH
Οι αγγλόφωνοι χρήστες έχουν την καλύτερη δυνατή τεχνολογική υποστήριξη, γεγονός που ευνοεί την περαιτέρω εξάπλωση της γλώσσας
ΜΕ GREEKLISH επικοινω-νούν πλέον µέσω µηνυµά-των ή email οι περισσότεροι νέοι της χώρας µας. Παρά το γεγονός ότι τα τελευ-ταία χρόνια υπάρχουν τα γλωσσικά εργαλεία, τα οποία επιτρέπουν τη χρήση της ελληνικής γραµµατο-σειράς, έφηβοι και νέοι ενήλικες φαίνεται ότι δεν έχουν «αγκαλιάσει» αυτές τις τεχνολογίες. Ο καθη-γητής Γλωσσολογίας, κ. Γιώργος Μπαµπινιώτης, λέει στον «Ε.Τ.»: «Τα greeklish είναι πρόβληµα για την ελληνική γλώσσα, ιδίως για ανθρώπους νέας ηλικίας για έναν καθαρά γλωσσικό λόγο. Με τη χρήση των greeklish αποξενώνονται από τη µορφή της λέξης ή όπως λέµε το ετυµολογικό ίνδαλµα που δηλώνεται µε την ορθογραφία της λέξης και συνδέεται και µε τη ση-µασία της λέξης και µε την προέλευσή της». Ο κίνδυνος, µε τον οποίο έρχονται αντι-µέτωποι οι νέοι άνθρωποι, είναι η αποξένωση από τη γραπτή µορφή της γλώσ-σας. Αυτή η «οικειότητα», όµως, βοηθάει και στην κατανόηση της σηµασίας αλλά και την προέλευση της λέξης. «Αυτή η αποξένωση δεν είναι άνευ σηµασίας», αναφέρει ο ειδικός, ο οποίος εξηγεί ότι η διαδικασία της γραφής βοηθάει να εντυπω-θεί η λέξη και να συνδεθεί µε άλλες οµόρριζες λέξεις. «Οταν χρησιµοποιείται αυτή η µορφή επικοινωνίας, κα-ταστρέφονται, ατονούν. ∆εν είναι προς θάνατο, αλλά θα κάνει ζηµιά», αναφέρει ο κ. Μπαµπινιώτης, ο οποίος συµβουλεύει τους χρήστες να επιλέγουν την ελληνική γραµµατοσειρά.
Γιώργος Μπαµπινιώτης.
Date 30 September 2012 Page 16
Copyright material. This may only be copied under the terms of a Newspaper Licensing Agency agreement (www.nla.co.uk) or with written publisher permission. For external republishing rights see www.nla-republishing.com
Press Campaign: Highlights
http://www.meta-net.eu 22
49KYPIAKH 30 ΣΕΠΤΕΜΒΡΙΟΥ 2012
Η 26η Σεπτεµβρίου έχει καθιε-ρωθεί από το Συµβούλιο τηςΕυρώπης ως η ΕυρωπαϊκήΗµέρα των Γλωσσών, αλλά,
σύµφωνα µε µια νέα ευρωπαϊκή επι-στηµονική έκθεση, οι 21 από τις 30γλώσσες της Ευρώπης -µεταξύ των οποί-ων και η Ελληνική- αντιµετωπίζουν κίν-δυνο ψηφιακής εξαφάνισης. Η έρευνα κρούει τον κώδωνα κινδύ-
νου, καθώς διαπίστωσε ότι η ψηφιακήβοήθεια για τις περισσότερες ευρωπαϊκέςγλώσσες είναι ελλιπής ή απολύτως ανύ-παρκτη για τους χρήστες.
Τις έφαγαν οι κοινέςΗ έκθεση, µε τη µορφή µιας σειράς
Λευκών Βίβλων (µε τίτλο «Γλώσσες στηνΕυρωπαϊκή Κοινωνία της Πληροφορίας»),από το επιστηµονικό δίκτυο ΜΕΤΑ-ΝΕΤ, το οποίο συνενώνει 60 ερευνητικάκέντρα σε 34 χώρες, επισηµαίνει ότι οιγλώσσες που µιλιούνται από σχετικάµικρό αριθµό ανθρώπων κινδυνεύουν,επειδή δεν έχουν τεχνολογική υποστή-ριξη όπως έχουν οι ευρέως χρησιµο-ποιούµενες γλώσσες. Λευκές Βίβλοιέχουν καταρτιστεί για τις εξής ευρω-παϊκές γλώσσες: αγγλικά, βασκικά,βουλγαρικά, γαλικιανά, γαλλικά, γερ-µανικά, δανικά, ελληνικά, εσθονικά,ιρλανδικά, ισλανδικά, ισπανικά, ιταλικά,καταλανικά, κροατικά, λετονικά, λι-θουανικά, µαλτέζικα, νορβηγικά (µπουκ-µόλ και νινόρσκ), ολλανδικά, ουγγρικά,πολωνικά, πορτογαλικά, ρουµανικά,σερβικά, σλοβακικά, σλοβενικά, σουη-δικά, τσεχικά και φινλανδικά. ΚάθεΛευκή Βίβλος είναι γραµµένη στη γλώσ-σα στην οποία αναφέρεται και είναιµεταφρασµένη στα αγγλικά.
Τέσσερις µεγάλοι κίνδυνοιΣύµφωνα µε τη νέα µελέτη, η Ισ-
λανδική, η Λετονική, η Λιθουανική καιη Μαλτέζικη αντιµετωπίζουν τον µε-γαλύτερο κίνδυνο εξαφάνισης σε µιαευρωπαϊκή τεχνολογική κοινωνία, πουολοένα περισσότερο προωθεί τη χρήσησυγκεκριµένων γλωσσών και ιδίως τηςΑγγλικής. Όµως και άλλες γλώσσες,όπως η Ελληνική, η Βουλγαρική, η Ουγ-γρική και η Πολωνική, επίσης κινδυ-νεύουν στον σύγχρονο ψηφιακό κόσµο. Η έρευνα του ΜΕΤΑ-ΝΕΤ, στην οποία
συνέβαλαν περισσότεροι από 200 ειδικοί,αξιολογεί τον κίνδυνο για κάθε γλώσσαµε βάση τέσσερα βασικά κριτήρια σετεχνολογικό/ψηφιακό επίπεδο: την ύπαρ-ξη αυτόµατης µετάφρασης στη συγκε-κριµένη γλώσσα, τη δυνατότητα φωνη-τικής αλληλεπίδρασης, τη δυνατότηταψηφιακής ανάλυσης κειµένου και τηδιαθεσιµότητα των σχετικών ψηφιακώνγλωσσικών πόρων/πηγών.
Οι δυνατέςΗ γλώσσα µε την καλύτερη βαθµο-
λογία στα κριτήρια είναι ασφαλώς ηΑγγλική, που απολαµβάνει τη συγκριτικάκαλύτερη τεχνολογική υποστήριξη (ανκαι όχι την καλύτερη δυνατή), γεγονόςπου διευκολύνει την περαιτέρω εξά-πλωσή της.
Ακολουθούν µε ικανοποιητική ή µέ-τρια τεχνολογική/ψηφιακή υποστήριξηη Ολλανδική, η Γαλλική, η Γερµανική,η Ιταλική και η Ισπανική. Η Ελληνική,όπως επίσης η Βασκική, η Καταλανική,η Πολωνική, η Ουγγρική κ.ά. κατα-τάσσονται στις γλώσσες µε «αποσπα-σµατική» µόνο υποστήριξη, γι’ αυτόακριβώς θεωρούνται γλώσσες υψηλούκινδύνου προς εξαφάνιση.
Δραµατικές διαφορές Σύµφωνα µε τους επιµελητές της µε-
λέτης Χανς Ουζκοράιτ και Γκέοργκ Ρεµ,«υπάρχουν δραµατικές διαφορές στηνυποστήριξη της γλωσσικής τεχνολογίαςανάµεσα στις διάφορες ευρωπαϊκέςγλώσσες και τεχνολογικές περιοχές. Τοχάσµα µεταξύ ‘µικρών’ και ‘µεγάλων’γλωσσών ολοένα και διευρύνεται. Πρέπεινα εξασφαλίσουµε τον εφοδιασµό τωνµικρότερων και λιγότερο πλούσιων -σεψηφιακούς πόρους- γλωσσών µε τιςαπαραίτητες βασικές τεχνολογίες, αλλιώςοι γλώσσες αυτές είναι καταδικασµένεςσε ψηφιακή εξαφάνιση».Ως ελπίδα αυτών των γλωσσών θεω-
ρείται η βελτίωση και η ευρύτερη αξιο-ποίηση του λογισµικού γλωσσικής τε-χνολογίας, το οποίο επιτρέπει τη φω-νητική και τη γραπτή επεξεργασία τωνδιαφόρων γλωσσών. Παραδείγµατα αυτών των δυνατοτή-
των είναι οι ηλεκτρονικοί ορθογραφικοίκαι συντακτικοί διορθωτές κειµένων,οι διαδραστικοί προσωπικοί «βοηθοί»των έξυπνων κινητών τηλεφώνων (π.χ.η Siri στο iPhone), τα συστήµατα αυ-τόµατης µετάφρασης, τα ηλεκτρονικάσυστήµατα διαλόγου των τηλεφωνικώνκέντρων, οι µηχανές αναζήτησης, ησυνθετική φωνή στα συστήµατα πλοή-γησης των αυτοκινήτων. κ.ά.
Το βασικό πρόβληµαΤο σηµαντικό, σύµφωνα µε την έκ-
θεση, είναι όλες αυτές οι δυνατότητεςνα προσφέρονται στους χρήστες και στηµητρική τους γλώσσα που κινδυνεύειµε εξαφάνιση. Χωρίς αποφασιστική δρά-ση, γίνεται η δυσοίωνη πρόβλεψη ότιοι γλώσσες αυτές δύσκολα θα επιβιώσουνστον ψηφιακό κόσµο του 21ου αιώνα.Ένα πρόβληµα είναι ότι το λογισµικό
αυτών των συστηµάτων γλωσσικής τε-χνολογίας στηρίζεται σε στατιστικές µε-θόδους που απαιτούν τεράστιες ποσό-τητες γραπτών ή φωνητικών δεδοµένων,όµως τόσα πολλά δεδοµένα είναι δύσκολονα αποκτηθούν για γλώσσες που οµι-λούνται από σχετικά λίγους ανθρώπους.Εξάλλου, ακόµα και για ευρέως χρη-
σιµοποιούµενες γλώσσες όπως τα αγ-γλικά, η σχετική γλωσσική τεχνολογίαέχει ακόµα αδυναµίες, που είναι π.χ.φανερές στις άκρως ανεπαρκείς και γε-µάτες λάθη αυτόµατες µεταφράσεις. Ηέκθεση προτείνει ότι πρέπει να αναληφθείµια συντονισµένη µεγάλης κλίµακαςπροσπάθεια στην Ευρώπη, προκειµένουσταδιακά να δηµιουργηθούν ή να βελ-τιωθούν οι αναγκαίες τεχνολογίες καινα βοηθηθούν οι γλώσσες που είναι ψη-φιακά παραγκωνισµένες.
Τη γλώσσα µού... έχασαν
Οι περισσότερες ευρωπαϊκές γλώσσες κινδυνεύουν µε ψηφιακή εξαφάνιση
Πρέπει να εξασφαλιστεί ο εφοδιασµός των µικρότερων και λιγότερο πλούσιων-σε ψηφιακούς πόρους- γλωσσών µε τις απαραίτητες βασικές τεχνολογίες
?049-ΚΟΣΜΟΣ 29/09/2012 1:41 ?Μ Page 49
Press Campaign: Highlights
http://www.meta-net.eu 23
Press Campaign: Highlights
http://www.meta-net.eu 24
Press Campaign: Highlights
http://www.meta-net.eu 25
Press Campaign: Highlights
http://www.meta-net.eu 26
Press Campaign: Highlights
http://www.meta-net.eu 27
Website: Visitors Overview
http://www.meta-net.eu 28
began sending out press release
European Day of Languages
unusually high traffic
Website: Visitors’ Cities
http://www.meta-net.eu 29
City with the most visits: Brussels!
Computational Morphology for Europe’s Languages
The State of Computational Morphology for Europe’s Languages
http://www.meta-net.eu 30
Computational Morphology?
q So, what is the state of Computational Morphology support? Do we have precise, good, reliable tools for all European languages?
q Answering this question is a non-trivial, difficult and complex task.
q However, we can provide a rough approximation.
q In META-NET we had a look at 30 languages (Basque, Bulgarian, Catalan, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, Galician, German, Greek, Hungarian, Icelandic, Irish, Italian, Latvian, Lithuanian, Maltese, Norwegian, Polish, Portuguese, Romanian, Serbian, Slovak, Slovene, Spanish, Swedish).
q We gathered data on several aspects that were used to prepare a cross-language comparison, along with statistics, discussions, comparisons, experts’ opinions, etc.
http://www.meta-net.eu 31
Coarse-Grained View
q We investigated four main areas: Machine Translation; Speech; Text Analytics; Language Resources.
q Computational Morphology is covered by Text Analytics.
q Text Analytics comprises, among others,
§ the quality and coverage of existing text analytics technologies (morphology, syntax, semantics),
§ coverage of linguistic phenomena and domains,
§ amount and variety of available applications,
§ quality and coverage of existing lexical resources and grammars.
http://www.meta-net.eu 32
Coarse-Grained View
http://www.meta-net.eu 33
English
good
Dutch French German Italian Spanish
moderate fragmentary
Basque Bulgarian Catalan Czech Danish Finnish Galician Greek Hungarian Norwegian Polish Portuguese Romanian Slovak Slovene Swedish
weak or no support
Croatian Estonian Icelandic Irish Latvian Lithuanian Maltese Serbian
excellent
Text
Ana
lyti
cs
Key Observations
http://www.meta-net.eu 34
q When it comes to Language Technology support, there are massive differences between Europe’s languages and technology areas.
q LT support for English is ahead of any other language.
q Even support for English is far from being perfect.
q The gap between English and the other languages keeps widening!
q Several languages – Icelandic, Latvian, Lithuanian, Maltese – receive this weakest score in all four areas!
Simplified Methodology
http://www.meta-net.eu 35
q Distributed data collection process in the respective countries.
q 30 tables provide data for all languages (tools, resources, gaps etc.).
q Reduce numbers to one final score per language and area.
q Calibration of tables across languages in smaller groups.
q Final scores for each area and language were derived from two central features (quality, coverage), resulting in one big table:
Basque Bulgarian Catalan Croatian Czech Danish Dutch English Estonian Finnish French Galician German Greek Hungarian Icelandic Irish Italian Latvian Lithuanian Maltese Norwegian Polish Portuguese Romanian Serbian Slovak Slovene Spanish Swedish
Tokenization, Morphology (tokenization, POS tagging, morphological analysis/generation)
5 5 5 5 0 5 3,1 4,1 5 4 4 4,1 5 4 4,1 4,1 4,1 3,1 4,1 3 3,1 4,1 5 4,1 5 5 3,1 4,1 5 4,1Parsing (shallow or deep syntactic analysis) 4 4 3 2 5 3,1 2,1 4,1 3,1 3,1 4 4,1 3 2,1 4 4 2 3,1 2,1 1,1 0 3,1 4 3,1 4 3,2 0 3,1 4 4,1Sentence Semantics (WSD, argument structure, semantic roles) 3,1 2,1 2 1,2 3,1 1,1 2,1 3,1 2 2 1,1 2,1 1,1 2 1,2 1,1 0 4 0 1,1 0 3,1 1,3 3,1 4 0 0 2,2 2,1 2Text Semantics(coreferenceresolution, context, pragmatics, inference)
1 2 1,1 0 3 1 2 1,1 2 1 2,1 2,1 2,1 2 0,2 0 0 3 0 1 0 3 1,2 1,2 4,1 0 0 0 2 2,1Advanced Discourse Processing (text structure, coherence, rhetorical structure/RST, argumentative zoning, argumentation,
1 0 2 0 3 1 0 2 0 0 2 0 2,1 1 0 0 0 2 0 1 0 3 1 2 3,1 0 0 0 1 1Information Retrieval(text indexing, multimedia IR, crosslingual IR)
4 2 1,2 2,3 0 3 3 4,1 3 3 4,1 2 3 3,1 1,1 0 3,1 4,1 0 1,2 0 4 2 0 5 3 2,1 0 2 3,1Information Extraction (named entity recognition, event/relation extraction, opinion/sentiment recognition, text
3 3 1,1 3,1 4,1 3 2,1 3,1 2 2 3,1 1,2 3 3 6 1 0 4,1 3 3 0 4 2 3,1 4,1 2 1 2,1 1,1 4Language Generation (sentence generation, report generation, text generation)
0 2 1,2 0,4 4 0 2,1 2 0 2,2 2 0 2 1,1 0 0 3 0 1,2 0 0 3,1 1 0 0 0 0 0 2 2,1Summarization, Question Answering,advanced Information Access Technologies
2 2 0 0,1 3 2,1 2,1 2 2 2 3 1,1 2 1,1 0 0 0 3 0 0,1 0 3,1 2 2,2 4,1 0,1 1 1,1 2,1 1Machine Translation 3,1 2 3,1 1,2 0 1,2 2,2 2,1 2,1 3 3,1 4,1 2,1 1 5 2 2,1 3,1 4 3 2,1 2,2 3 2,1 3,1 0,1 2 3,1 4,1 2,2Speech Recognition 1 3 3 3 2,1 1,2 3,1 4 4 3 4 5 4 3,1 2,2 1,1 3,1 4,1 0 1,1 1 1,1 3,1 2,2 2,1 1 2 2,1 3,1 3,1Speech Synthesis 2,4 3 4 3,1 4 2,1 4 4,1 4 4 4 5 4,1 4,1 4 2,1 3,1 4 3,1 3 4 2,1 5,1 4 2 4 3 3,2 4 3Dialogue Management (dialogue capabilities and user modelling)
0 0 2,2 1 3,1 1 2,1 3,1 3 1,1 3 1 3,1 1,2 0 0 0 3 0 0 0 1,1 1 3 0 0 0 2,1 2 3
Reference Corpora 2,3 4,1 3,1 3,1 5 3,1 2,2 4,1 4 3,1 3,1 5 3,1 3 6 3,1 3,2 3 4,1 4 3 3 4 4,1 1,1 2,2 4,1 4,1 3,1 3,1Syntax-Corpora(treebanks, dependency banks) 2,2 2,1 3 3,1 3,3 1,3 2,2 4,2 2,1 3,2 3 2 3 3,1 5,1 2,2 1,2 3 1 1 0 3,1 4 4 4,1 0 2 3,2 2 3Semantics-Corpora 1 4,1 1 0 3,1 1,2 1,2 3 2 0 1,1 1 1,1 2,1 1,5 0 0 4 1 0 0 2,1 2,2 3,1 2,1 0 0 1,4 2 1Discourse-Corpora 0 2 2 0 2,1 1,3 0 3 2,1 2,1 2 0 2 0 0 0 0 2,2 0 0 0 1,1 1,1 2 2,1 0 1,1 0 3 1Parallel Corpora, Translation Memories 0 2,2 2,1 3 3,1 2,1 2,1 4 2,1 3 3,1 5 2 2 6 1,1 3,2 3,1 3,1 3,1 2,1 4,1 4 2,1 4,1 2,1 2 2,2 3,1 3,2Speech-Corpora (raw speech data, labelled/annotated speech data, speech dialogue data)
2,2 2,1 3,1 3 2,2 1,2 4,1 5,1 3,1 2,1 3,1 4,1 2,1 2,1 2,2 2 2,2 2,1 1 2 2,1 3,2 3 4 2,2 4 2 3,1 2,1 3Multimedia and multimodal data 5 1 2 3,1 2,2 1,2 1,3 1,1 1 2,1 1,2 2,2 1,2 2,1 1 1 1,1 3,1 0 1 0 4,1 1 0 0 1,1 2,1 0 2 1Language Models 2 2 2,1 0 4 3 2,1 5 3 2 3 4,1 3 2,1 3,1 3 0 0 3,1 3,1 3 1 1 0 4 2,1 1,2 2,2 2 4Lexicons, Terminologies 5,1 3,1 3,1 3,1 3,1 4 3,1 4,1 5 4 3,1 4,1 3,1 3 6 3 4 4,1 5 3,1 2,1 5 4 4,1 4,1 4 3,1 2,2 3 4,1Grammars 3,1 3 2 0 2,1 1,3 2,1 3 4 4 3 2 3 1 5,1 3 3 3 3,1 0 0 3,2 4 2,3 2,1 0,1 2,1 2,1 3 3Thesauri, WordNets 4 4,1 2,2 3,1 3,1 3 2,1 4,1 3,1 3,1 1,1 4 2,1 1,1 3,3 3 3,1 3,1 2,1 1 0 0 4 2,2 4 2,1 1,1 3 3 4,1Ontological Resources for World Knowledge (e.g. upper models, Linked Data)
2 3 2,1 0 2,1 1,1 0 4 0 2,1 1,1 1 2,1 2 1 0 0 3,1 1 1,1 0 0 2,2 2 2 0,1 0 0 2 1
Language Technology (Tools, Technologies, Applications)
Language Resources (Resources, Data, Knowledge Bases)
Simplified LR/LT Table (German)
http://www.meta-net.eu 36
0: very low 6: very high
Coarse-Grained View
37
English (4.50)
good
Dutch (3.94) French (3.71) German (3.36) Italian (3.50) Spanish (3.77)
moderate fragmentary
Basque (3.36) Bulgarian (2.80) Catalan (3.21) Czech (3.29) Danish (3.00) Finnish (3.64) Galician (3.43) Greek (2.71) Hungarian (3.79) Norwegian (4.36) Polish (4.07) Portuguese (3.64) Romanian (3.87) Slovak (2.43) Slovene (3.57) Swedish (4.57)
weak or no support
Croatian (2.43) Estonian (3.14) Icelandic (3.50) Irish (3.71) Latvian (3.14) Lithuanian (1.79) Maltese (0.80) Serbian (1.64)
excellent
Text
Ana
lyti
cs
In parenthesis: average scores of the grammatical analysis feature. Several additional categories and features informed and influenced the overall ranking of a language in
one of the five categories. Neither the individual scores nor the avg. scores have been calibrated with regard to the scores assigned to the LT support of other languages. These scores cannot be used for a
cross-language comparison alone; nevertheless, the avg. scores show how the authoring teams perceive the state of the grammatical analysis category for their respective language themselves.
“Grammatical Analysis” Feature
Language QuanBty Availability Quality Coverage Maturity Sustainability Adaptability Average Level of support (Text AnalyBcs)
Basque 4 2.5 4 4 4 2.5 2.5 3.36 fragmentary Bulgarian 2.4 2 3.6 3.6 2.8 2.4 2.8 2.80 fragmentary Catalan 3 2.5 4 4 4 2.5 2.5 3.21 fragmentary CroaBan 2 1.5 3.5 3 2 1 4 2.43 weak/no Czech 4 2 4 4 3 2 4 3.29 fragmentary Danish 3 2 4 4 3 2 3 3.00 fragmentary Dutch 3.6 5.4 4.8 3.6 4.8 3.6 1.8 3.94 moderate English 5 5 5.5 4.5 4.5 3 4 4.50 good Estonian 2.5 3.5 3.2 2.8 4 2.5 3.5 3.14 weak/no Finnish 3.5 3.5 3.5 4 4 3.5 3.5 3.64 fragmentary French 4 4 4 4 4 3 3 3.71 moderate Galician 3 5 4 4 3 2 3 3.43 fragmentary German 4 2.5 4 4 4 2.5 2.5 3.36 moderate Greek 2 1.5 3.5 3 3 3 3 2.71 fragmentary Hungarian 4.5 2 4 4.5 4 3 4.5 3.79 fragmentary Icelandic 2 5.5 4 3 3.5 3.5 3 3.50 weak/no Irish 4 4 3 3 4 4 4 3.71 weak/no Italian 3.5 3 4 5 4 3 2 3.50 moderate Latvian 2.5 2 3 3.5 4 3 4 3.14 weak/no Lithuanian 2 1.5 2.5 2 1.5 1 2 1.79 weak/no Maltese 0.8 0.8 0.8 0.8 0.8 0.8 0.8 0.80 weak/no Norwegian 4 4.5 4 4 4.5 4.5 5 4.36 fragmentary Polish 4 4.5 4.5 4.5 4 4 3 4.07 fragmentary Portuguese 3 3 4 4 4.5 2.5 4.5 3.64 fragmentary Romanian 4 3.5 4 3.6 4.5 3.5 4 3.87 fragmentary Serbian 1 1 2.5 2 2 1.5 1.5 1.64 weak/no Slovak 2 2 3 2 2 3 3 2.43 fragmentary Slovene 2.5 4 4.5 3.5 3 3 4.5 3.57 fragmentary Spanish 3.5 3 5.4 4.5 3.5 3 3.5 3.77 moderate Swedish 4.5 3.5 5 4 5 5 5 4.57 fragmentary
http://www.meta-net.eu 38
Across Categories
q The four area rankings of the 30 languages on the five point scale (from “excellent support” to “weak/no support”) take many different features and factors into account.
q The “grammatical analysis” data are only one single piece of the puzzle – the piece that is closest to Computational Morphology.
q Let’s have a look at the individual White Papers and the languages as they are ranked – from “good support” to “weak/no” support.
q The following ranking is in terms of Text Analytics, the excerpts taken from the White Papers refer to morphological tools.
http://www.meta-net.eu 39
Good Support
q Only language that is considered to have “good support” in terms of Text Analytics is English.
q In comparison to certain other languages and language families, the morphology of English is usually considered as being rather simple and straight-forward.
q Many robust and precise off-the-shelf technologies exist. q This is most probably the main reason why the authors of the white
paper on English do not discuss morphology components at all, nor any related issues or challenges.
http://www.meta-net.eu 40
Moderate Support
q Same trend in this category concerning morphological tools. q Authors mainly discuss other research and technology gaps,
mentioning the existence of, for example, “medium- to high-quality software for basic text analysis, such as tools for morphological analysis and syntactic parsing” (German),
q Some authors mention morphology on a more superficial level (Italian, Spanish) or not at all (Dutch).
q The authors of the white paper on French emphasise that large programmes were set up (1994–2000; 2003–2005) to build a set of basic technologies for French, from spoken and written language resources to spoken and written language processing systems.
http://www.meta-net.eu 41
Fragmentary Support 1/4
q 16 languages only have fragmentary support in Text Analytics. q The respective authoring teams report the existence of one or two
morphological tools per language. q Clear tendency: these tools have limited functionality and a long
history including an unclear copyright situation (Hungarian). q Neither freely nor immediately available (Danish, Romanian). q However, these tools are usually employed in the large office suites
(MS Office, Open Office), localisation frameworks or national search engines (Norwegian, Czech, Slovak).
http://www.meta-net.eu 42
Fragmentary Support 2/4
q Key contributing factor that only few morphological components exist: rich morphological systems; high degree of inflection; lack of morphological distinction for certain nominal cases.
q These linguistic properties make morphological processing, as well as all approaches based primarily on statistics, a challenge (Basque, Polish, Slovene and other languages).
q Special characters and encoding systems are mentioned for languages with alphabets that go beyond plain ASCII: processing words when diacritics are missing (web, email) is a challenge. Experts demand more robust error detection algorithms (Czech).
q Important observation (Basque, Greek): algorithms and approaches developed for English cannot be directly transferred to other languages.
http://www.meta-net.eu 43
Fragmentary Support 3/4
q Languages spoken in smaller countries usually do not receive as much attention and research funding as larger languages in which typically also a larger base of researchers works on building actual technologies, maybe even breaking new ground (Greek).
q Hungarian: a lack of synchronisation between parallel efforts to build morphological processors lead to substantial friction loss. This is why several morphological parsers for Hungarian exist but they use conflicting and incompatible formalisms.
q Some authors discuss related technologies such as, for example, e-learning tools and systems for second language learners that employ complex morphological components (Czech).
http://www.meta-net.eu 44
Fragmentary Support 4/4
q Portugal set up a project in 2005 to enable the development of a set of linguistic resources and components to support the processing of Portuguese. Outcome: large corpus and tools for tokenisation, morphosyntactic tagging, inflection analysis, and lemmatisation.
q Slovakia set up a project to provide processing of Slovak for linguistic research purposes within the National Research and Development Programme. Outcome: tools and data sets that include processors and morphologically annotated corpora.
q French (1994-2000) had a clear head-start over Portuguese and Slovak in addition to a longer, more established research tradition in this area, which is why it was ranked higher.
q The Slovak experts conclude that, while certain morphological tools do exist, “those must be further developed and supported.”
http://www.meta-net.eu 45
Weak or No Support 1/2
q This category concerns eight languages. q A small or very small number of morphological tools or components
exist (Irish) and are used, even in well known applications, but they are neither freely available nor accessible for research purposes.
q Tools are based on very simple approaches that rely on word lists (Lithuanian, Estonian, Croatian).
q Several of these tools have been in development since the 1980ies and are under the control of companies. Researchers often use ispell or aspell (open source) as a technological fallback solution.
q The complex morphology of languages is mentioned in almost all cases along with the statement that morphology processing must be further developed (Icelandic, Estonian, Croatian, Maltese, Serbian).
http://www.meta-net.eu 46
Weak or No Support 2/2
q Authors demand more development for basic morphological tools. q Perceived as very important: to design and model approaches to the
specific linguistic properties of a language without trying to adapt an approach developed for English (Serbian, Estonian).
q One such step is to set up specific language technology programmes, as has been done, among others, in France, Slovakia and Portugal.
q In 2000, the Icelandic government set up a national programme with the aim of supporting institutions and companies in creating resources for Icelandic. Outcome: several projects, huge impact on the national field. Among its results are a full-form morphological database of Modern Icelandic inflections, a balanced morphosyntactically tagged corpus and a training model for data-driven POS taggers and an improved spell checker.
http://www.meta-net.eu 47
Summary
q Solid computational morphology tools only exist for a handful of European languages – i.e., those with many speakers (and funding).
q The smaller the language, the less tools exist. q “Fragmentary” or “weak/no support”: 24 of the 30 languages – very few
tools; very limited functionality; availability is a problem. q In terms of the full NLP stack, computational morphology cannot be
taken for granted, it is by no means a “solved problem”. q More original research off the beaten track (i.e., English) needed.
q More coordination, synergies and research transfer between the languages needed.
q France, Iceland, Portugal, Slovakia show that large, dedicated funding programmes are needed to support the development of basic LRs/LTs.
http://www.meta-net.eu 48
Strategic Research Agenda The META-NET Strategic Research Agenda for Multilingual Europe
http://www.meta-net.eu 49
Three Ingredients
50
Appropriate Programme
Vision & Agenda
Appropriate Actors
Research & Commercialisation
Appropriate Support
Funding
http://www.meta-net.eu
Three Vision Groups
http://www.meta-net.eu 51
q Translation and Localisation (technical documentation, official bulletins, GUI localisation, games, services etc.) § Target stakeholders: large users of translation services, (machine)
translation, software companies, game companies, localisation industry
q Media and Information Services (audiovisual sector, news, digital libraries, portals, search engines etc.) § Target stakeholders: media industries, search engine providers, archives
q Interactive Systems (mobile assistance, dialogue translation, call centres, etc.) § Target stakeholders: mobile software and service providers, telecom
industry, call centres
52 http://www.meta-net.eu
Vision Group Meetings
q Vision Group Translation and Localisation § July 23, 2010 Berlin, Germany § September 28, 2010 Brussels, Belgium § April 7/8, 2011 Prague, Czech Republic
q Vision Group Media and Information Services § September 10, 2010 Paris, France § October 15, 2010 Barcelona, Spain § April 1, 2011 Vienna, Austria
q Vision Group Interactive Systems § September 10, 2010 Paris, France § October 5, 2010 Prague, Czech Republic § March 28, 2011 Rotterdam, The Netherlands
Vision Paper
Vision Group Translation and
Localisation Report
Vision Group Interactive
Systems Report
Vision Group Media and
Information Services Report
Priority Themes Paper
Expert meeting minutes
Expert meeting minutes
Expert meeting minutes
Planning Process
Strategic Research Agenda
2010 2011 2012
Vision Paper
Vision Group Translation and
Localisation Report
Vision Group Interactive
Systems Report
Vision Group Media and
Information Services Report
Priority Themes Paper
Expert meeting minutes
Expert meeting minutes
Expert meeting minutes
Planning Process: Documents
Strategic Research Agenda
2010 2011 2012
www.meta-net.eu [email protected] T: +49 30 23895 1833
The Future European Multilingual Information Society
Vision Paper for a Strategic Research Agenda
“People can’t share knowledge if they don’t speak a common language.” Davenport, Thomas H, and Laurence Prusak, Working Knowledge: How Organizations Manage What They Know, Harvard Business School, Boston, 1997, p. 98.
Join the discussion at www.meta-et.eu/forum
LT 2020 Vision and Priority Themes for Language Technology Research in Europe until the Year 2020 Towards the META-NET Strategic Research Agenda
The development of this paper has been funded by the Seventh Framework Programme and the ICT Policy Support Programme of the Euro-pean Commission under contracts T4ME (Grant Agreement 249119), CESAR (Grant Agreement 271022), METANET4U (Grant Agreement 270893) and META-NORD (Grant Agreement 270899).
Do you have comments, ideas or suggestions
with regard to the content of this document?
Please send them to [email protected] or
discuss them online: http://www.meta-net.eu/sra.
This document is part of the Network of Excellence “Multilingual Europe Technology Alliance (META-NET)”, co- funded by the 7th Framework Programme of the European Commission through the T4ME grant agreement no.: 249119.
A Network of Excellence forging the
Multilingual Europe Technology Alliance
Vision Document
Vision Group Translation and Localisation Results of first two meetings
Editors: Aljoscha Burchardt, Georg Rehm
Dissemination Level: Public
Date: 3 December 2010
This document is part of the Network of Excellence “Multilingual Europe Technology Alliance (META-NET)”, co- funded by the 7th Framework Programme of the European Commission through the T4ME grant agreement no.: 249119.
A Network of Excellence forging the
Multilingual Europe Technology Alliance
Vision Document
Vision Group Media and Information Services: Results of first two meetings
Editors: Maria Koutsombogera, Stelios Piperidis
Dissemination Level: Public
Date: 10 November 2010
This document is part of the Network of Excellence “Multilingual Europe Technology Alliance (META-NET)”, co- funded by the 7th Framework Programme of the European Commission through the T4ME grant agreement no.: 249119.
A Network of Excellence forging the
Multilingual Europe Technology Alliance
Vision Document
Vision Group Interactive Systems: Results of first two meetings
Editors: Joseph Mariani, Bernardo Magnini
Dissemination Level: Public
Date: 28 December 2010
Preparation of the SRA
q Strategic Research Agendas of other initiatives were screened. q Many suggestions as input from Vision Group members. q We discussed procedures, input and structure of the SRA in four
meetings of the META Technology Council. § Brussels, Belgium, November 16, 2010 § Venice, Italy, May 25, 2011 § Berlin, Germany, September 30, 2011 § Brussels, Belgium, June 19, 2012
q Additional input in talks, meetings, workshops, discussions, etc. § Example: Three HLT Expert Meetings organised by the EC (end of 2011)
q Almost 200 experts contributed to the SRA (54% from industry; 46% from research; 4% from national/international institutions).
http://www.meta-net.eu 55
Strategic Research Agenda
http://www.meta-net.eu 56
q Addresses the problems we identified when preparing the white papers.
q Three priority research themes and application/innovation scenarios.
q Can put Europe ahead of its competitors in this technology area.
q >190 contributors; >2 years.
q Presented and discussed at 83 conferences and major workshops.
q Final version ready on Dec. 1, 2012.
q http://www.meta-net.eu/sra
SRA: Contents – Brief Glimpse
http://www.meta-net.eu 57
q Set the stage and describe the Euro-pean situation, the needs and the LT research and industry.
q Discuss the state of IT, predictions and mega-trends.
q Our technology vision for 2020.
q Select and specify priority themes.
q Suggest a model for speeding up innovation.
q Outline proposals for the organisation of research and innovation.
Priority Themes: 3 + 2
q We decided on priority themes that (a) support technology progress, (b) lead to solutions that European society needs and (c) solutions from which European industry will benefit as users or as providers.
§ Translingual Cloud § Social Intelligence and e-Participation
§ Socially-Aware Interactive Assistants
q Two additional themes:
§ European Service Platform for Language Technologies
§ Core Technologies for Language Analysis and Production
http://www.meta-net.eu 58
PT1: Translingual Cloud
q Europe has a big need for translations of publishable quality. q Focus on high-quality translation. q New research paradigms
§ Inclusion of professional translators into the research process
§ Inclusion of technologists into research on human translation processes
q Different technological approaches § Stronger emphasis on the properties of
individual languages § A central role for semantics
q Methods for specific genres & domains
http://www.meta-net.eu 59
Priority Research Theme 1: Translingual Cloud
Anydevice
Target groups: European citizen, language professional, organisations, companies, European
institutions, software applications
Multiple target formats
Single accesspoint
Automatic translation and interpretation
Language checking Post-editing Workbenches for creative
translations Novel translation and authoring
workflows
Quality assurance Computer-supported human
translation Multilingual content production and
text authoring Trusted service centre (privacy,
confidentiality, security of source data)
Services and Technologies:
Crosslingual communication, translation and search
Real-time subtitling, voice-over generation and translating speech from live events
Mobile interactive interpretation
Multilingual content production (media, web, technical, legal documents)
Showcases: translingual spaces for ambient translation
Applications:
Written (twitter, blog, article, newspaper,text with/without metadata etc.) orspoken input (spontaneous spoken
language, video/audio, multiple speakers)
Modular combination of analysis, transfer
and generation models
From very fast but lower quality to slower but very
high quality (including instant quality upgrades)
Exploiting strong monolingual analysis
and generation methods and resources
Multiple target formats
Domain, task and genre specialisation
models
Extending translation with
semantic data and linked open data
PT2: Social Intelligence
q Better decisions by monitoring social media q Inclusion of citizens into collective decision processes q Opinion formation, consensus building, decision making q Evolution of new solutions q New forms of democracy: e-democracy,
massive participation, transparency q Dialogues and debates across language
boundaries and across parties, political alliances, social classes
q Better than binary voting q Documented transparent
decision processes
http://www.meta-net.eu 61
Priority Research Theme 2: Social Intelligence and e-Participation
From shallow to deep, from coarse-grained to
detailed processing techniques
Making language technologies interoperable
with knowledge representa-tion and the semantic web
“Semantification” of the web: tight integration with the Semantic Web and Linked Open Data
Mapping large, heterogeneous, unstructured volumes of online content to structured, actionable
representations
Unleashing social intelligence by detecting and monitoring opinions,
demands, needs and problems
Target groups: European citizen, European institutions, discussion
participants, companies
Make use of the wisdom of the
crowds
Improved efficiency and
quality of decision processes
Understanding influence diffusion across social media
especially social media, comments, blogs, forums
decision-relevant information
support
sentiment analysis and opinion mining including the temporal dimension)
cues
from arbitrary online content
visualising discussions and opinion statements
Services and Technologies:
collective deliberation and e-participation
-wide deliberation on pressing issues
and processes; modeling evolution of opinions
analysis technologies
Applications:
Priority Research Theme 3: Socially-Aware Interactive Assistants
Interacting naturally
with and in groups
Learning and
forgetting information
Adaptable to the user’s needs and preferences and the environment
Include human-computer, human-artificial agent and
computer-mediated human-human communication
Proactive, self-aware,
user-adaptable
Interacts naturally with humans, in any
language and modality
Can be personalised to individual communication
abilities including special needs
Can learn incrementally from all interactions and
other sources of information
recognition
and synthesis, providing expressive voices
understanding
incremental conversational speech
models of human communication
inter-dependencies
priority themes
Services and Technologies:
Applications:
dialogue systems
environment
modalities (visual, tactile, haptic) verbal/non-verbal behaviour, social context
ments, any
vocabulary
recovery,self-
assessment
Multilingualcapabilities
Providers of operational and research technologies and services
ResearchCentres
EuropeanInstitutions
Othercompanies (SMEs,
startups etc.)
NationalLanguageInstitutions
LanguageTechnologyProviders
LanguageService
ProvidersUniversities
EuropeanInstitutions
ResearchCentres
Public Administrations Enterprises LT User
Industries UniversitiesEuropeanCitizens
Beneficiaries/users of the platform
Interfaces (web, speech, mobile etc.)
Priority Research Theme 1:Translingual
Cloud
Priority Research Theme 2:Social Intelligence& e-Participation
Priority Research Theme 3:Socially Aware
Interactive Assistants
European Service Platform for Language Technologies(Cloud or Sky Computing Platform)
Multilingualtechnologies
Textanalytics
Textgeneration
Languagechecking
Sentimentanalysis
Named entityrecognition
Summari-sation
Knowledge accessand management
Information andrelation extraction
LanguageProcessing
LanguageUnderstanding
Knowledge
Emotion/Sentiment
Data protectionToolsData SetsResourcesComponentsMetadataStandardsInterfacesAPIsCataloguesQuality AssuranceData Import/ExportInput/OutputStoragePerformanceAvailabilityScalability
Featu
res
Core Resources & Technologies
Icelandic
French
CatalanItalian
Maltese
Greek
Bulgarian
Romanian
Serbian
Croatian
Slovene Hungarian
Slovak
Czech
German
Danish Lithuanian
Latvian
Estonian
Finnish
Swedish
Norwegian
Basque
SpanishPortuguese
Galician
English
Irish
PolishDutch
Polish
English
Irish
Icelandic
Italian
Maltese
Greek
Bulgarian
Romanian
SerbianCroatian
SloveneHungarian
Slovak
Czech
German
Dutch
DanishLithuanian
Latvian
Estonian
Finnish
Swedish
Norwegian
Basque
Spanish
Portuguese
Galician
French
Catalan
http://www.meta-net.eu 65
Conclusions and Next Steps META-NET
http://www.meta-net.eu 66
Conclusions and Next Steps
q The White Paper Series clearly shows that Computational Morphology cannot and must not be considered a “solved problem”.
q Quite the contrary: several good technologies exist only for a small number of languages; many languages lack adequate support.
q The research community needs to team up to discuss synergies and to boost research and technology transfer between its languages.
q The goal should be adequate, precise, robust, scalable and freely available morphology components for all European languages.
q New challenges and opportunities: real-time processing, web-scale processing of and training on documents using big data technologies such as Hadoop, interoperability and standardisation of data formats, morphology as a service etc.
q The sophisticated applications foreseen in our META-NET SRA are critically dependent on reliable and precise basic processing components — including computational morphology!
http://www.meta-net.eu 67
Conclusions and Next Steps
q Europe is extremely interested in and passionate about its languages. q Our Strategic Research Agenda for LT research and innovation can put
Europe ahead of its competitors in this technology area. q Provides useful and attractive solutions to European society, at the same
time creating huge business opportunities for European industry. q Now is the time to move forward with a continent-wide, systematic push
and to invest in strategic research. A modest investment is required. q We are very confident that we can help build applications that break
down language barriers in Europe and beyond. q This push will generate a countless number of opportunities. q This year is important: H2020 and CEF can provide sufficient resources
to make our visions for Europe’s citizens and economy a reality. q META-FORUM 2013, September 19/20, Berlin, Germany.
http://www.meta-net.eu 68
META-FORUM 2013 — Connecting Europe for New Horizons is an international conference on powerful language technologies for the multilingual information society, the data value chain and the information market place. The two special themes of this year's edition of the conference are Big Data Text Analytics and Multilingual Web Services for Multilingual Europe.
HighlightsKeynote lectures by Daniel Marcu (Chief Science Offi cer, SDL) andWolfgang Wahlster (CEO, German Research Center for Artifi cial Intelligence, DFKI)Horizon 2020 and Connecting Europe Facility (CEF): Current State of PlayDynamic Discussions on:
Technologies for the Multilingual WebMT for ProfessionalsServices for Multilingual EuropeNeeds of Europe's LanguagesConnecting Towards New HorizonsQuality Translation and Innovation
New Stakeholders: GALA (Globalization and Localization Association); NPLD (Network to Promote Linguistic Diversity); Council of Europe Committee of Experts on the Charter of Regional and Minority LanguagesTowards a European Language Technology PlatformPanel discussionsAwards Ceremony: META Prize and META Seal of RecognitionMETA Exhibition (industry and research exhibition – software demos and posters)
Connecting Europe for New Horizons
Economics and Technology — Berlin, Germany
http://www.meta-forum.eu
Register now!
http://www.meta-net.eu
META-FORUM 2013 will be held jointly by META-NET and the German Federal Ministry of Economics and Technology, co-organised with MultilingualWeb-LT, QTLaunchPad and LT Berlin.
Vision GroupTranslation and Localisation
Vision GroupInteractive Systems
Vision GroupMedia and Information Services
StrategicResearchAgenda
META-NET Website
Language White Paper Series
ConneDeliverin
2014-2020TransportEnergyConnect
http://www.meta-net.eu
2010
2011
2012
2013
Horizon 2020
Thank you!
META-FORUM 2013 September 19/20, Berlin http://www.meta-forum.eu
http://www.meta-net.eu http://www.facebook.com/META.Alliance
70
Q/A
META-FORUM 2013 — Connecting Europe for New Horizons is an international conference on powerful language technologies for the multilingual information society, the data value chain and the information market place. The two special themes of this year's edition of the conference are Big Data Text Analytics and Multilingual Web Services for Multilingual Europe.
HighlightsKeynote lectures by Daniel Marcu (Chief Science Offi cer, SDL) andWolfgang Wahlster (CEO, German Research Center for Artifi cial Intelligence, DFKI)Horizon 2020 and Connecting Europe Facility (CEF): Current State of PlayDynamic Discussions on:
Technologies for the Multilingual WebMT for ProfessionalsServices for Multilingual EuropeNeeds of Europe's LanguagesConnecting Towards New HorizonsQuality Translation and Innovation
New Stakeholders: GALA (Globalization and Localization Association); NPLD (Network to Promote Linguistic Diversity); Council of Europe Committee of Experts on the Charter of Regional and Minority LanguagesTowards a European Language Technology PlatformPanel discussionsAwards Ceremony: META Prize and META Seal of RecognitionMETA Exhibition (industry and research exhibition – software demos and posters)
Connecting Europe for New Horizons
Economics and Technology — Berlin, Germany
http://www.meta-forum.eu
Register now!
http://www.meta-net.eu
META-FORUM 2013 will be held jointly by META-NET and the German Federal Ministry of Economics and Technology, co-organised with MultilingualWeb-LT, QTLaunchPad and LT Berlin.
Acknowledgements: This work would not have been possible without the dedication and commitment of our colleagues Aljoscha Burchardt, Kathrin Eichler, Tina Klüwer, Arle Lommel, Felix Sasaki and Hans Uszkoreit (all DFKI), the 60 member organisations of the META-NET network of excellence, the ca. 70 members of the Vision Groups, the ca. 30 members of the META Technology Council, the more than 200 authors of and contributors to the META-NET Language White Paper Series and the ca. 200 representatives from industry and research who contributed to the META-NET Strategic Research Agenda.