Language, Speech and Multimedia Technologies Observatory
12/19/2011 - 19:20

How mobile applications have been changed forever.
11/17/2011 - 19:45

Siri will likely be taking on a new personal assistant duty soon – that of omniscient navigator. A new article reports, “Traveling is pretty complicated stuff, if you think about it. It’s not just plugging in the destination address, and arriving, on time, with no changes or distractions, to the venue. There might be a last minute changes to the venue – that you find while you’re already on the way, and have to change the route mid-drive. You might be early for an appointment, and would like to just go to the nearest Starbucks to hang out in the meantime. You’re running low on gas. Will you be able to make it to the venue with gas to spare? Should you fill up now, or can you wait till later?” continued…

New Career Opportunities Daily: The best jobs in media.
11/15/2011 - 18:10

Hitz anitzeko terminoen detekzioa eta ulermena ez da arazo erraza. Alemaniako Saarbruken-eko  DFKI laborategi ospetsutik bisitan datorkigun Valia Kordoni ikertzaileak horretaz hitz egingo digu: Nola erauzi automatikoki hitz anitzeko terminoak eta nola erabili horiek gramatika eleanitzak sortzeko.

Gaia: Automated Annotation and Acquisition of Linguistic Knowledge for Efficient Multilingual Grammar Engineering
(Hitz anitzeko terminoen erauzketa automatikoa  gramatika eleanitzak sortzeko).
Tokia:  3.2 aretoa. Informatika Fakultatea
Valia Kordoni (LT-Lab DFKI GmbH & Dept. of Computational Linguistics, Saarland University).
Eguna: Azaroaren 18an

Ordua: 16:00-17:00


In this talk, I mainly deal with automated acquisition
of linguistic knowledge as a means of enhancing
robustness of lexicalised grammars for real life applications. The
case study I focus on in the best part of this talk is
Multiword Expressions (henceforward MWEs). Specifically,
in the first part of the talk I am taking a closer
look at the linguistic properties of MWEs, in particular,
their lexical, syntactic, as well as semantic
characteristics. The term Multiword Expressions has been
used to describe expressions for which the
syntactic or semantic properties of the whole expression
cannot be derived from its parts (cf., Sag et al., 2002),
including a large number of related but distinct
phenomena, such as phrasal verbs (e.g., “come along”),
nominal compounds (e.g., “frying pan”), institutionalised
phrases (e.g., “breadand butter”), and many
others. Jackendoff (1997) estimates the number of MWEs in a
speaker’s lexicon to be comparable to the number of single
However, due to their heterogeneous characteristics,
MWEs present a tough challenge for both
linguistic and computational work (cf., Sag et al., 2002).
For instance, some MWEs are fixed, and do not present
internal variation, such as “ad hoc”, while
others allow different degrees of internal variability and
modification, such as “spill beans” (“spill
several/musical/mountains of beans”). With the observations
about the linguistic properties of MWEs at hand, I turn in
the second part of the talk to methods for the automated
acquisition of these properties for robust grammar
engineering. To this effect, I first investigate
the hypothesis that MWEs can be detected by the distinct statistical
properties of their component words, regardless of their
type, comparing various statistical measures, a
procedure which leads to extremely
interesting conclusions. I then investigate the
influence of the size and quality of different
corpora, using the BNC and the Web search engines Google and
Yahoo. I conclude that, in terms of language usage, web
generated corpora are fairly similar to more
carefully built corpora, like the BNC, indicating that the
lack of control and balance of these corpora are probably
compensated by their size.
Then, I show a qualitative evaluation of the results of
automatically adding extracted MWEs to existing
linguistic resources. To this effect, I first discuss two
main approaches commonly employed in NLP for treating MWEs:
the words-with-spaces approach which models an MWE as a
single lexical entry and it can adequately capture fixed
MWEs like “by and large”, and compositional approaches which
treat MWEs by general and compositional methods of
linguistic analysis, being able to capture more
syntactically flexible MWEs, like “rock boat”, which cannot
be satisfactorily captured by a wordswith-spaces
approach, since this would require lexical entries to be
added for all the possible variations of an MWE (e.g.,
“rock/rocks/rocking this/that/his…boat”). On this basis, I
argue that the process of the automatic addition of
extracted MWEs to existing linguistic resources improves qualitatively,
if a more compositional approach to grammar/lexicon
extension is adopted.
Finally, I also propose that the methods developed for
the acquisition of linguistic knowledge in the case of the
English MWEs can be tuned to enhance robustness
of lexicalised grammars for languages with richer morphology
and freer word order, as is the case of German, and can
benefit from gold standard syntactically and
semantically annotated corpora, for the (semi-automated)
development of which I am briefly
showing a very simple statistical ranking model which
significantly improves treebanking efficiency by
prompting human annotators to the most relevant linguistic
annotation decisions.
10/28/2011 - 13:50

A new article reports, “Perhaps the biggest announcement at Apple’s iPhone event (about one hour from this posting) will be Assistant, Apple’s evolution of the Siri Personal Assistant Software. Siri, you’ll remember, is the company Apple picked up for a rumored $200 million in April of last year for, in Steve Jobs’ words, its “Artificial Intelligence”, not search or speech recognition.”

Before Apple bought the company, Siri described itself as a Virtual Personal Assistant continued…

New Career Opportunities Daily: The best jobs in media.
10/14/2011 - 03:10

(Cross-posted on the Mobile Blog and the Translate Blog

Mobile technology and the web have made it easier for people around the world to access information and communicate with each other. But there’s still a daunting obstacle: the language barrier. We’re trying to knock down that barrier so everyone can communicate and connect more easily.

Earlier this year, we launched an update to Google Translate for Android with an experimental feature called Conversation Mode, which enables you to you translate speech back and forth between languages. We began with just English and Spanish, but today we’re expanding to 14 languages, adding Brazilian Portuguese, Czech, Dutch, French, German, Italian, Japanese, Korean, Mandarin Chinese, Polish, Russian and Turkish.

To use Conversation Mode, speak into your phone’s microphone, and the Translate app will translate what you’ve said and read the translation out loud. The person you’re speaking with can then reply in their language, and Conversation Mode will translate what they said and read it back to you.

This technology is still in alpha, so factors like background noise and regional accents may affect accuracy. But since it depends on examples to learn, the quality will improve as people use it more. We wanted to get this early version out to help start the conversation no matter where you are in the world.

We’ve also added some other features to make it easier to speak and read as you translate. For example, if you wanted to say “Where is the train?” but Google Translate recognizes your speech as “Where is the rain?”, you can now correct the text before you translate it. You can also add unrecognized words to your personal dictionary.

When viewing written translation results, you can tap the magnifying glass icon to view the translated text in full screen mode so you can easily show it to someone nearby, or just pinch to zoom in for a close-up view.

Tap the magnifying glass icon to view translations full screen.

Finally, we’ve also optimized the app for larger screens like your Android tablet.

While we work to expand full Conversation Mode to even more languages, Google Translate for Android still supports text translation among 63 languages, voice input in 17 of those languages, and text-to-speech in 24 of them.

Download the Google Translate app in Android Market—it’s available for tablets and mobile phones running Android 2.2 and up.

Posted by Jeff Chin, Product Manager
10/13/2011 - 11:55
Bi hitzaldi izango dira bihar Ixa Taldeko mintegian. Australiatik datoz hizlaria biak, baina David oso ezaguna dugu, ixakide ohia da-eta. Bisitan datorkigu beste behin. 

Tokia: Fakultateko 3.2 gela
Eguna: Urriaren 14a, ostirala
Ordua: 15.00

Izenburua: Word classes in Indonesian: A linguistic reality or a convenient fallacy in natural language processing?
Hizlaria: Meladel Mistica (Australian National University)
Laburpena: In this talk I will be presenting work on Indonesian (Bahasa Indonesia), and the claim that there is no noun-verb distinction within the language as it is spoken in regions such as Riau and Jakarta. We test this claim for the language as it is written by a variety of Indonesian speakers using empirical methods traditionally used in part-of-speech induction. In this study we use only morphological patterns that we generate from a pre-existing morphological analyser. We find that once the distribution of the data points in our experiments match the distribution of the text from which we gather our data, we obtain results that show a significant distinction between the class of nouns and the class of verbs in Indonesian. Furthermore it shows promise that the labelling of word classes may be achieved only with morphological features, which could be applied to out-of-vocabulary items.
 [2] Izenburua: Text classification of patient reports and event-modifier identification for the biomedical literature Hizlaria: David Martinez (NICTA - National ICT Australia)
Laburpena[2]: The first short talk describes the implementation and evaluation of a text classification system of pathology reports for the Royal Melbourne Hospital, which relied on document-level annotations obtained from the medical workflow. We observed that a basic machine learning framework with linguistic features carries the potential to make an impact in their process. The second talk describes our work on modifiers of biomedical events over the BioNLP-2009 dataset. Our system combines a simple bag-of-words method with two grammar-based approaches, namely the English Resource Grammar and the RASP parser. We interpret the output of the respective parsers via MRS (Minimal Recursion Semantics), and feed them into a machine learner. Our results indicate that grammar-based techniques can enhance the accuracy of methods for detecting event modification.
10/11/2011 - 07:40 is striving to stand apart from other search engines by utilizing semantic search technology. According to the article, “Today, Google captures nearly 65% of all U.S. search queries, according to ComScore. But, the granddaddy of question and answer based search sites, isn’t bowing out just yet. Established in 1996, (it was then known as Ask Jeeves and is owned by IAC) has survived stiff competition from old-timers like Yahoo! and Google, but also newcomers, from Facebook to Quora. The company’s latest effort to hold and gain a new piece of the overall search market entails a brand-new Q&A community and mobile apps for iPhone and Android operating systems. How is this going to work?” continued…

New Career Opportunities Daily: The best jobs in media.
10/06/2011 - 01:30

Konputagailu bidez jakin omen daiteke zein den pertsona baten jarrera elkarrizketa batean. Stanford-eko Unibertsitateko Dan Jurafsky
irakasleak esperimentu batean saiatu da hori aztertzen. Elkarrizketa
guztia ulertu ez, baina emaitza onak lortu du bereizten ea zein zen kide
bakoitzaren jarrera. Interbentzioak entzunda mintzakideen jarreraren
lau ezaugarri hauek aztertzen ditu makinak: adiskide-moduan
(Friendliness), deseroso (Awkwardness), limurtu nahian (Flirtation) edo
neutroa (Assertiveness). Beti ez du asmatzen, %68an bakarrik, baina
horrek esan nahi du susmo onak dituela.

Testu edo hizketa batean ordenagailuek ezin dute ulertu pertsonok
ulertzen dugun mailan. Hori jakina da. Baina zerbait bai, konputagailuek
hasi dira zerbait ulertzen. Informazio partzialak dira oraindik, baina
esanahiaren zati bat harrapatzeko ahalmena dute. Stanford-eko
Unibertsitate: ospetsutik Ixa taldera pasa den astean bisitan etorri zitzaigun Dan Jurafsky adituak asko daki horretaz. Berak lankideekin egindako lau esperimentu azaldu zizkigun HAP Masterreko hitzaldi batean, ondo aukeratuak gainera, arlo honetan dauden teknika eta aplikazio posibleen berri emateko:

  1. APLIKAZIOA: Solaskideen jarrera detektatzen.;
    TEKNIKA: ikasketa automatiko gainbegiratua.
  2. APLIKAZIOA: Korreferrentzia (testuan gauza bera modu desberdinetan aipatuta).
    TEKNIKA: Eskuz idatzitako erregelak.
  3. APLIKAZIOA: Testuaren esanahiko erlazioak erauzten.
    TEKNIKA:ikasketa automatiko erdigainbegiratua.
  4. APLIKAZIOA: Testuko gertaerak erauzten.
    TEKNIKA: ikasketa automatiko gainbegiratugabea.

Xelebre samar izan zen lehenengo aplikazioa. Goian esan bezala, gauza
izan dira elkarrizketa batean kide bakoitzak nola jokatzen duen
esateko. Horretarako mila elkarrizketa grabatu zituzten hainbat ikasle
bikoteka jarrita 2005ean. Hizketarekin guztira 60 ordu, eta horien
transkripzioan 800.000 hitz lortu zuten. Datu asko, bai.

Datu guzti horien gainean ikasketa gainbegiratuko teknika bat
aplikatu zuten. Horretarako hainbat ezaugarri prosodiko-linguistikoak
identifikatu zituzten testu eta grabazioetan: tonua (pitch),
interbentzioaren luzera, altu edo baxu hitz egitea, galdera, irria,
besteak esandakoa errepikatzea adostasuna adierazteko, azpimarratzeko
interbentzioak (Wow, That’s true, Oh, great!, Oh, gosh!). osagai horien
sekuentziak ere lortu ziren "adierazpen erregularren" bitartez.

Ondorioetako batzuk xelebreak dira:

  • Nabaritu zuten, adibidez, gehienetan elkarrizketan pertsonak bere
    jokaera aldatzen duela bere kidearen jokaerara hurbiltzeko (edo
    urruntzeko, giro txarra dagoenan).
  • Sistemak hobeto antzematen omen die mutilei noiz jarduten duten giro
    "friendly" ederrean (%71ean asmatzen du), neskei baino (%64ean
    bakarrik). Aldiz, errazago antzematen die neskei limurtzeko jarrera
    (%78) mutilei baino (%65).
  • Automatikoki ikasitakoa aztertuta ikusi dute neskek limurtu nahi
    dutenean honelakoak egiten dituztela: tonua igo, beren buruaz parre
    egin, arinago hitz egin, "I" ("ni") esan, muletilla gehiago erabiltzen
    dituzte (kind of, sort of, a little, I don’t know, I guess); eta
    mutilek, aldiz, honela: tonuaren oinarria igotzen dute (pitch floor),
    "you" ("zu") esan, par egin (adarra jo?) solaskideaz, eta "hitz
    akademikorik" ez erabili (academia, interview, teacher, phd, advisor,
    lab, research, management).

Esperimentuko elkarrizketa batzuk
10/05/2011 - 22:45

The W3C has announced a workshop on Linked Enterprise Data in December. The announcement states, “Linked Data technology offers huge potential for enterprise applications such as the integration and the management of data within and across enterprises. The distributed nature of Linked Data enables loose-coupling for data sharing within and between organizations. With Linked Data, enterprises have a unique opportunity to cooperate in their use of shared data without the costs of extensive coordination.” continued…

New Career Opportunities Daily: The best jobs in media.
09/23/2011 - 22:00

Iturria: Diagnosticimaging

Hizkuntzaren prozesamenduak, ahotsa ezagutzeko softwarearen hurrengo belaunalditzat hartzen denak, erraztu egiten du erradiologia-txostenak laburbiltzea, haietan bilaketak egitea eta datuak aurkitzea. Baina berrikitan egindako ikerlan batek erakutsi du oraindik ere profesioanl askok ez dutela erabiltzen.

Duela 50 urte gutxi gorabehera, hizketa ezagutzeko softwarea hasi ziren erabiltzen osasun-arloan, eta hornitzaileek erradiologia-txostenen emaitzak grabatzeko erabiltzen zuten. Teknologia aldetik egindako hobekuntzek softwarea beste maila batera eraman dute hizkuntzaren prozesamenduarekin, eta orain oso garrantzitsua da kalitatea hobetzeko egiten diren ahaleginen ikuspegitik, esan zuen Brigham & Women´s Hospital-en lan egiten duen erradiologia-esparruko Ronylda Lacson ikertzaileak. Hizkuntzaren prozesamenduak ahotsez egindako kontakizunak hartu eta haiek egituratu eta bilaketak egiteko  prestatzen ditu.

“Hizkuntzaren prozesamenduak medikuei txostenetan bilaketa egokiak egitea bermatzen die” esan zuen  Lacson-ek. “Informazioa hain forma laburbilduan graba dezakete non pazienteen historiak ateratzen dituztenenan aztertzeko, datu esanguratsuen ikuspegi guztiz zehatza izan baitezakete.”
Journal of the American College of Radiology aldizkariaren iraileko zenbakian argitaratu zuen lan batean, Lacson-ek eta bere lankideek hizkuntzaren prozesamenduaren hiru erabilera nagusi identifikatu zituzten. Sofwareak atera ditzake irizpide jakinen araberako grabazioak, emaitza eragingarrien bilaketan laguntzeko. Zenbait bertsiok datu jakinak zehaztea ahalbidetzen dute, hala nola irudi bakoitzeko aurkikuntzak, analisirako eta kalitate-hobekuntzetarako. Alabaina, hizkuntzaren prozesamenduaren erabilera baliagarriena epe luzera, zioen Lacson-ek, funtsezko edukia eta aurkikuntza kritikoak nabarmentzeko txosten laburrak egin ahal izatea da. Beste erradiologo batzuek azter ditzakete laburpen horiek, beren etorkizuneko dokumentazioa hobetzeko.

Lacson-ek esan zuen teknologia ez dela erabiltzen erabil litekeen guztia, baina bere lanak ez ditu aipatzen erabilera-portzentaiak irudigintzaren sektorean. Lacson-en ikerlanaren arabera, badira oztopoak hizkuntzaren prozesamendua modu eraginkorrean inplementatzeko, eta berrikitan Diagnostic Imagin-en irakurleen artean egindako inkesta ez-zientifiko batek erakutsi du, industria gisa, erabiltzen dutenak eta gustatzen zaienak bereiztea ekarri dutela zailtasun horiek.

Erantzuleak guztira 145 izan ziren. Horietatik erdiak gutxi gorabehera pozik zeuden ahotsa ezagutzeko softwarearekin. Baina ia  % 30i ez zitzaion gustatzen. 
Behaztopa horiek informazio-faltak eraginak dira, zioen Columbia Unibertsitatean informatika biomedikoko irakasle den  George Hripcsak, doktoreak. Bere karreraren parte handi batean, Hripcsak-ek hizkuntzaren prozesamenduak ikerketa klinikoan eta pazienteen segurtasunaren aldeko ahaleginetan nola lagundu dezakeen ikertu du, eta esan zuen erronka asko gainditu behar direla inplementazioa hedatzeko.

“Erradiologo askok ez dakite zer programa dauden edo zer egin dezaketen haiekin” esan zuen. “Hori bakarrik ez, gainera erradiologiaren merkatua txikia ere bada. Ez du erakartzen hizkuntzaren prozesamenduko sistemak saldu nahi dituzten enpresen interes handirik."

Gainera, Lacson-ek azpimarratu zuen hizkuntza-prozesamenduko teknologiak eskatzen duen ikaskuntza-kurba pikoa eta softwarearen erabilgarritasuna neurtzeko estandarren falta direla gainditu beharreko oztopoak.

Eragozpen eta guzti, Hripcsak-ek esan zuen, hizkuntzaren prozesamenduak aukera asko eskaintzen dituela medikuntzako formazioa eta pazienteen segurtasuna hobetzeko. Hizkuntza-prozesamendua erabil daiteke pazienteen datu-baseetan aurkikuntza jakin erkideak dituzten fitxa-taldeak bilatzeko, esan zuen. Irakasteko taktika horrek egoiliarrak antzeko ezaugarriak dituzten kasu askoren aurrean jartzen ditu eta haiei aukera ematen die beren diagnostikorako gaitasunak praktikatzeko.
Hizkuntza-prozesamenduko zenbait bertsiok lagundu diezaiekete hornitzaileei talde gisa lan egiten, oharkabean pasatu diren aurkikuntza susmagarriak dituzten adibideak harrapatzeko. Kasu horietan, hizkuntzaren prozesamenduak argi gorri bat pizten digu baldin eta esplorazio erradiologiko batean identifikatu eta pazientearen fitxan erregistratu den edozein asaldura ez bada azkeneraino aztertu.
Pazienteei beren historia klinikoa berehala eskuratzeko aukera ematen dieten osasun-portaleen aroan, hizkuntza -prozesamendua itzulpen-tresna bat izan daiteke, formazio medikorik ez duten pertsonentzat, esan zuen Hripcsak-ek. 
“Pertsona askok oso osasun-kultura eskasa dute” esan zuen. “Eta, garrantzitsua da haiek ulertzea beren erresonantzia magnetikoari edo eskanerrari buruz erradiologoak esaten duena. Hizkuntza-prozesamenduak erradiologoaren txostena hizkera ulerterrazean jar dezake.”

Syndicate content