Accepted Papers and Workshop @ COLING 2018, New Mexico, USA

Accepted Papers @ the International Conference on Computational Linguistics (COLING 2018), Santa Fe, New Mexico, USA,  August 20-25, 2018:

  • Measuring the Diversity of Automatic Image Descriptions – Emiel van Miltenburg, Desmond Elliott and Piek Vossen.
  • Systematic Study of Long Tail Phenomena in Entity Linking – Filip Ilievski, Piek Vossen and Stefan Schlobach.
  • A Deep Dive into Word Sense Disambiguation with LSTM – Minh Le, Marten Postma, Jacopo Urbani and Piek Vossen.
  • Scoring and Classifying Implicit Positive Interpretations: A Challenge of Class Imbalance – Chantal van Son, Roser Morante, Lora Aroyo and Piek Vossen.
  • DIDEC: The Dutch Image Description and Eye-tracking Corpus – Emiel van Miltenburg, Ákos Kádár, Ruud Koolen and Emiel Krahmer

Vacancy @CLTL: Postdoc Researcher NWO project “Framing situations in the Dutch language” (closing Sept. 1, 2018)

Job position: PostDoc Researcher

Fte: 0.8 fte, 4 years (expandable to a full time position with an educational task for the Research Master track Human Language Technology)

Vacancy number : 18254

VU unit: Faculty of Humanities, VU University Amsterdam

Closing date: September 1, 2018

Vrije Universiteit Amsterdam is a leading, innovative and growing university that is at the heart of society and actively contributes to new developments in teaching and research. Our university has ten faculties which span a wide range of disciplines, as well as several institutes, foundations, research centres, and support services. Its campus is located in the fastest-growing economic region in the Netherlands (the Zuidas district of Amsterdam), and provides work for over 4,500 staff and scientific education for more than 23,000 students.

The work is carried out at the Computational Lexicology and Terminology Lab (CLTL), which is led by the Spinozalaureate Prof. Dr. Piek Vossen. CLTL is a renowned research group that studies language understanding, generation and interaction through computer models, based on latest technologies, such as neural networks and (un)supervised data approaches, in combination with symbolic knowledge resources, such as computational lexicons and linked-open data repositories. CLTL also runs the Research Master track Human Language Technology, which attracts many international students. More information on the research projects and the master program can be found at the website: www.cltl.nl.

CLTL is part of the Network Institute (http://networkinstitute.org): an interfacultary research institute in which more than 200 researchers from different disciplines collaborate. CLTL has strong ties with the Faculty of Computer Science and Social Sciences at the VU University, through joined projects, PhDs and education programs. Furthermore, CLTL has a long tradition of collaboration both nationally and internationally with research groups at other universities and companies. In this project, we will specifically collaborate with Prof. Dr. Johan Bos from the University of Groningen and Prof. Dr. Collin Baker from Berkeley University.

The PostDoc will work on the NWO project Framing situations in the Dutch language. Language plays a central role in framing, as we daily choose which nouns and verbs describe or frame, a given situation. For some languages, researchers created databases (called FrameNets) containing rich collections of conceptual schemas (frames) that describe situations from a certain perspective. These frames are connected to words and sentences that express them. Several lexical resources exist for Dutch, but no FrameNet. Moreover, we have limited knowledge of the variation of framing in Dutch and how this compares to other languages. The project’s objectives are 1) to create a unique data set where similar situations are framed by many different sources and texts using a newly developed data-to-text method, 2) to capture the variation in framing these situations in Dutch and other languages, 3) to capture semantic-pragmatic factors underlying the usage of different frames for similar situations, and 4) to develop semantic frame and role annotation software. An additional concrete outcome of this project is a Dutch FrameNet contributing to the renowned Berkeley Multilingual FrameNet project, which assesses the cross-linguistic validity of frames and investigates crosslinguistic variation in framing. The insights, resources and technologies created by this project provide new possibilities for (industrial) data analysts and researchers from the Humanities and Social Sciences. The project is coordinated by

Prof. Dr. Piek Vossen and the PostDoc will collaborate with two PhDs (to be hired) and the research staff at CLTL and the University of Groningen.

Tasks

– Further develop and apply the data-to-text method to collect Dutch, English, and Italian texts that frame the same and similar situations, starting from event-registries and structured data on events and automaticallyacquiring texts that make reference to these events.
– Develop an annotation environment that exploits the structured event data shared by different texts, uses the background FrameNet lexicon and ontology and results in faster and more consistent annotation.
– Train and supervise student-assistents that will carry out the annotation with the environment.
– Define crowd-annotation tasks to expand the student annotations.
– Derive a FrameNet lexicon supported by and as an abstraction of the corpus annotations; deploy this lexicon for supporting further annotations.
– Set up an evaluation framework to test the annotation and the lexicon in acquisition, annotation and lexicalisation cycles.
– Co-supervise one PhD in the project.
– Work on the synthesis of the project in colloboration with the Multilingual FrameNet project of Berkeley University.

Requirements: The ideal candidate:

1. has a PhD in (computational) linguistics or computer science
2. has good knowledge and control of the Dutch language and English, preferably also Italian.
3. has a strong computational linguistics background or a computer science background with experience in natural language processing and machine learning
4. has good programming skills
5. can work well in a team
6. preferably has knowledge/experience of/with FrameNet
7. preferably has knowledge and experience in lexicology

Further particulars:

The position is possibly expanded to a full time position with an educational task for the Research Master track Human Language Technology. We aim at starting the project between September 2018 and January 2019. The appointment will be initially for 1 year After satisfactory evaluation of the initial appointment, it can be extended for a total duration of 4 years.

You can find information about our excellent fringe benefits of employment at www.workingatvu.nl like:

• remuneration of 8,3% end-of-year bonus and 8% holiday allowance
• solid pension scheme (ABP);
• a minimum of 29 holidays in case of full-time employment;
• participation in Individual Choices model;
• a wide range of sports facilities which staff may use at a modest charge.

Salary:

The starting salary will be in accordance with university regulations for academic personnel, and depending on experience, range from a minimum of €3238,- gross per month up to a maximum of €3475,- gross per month (salary scale 10) based on a fulltime employment.

For additional informationplease contact:

Piek Vossen phone: +31 20 59 86457
e-mail: piek.vossen@vu.nl
website: cltl.nl

Application

Applicants are requested to write a letter in which they describe their abilities and motivation, accompanied by a curriculum vitae and one or two references (name(s) and e-mail address(es). Please send your application before August 15 to:

Prof. Dr. Piek Vossen
Faculty of Humanities
Vrije Universiteit Amsterdam

Applications should be send by email: piek.vossen@vu.nl
Please mention the vacancy number in the e-mail header.

Any other correspondence in response to this advertisement will not be dealt with.

Symposium “NLP for the Vaccination Debate”

On Wednesday 27 June, the CLTL organizes the symposium “NLP for the Vaccination Debate” in collocation with the PhD defense of Isa Maks. We discuss some of the recent developments in Natural Language processing and their application to the online vaccination debate. The invited speakers are Sabine Bergler and Antal van den Bosch.

When: Wednesday 27 June 2018 (13:00 – 15:00)
Where: Vrije Universiteit Amsterdam, room Agora-2 (main building)

More information: http://www.cltl.nl/event/symposium-nlp-for-the-vaccination-debate
Registration: https://goo.gl/forms/2Y8Ltgz5WZaFKV803

Vacancy @ CLTL: Scientific Programmer

Immediate Opening in the following position: Scientific Programmer (Dutch below)

Pia_Vacancy_New

The Computational Lexicology and Terminology Lab (CLTL), led by Spinoza prize winner Prof. dr. Piek Vossen, is looking for a scientific programmer with an interest in language technology.

Function title: Scientific Programmer
Fte: 0.6-0.8
VU Faculty: Humanities
Vacancy number: 17341
Closing date: Open until filled

Location: Vrije Universiteit Amsterdam, Netherlands

In the NewsReader project, CLTL has developed a pipeline architecture containing software modules with which Dutch texts can be interpreted semantically. De software determines which events are named, who is involved, where, and when the events have taken place, what the sentiment of the named sources of the events is, et cetera. These interpretations are stored in XLM format, the so called Natural Language Annotation Format (NAF). Furthermore, the software generates a representation in RDF supporting (automatic) reasoning over the data. De RDF representations are stored in the so called Triple store and can be queried by means of SPARQL. The candidate takes care of the management of this unique Dutch Natural Language Processing (NLP) pipeline.

The tasks are being executed in the context of the NWO (Netherlands Organisation for Scientific Research)-roadmap project CLARIAH. In the project cooperations take place with researchers across the Netherlands to develop a research infrastructure for the humanities. There are also cooperations with the department of computer science at the VU, with the eScience institute to develop demonstrators, and with researchers abroad.

Requirements
Candidate is expected to support maintenance, usage and further development of the pipeline mentioned above:

• Standardisation and meta data management;
• Software release and versioning of modules;
• Testing;
• Logging;
• Distributive parallel installations and processing;
• Compilation, installation and packaging (e.g.: VMs, Docker);
• Process management;
• Integration in Virtual Research Environment for students and researchers in the Humanities;
• Installation and maintenance of demonstrators.

Ideal Applicant Requirements:
• MSc/MA in computer science/computational linguistics or equivalent title and/or experience;
• Extensive experience in programming languages among which Java and Python;
• Extensive experience with Unix-like systems (Linux and Mac);
• Experience with working within a team of researchers;
• Service oriented;
• Experience with large scale and complex Big Data processing flows;
• Knowledge of standardisation of data in both NLP and Semantic Web;
• Experience with NLP software;
• (preferably) Experience with Sparql and triple stores;
• (preferably) Experience with web-based clients for visualisation and demonstration.

Further particulars
The appointment will be initially for a period of 1 year with the possibility of an extension.

For the completion of the CLARIAH tasks a minimum of 0.6 fte and a maximum 0.8 fte is required.

You may find information about our excellent fringe benefits of employment at www.workingatvu.nl including:
• remuneration of 8,3% end-of-year bonus and 8% holiday allowance;
• solid pension scheme (ABP);
• a minimum of 29 holidays in case of full-time employment.

Salary
The salary will be depending on education and experience, and range from a minimum of € 2.588,- gross per month up to a maximum of € 4.084,- gross per month (salary scale 10) based on a fulltime employment.

Information
For additional information please contact:
Prof dr. Piek Vossen
phone: 020 59 86457
e-mail: piek.vossen@vu.nl
website: www.cltl.nl

Application
Applicants are requested to write a letter in which they describe their abilities and motivation, accompanied by a curriculum vitae and a list of software projects executed and publications.

Please send your application to: piek.vossen@vu.nl

Vrije Universiteit Amsterdam
Attn. Faculty of Humanities
Prof dr. Piek Vossen

Please mention the vacancy number in the e-mail header or at the top of your letter and on the envelope.

Any other correspondence in response to this advertisement will not be dealt with.

Vrije Universiteit Amsterdam
Vrije Universiteit Amsterdam is a leading, innovative and growing university that is at the heart of society and actively contributes to new developments in teaching and research. Our university has ten faculties which span a wide range of disciplines, as well as several institutes, foundations, research centres, and support services. Its campus is located in the fastest-growing economic region in the Netherlands (the Zuidas district of Amsterdam), and provides work for over 4,500 staff and scientific education for more than 23,000 students

Pia_Vacancy

Functietitel: Wetenschappelijk Programmeur
Fte: 0.6-0.8
VU eenheid: FGW
Vacaturenummer: 17341
Sluitingsdatum: Open tot ingevuld

Het Computational Lexicology and Terminology Lab (CLTL) onder leiding van Spinozaprijswinnaar Prof. Dr. Piek Vossen zoekt per direct een wetenschappelijke programmeur met interesse voor taaltechnologie. CLTL heeft in het NewsReader project een pipeline architectuur ontwikkeld met software modules waarmee Nederlandse teksten semantisch geïnterpreteerd kunnen worden. De software bepaalt welke gebeurtenissen worden genoemd, wie er bij betrokken zijn, waar en wanneer die hebben plaatsgevonden, wat het sentiment is van de genoemde bronnen over die gebeurtenissen, etc. Deze interpretaties worden opgeslagen in een XML formaat, het zogenaamde Natural Language Annotation Format (NAF). Verder genereert de software een representatie in RDF die het mogelijk maakt om over de data te redeneren. De RDF representaties worden opgeslagen in een zogenaamde Triple store waar ze door middel van SPARQL bevraagd kunnen worden. De kandidaat zal zorgdragen voor het beheer van deze unieke Nederlandse Natural Language Processing (NLP) pipeline.

De werkzaamheden worden uitgevoerd in het NWO-roadmap project CLARIAH waarin samengewerkt wordt met onderzoekers uit heel Nederland om een onderzoeksinfrastructuur te ontwikkelen voor de geesteswetenschappen. Verder wordt er samengewerkt met het departement van computer science bij de VU, het eScience instituut voor demonstrators en met buitenlandse onderzoekers.

Functie-inhoud
De kandidaat wordt verwacht ondersteuning te leveren aan het onderhoud, gebruik en verdere ontwikkeling van deze pipeline:
• Standaardisatie en metadatabeheer;
• Software release en versioning van modules;
• Testing;
• Logging;
• Distributieve parallelle installaties en processing;
• Compilatie, installatie en packaging (bijv. VMs, Docker);
• Procesmanagement;
• Integratie in Virtual Research Environment voor studenten en onderzoekers in de geesteswetenschappen;
• Installatie en onderhoud van demonstrators.

Functie-eisen
• MA in computer science of een vergelijkbare titel en/of ervaring;
• Ruime ervaring met diverse programmeertalen, waaronder Java en Python;
• Ruime ervaring met Unix-achtige systemen (Linux en Mac);
• Ervaring in het werken binnen een team van onderzoekers;
• Servicegericht zijn.
• Ervaring met grootschalige en complexe Big Data processing flows;
• Kennis van standaardisatie van data in zowel NLP als Semantic Web;
• Ervaring met het werken met NLP software;
• (bij voorkeur) ervaring met Sparql en triple stores;
• (bij voorkeur) ervaring met web-based clients voor visualisatie en demonstratie;

Bijzonderheden
De arbeidsovereenkomst wordt in eerste instantie aangegaan voor een periode van
1 jaar. Verlenging van de arbeidsovereenkomst behoort tot de mogelijkheden.
Voor de CLARIAH werkzaamheden is een invulling van minimaal 0.6 fte en maximaal 0.8 fte nodig.

De Vrije Universiteit heeft aantrekkelijke secundaire arbeidsvoorwaarden en regelingen zoals:
• 8,3% eindejaarsuitkering en 8% vakantietoeslag;
• Goede pensioenregeling (ABP);
• Minimaal 29 vakantiedagen bij volledige arbeidsduur;

Salaris
Het salaris bedraagt afhankelijk van opleiding en ervaring minimaal € 2.588,- en maximaal € 4.084,- bruto per maand (salarisschaal 10) bij een voltijds dienstverband.

Informatie
Voor meer informatie kunt u contact opnemen met:
Prof dr. Piek Vossen
tel.: 020 59 86457
e-mail: piek.vossen@vu.nl
website: www.cltl.nl

Sollicitatie
Kandidaten kunnen solliciteren naar deze functie door een motivatiebrief, curriculum vitae en lijst van uitgevoerde softwareprojecten en publicaties onder vermelding van het vacaturenummer in de e-mail header te sturen aan:

Vrije Universiteit Amsterdam
T.a.v. Faculteit der Geesteswetenschappen
Prof dr. Piek Vossen

Het vacaturenummer graag vermelden in de e-mail header of linksboven op uw brief en envelop.

De Vrije Universiteit Amsterdam (VU) is een vooraanstaande, innovatieve en groeiende universiteit die midden in de samenleving staat en actief bijdraagt aan de ontwikkelingen in onderwijs en onderzoek. Onze breed georiënteerde universiteit telt tien faculteiten, verschillende instituten, stichtingen en onderzoekscentra, en ondersteunende diensten. Op de campus aan de snelst groeiende economische regio van Nederland (de Zuidas), werken ruim 4.500 medewerkers en volgen ruim 23.000 studenten wetenschappelijk onderwijs.

Acquisitie naar aanleiding van deze advertentie wordt niet op prijs gesteld.

Research Masters meet Language Industry

MEET & GREET Human Language Technology (CLTL) & Language Industry

20171207_HLT_FooterThe Computational Lexicology and Terminology Lab (CLTL) organized a MEET & GREET between companies and master students on Friday December 08, 2017 13:30 – 18:00.

Research Masters Meet Language Industry
In the afternoon of Friday December 8th, 2017m students from the Humanities Research Master meet companies and organizations interested in students in Language Technology and other disciplines for internships and theses. The meeting is organized by the Computational Lexicology and Terminology Lab at the VU, in cooperation with the VU Humanities Graduate School.

CLTL is one of the world’s leading research institutes in Human Language technology. Prof. Dr. Piek Vossen, recipient of the NWO Spinoza Prize, heads the group of international researchers that are working on interdisciplinary projects, including the Spinoza project ‘Understanding Language by Machines’. At CLTL we are training the next generation language technology experts. The two-year Research Master Human Language Technology is a program by CLTL.

The Meet & Greet is an excellent opportunity to introduce your company or organisation to Human Language Technology students, and for master students to present their research topic or area of expertise to you.

Join our afternoon program in the presence of the Reference Machine, LeoLani a Pepper robot!

Location
Lecture hall HG 10A.00 (main building at Floor 10, Wing: A), Main building , Vrije Universiteit Amsterdam, De Boelelaan 1105, 1081 HV Amsterdam.

Program
13:30 – 14:00 Walk-In / Doors open / Registration & Coffee
14:00 – 14:05 Introduction: Prof. Dr. Piek Vossen
14:05 – 14:45 Company pitches I
14:45 – 15:15 Student pitches I
15:15 – 15:30 Coffee Break
15:30 – 16:15 Company pitches II
16:15 – 16:45 Student pitches II
16:45 – 17:00 Q&A Reference Machine
17:00 – 18:00 Networking drinks

Pepper_Reference_Machine LeoLani, a Reference Machine

9th Global WordNet Conference Jan. 8—12, 2018

GWC 2018

The 9th Global WordNet Conference

8 — 12 January, 2018
Conference venue: Nanyang Technological University (NTU), Singapore

GWC 2018 The 9th Global WordNet Conference

Registration is now open

Who’s Going? Please also attend the event on Facebook

The ninth Global WordNet Conference (GWC 2018) is an opportunity for researchers and developers to present and discuss their latest results on the development, enrichment and exploitation of wordnets for various languages around the world.

This conference is hosted by the Computational Linguistics Lab at Nanyang Technological University, Singapore and the Global WordNet Association.

Conference Chairs:

Christiane Fellbaum, fellbaum@princeton.edu
Piek Vossen, piek.Vossen@vu.nl

Local Organizing Chair:

Francis Bond, Luís Morgado da Costa, František Kratochvíl, Takayuki Kuribayashi

Call for Student Assistants

We’re hiring academic assistants!

Are you a Master Student in Linguistics, Computer Science, AI or Communication Science? Do you want to get paid for working in an exciting research project that combines research strengths from different disciplines?

We are always looking for talented students for projects involving computational linguistics, computer science and communication science. Positions are for 1 day per week during the academic year 2017-2018.

Here you can recent annotation projects at CLTL to get an idea:  Annotation projects

2017-2018_Call_for_Student_Assistants_FB

The projects that are now looking for students:

If you are interested but want to know more about the possible projects and what to do please get in touch with Chantal van Son. Otherwise, send a motivation letter and CV to the contact person for each project.

Preferred knowledge and skills are:

  • strong background in linguistics and affinity with technology (programming skills are a plus), or;
  • strong technological background and an interest in language technology.
  • some projects require knowledge of Dutch, some good understanding of English

Why you should apply:

  • You will be taking part in a real research project and become knowledgeable about the research field
  • You will be collaborating with fellow students and researchers and learn how to do interdisciplinary research;
  • Topics of interest can be used for term paper and thesis;
  • You might even have the chance to publish a paper and attend a conference;
  • The work hours are flexible;
  • An excellent opportunity to boost your CV.
  • And you get paid!!!

Call for VU University Research Fellow 2017-2018

Apply for University Research Fellow 2017-2018
Deadline Friday 30 June 2017

Who makes our robots talk?

Who takes up this challenge and the exciting opportunity to work in an inspiring research group that is among the best in the world in the area of natural language understanding?

Spinoza prize winner Prof. dr. Piek Vossen has the honour to invite you to apply for the position of University Research Fellow for the academic year 2017-2018. As a University Research Fellow, you work for one year one day a week on a prestigious research project within the research group of Prof. Vossen: the Computational Lexicology and Terminology Lab (CLTL).Call for VU University Research Fellow 2017-2018Humanoid robots: Pepper by Aldebaran Robotics and SoftBank, and NAO by Aldebaran Robotics.

We recently bought a robot and now want to you to plug in our natural language processing technology so that the robot can respond to people in an intelligent way. If you are a wise girl or wise guy and you are interested in Artificial Intelligence, Natural Language Processing and robotics, then you are the perfect candidate to turn our robots into wise bots.

You will work with a real Pepper or NAO robot. The programming environment is Choregraphe and some programming skills in Python are recommended.

As an URF, you will have the chance to publish a paper and attend a conference. It is also an honorable position that looks great on your CV. You will work with PhD students and PostDocs that do exciting work in the area of natural language understanding. There is an opportunity to present a talking robot at the Weekend of Science (“Weekend van de Wetenschap”) to a general audience and basic school kids in October and your robot can be present with you at the opening of the new Computer Science building in 2018.

When you win the prize your activities will be funded for one day a week for one year starting September 2017.

Piek Vossen appointed Pia Sommerauer as VU Fellow for the 2016-2017 academic year.Piek Vossen appointed Pia Sommerauer as VU Fellow for the 2016-2017 academic year.

If you are interested, send an email to Selene Kolman by Friday 30 June 2017, listing:

— a brief motivation
— your interests and ideas related to Natural Language Processing and robotics
— your (Python) programming skills
— your undergraduate degree
— the master courses you have taken and intend to take
— your list of grades

For more information visit websites below or contact:
Prof. dr. Piek Vossen
Selene Kolman

Further information on VU University Research Fellowship (URF)

Prof. dr. Piek Vossen

Professor Computational Lexicology
Language, Literature and Communication
Faculty of Humanities, VU University
de Boelelaan 1105, 1081 HV Amsterdam, The Netherlands

VU University Research Fellow 2015-2016 Soufyan BelkaidPiek Vossen appointed Soufyan Belkaid as VU Fellow for the 2015-2016 academic year.

Piek Vossen appointed Chantal van Son as VU University Research Fellow for the 2014-2015 academic yearPiek Vossen appointed Chantal van Son as VU Fellow for the 2014-2015 academic year.

Controversy in Web Data — ADS Coffee & Data

ADS Coffee & Data: Controversy in Web Data
by Amsterdam Data Science
Screen Shot 2017-06-07 at 15.57.48

Date: Friday 09 June
Time: 0900-1100

Location: VU Amsterdam, HG-16A00 Kerkzaal, 16th floor main building VU
De Boelelaan, Amsterdam, Nederland

Overview: This edition of the ADS meetup will focus on the topic of “How to deal with controversy, bias, quality and opinions on the Web” and will be organised in the context of the COMMIT/ ControCurator project, in which VU and UvA computer scientists and humanities researchers investigate jointly the computational modeling of controversial issues on the Web, and explore its application within real use cases in existing organisational pipelines, e.g. Crowdynews and Netherlands Institute for Sound and Vision.
09:00-09:10 Coffee

Introduction & Chair by Lora Aroyo, Full Professor at the Web & Media group, VU Computer Science

09:10-9:20: Kaspar Beelen – Detecting Controversies in Online News Media (UvA, Faculty of Humanities)

09:20-09:30: Benjamin Timmermans – Understanding Controversy Using Collective Intelligence (VU, Computer Science)

09:30-09:45: Gerben van Eerten – Crowdynews deploying ControCurator

09:45-10:00: Davide Ceolin – (VU, Computer Science)

10:00-10:15: Damian Trilling – (UvA, Faculty of Social and Behavioural Sciences)

10:15-10:30: Daan Oodijk (Blendle)

10:30-10:45: Andy Tanenbaum – “Skewing the data”

10:45-11:00: Q&A Coffee

Registration & further information: https://www.meetup.com/Amsterdam-Data-Science/events/239903981/

Minh Le and Antske Fokkens’ long paper accepted for EACL 2017

Title: Tackling Error Propagation through Reinforcement Learning: A Case of Greedy Dependency Parsing

Conference: EACL 2017 (European Chapter of the Association for Computational Linguistics), at Valencia, 3-7 April 2017.

Authors: Minh Le and Antske Fokkens Title: Tackling Error Propagation through Reinforcement Learning: A Case of Greedy Dependency ParsingTackling Error Propagation through Reinforcement Learning: A Case of Greedy Dependency Parsing by Minh Le and Antske Fokkens

Abstract:
Error propagation is a common problem in NLP. Reinforcement learning explores erroneous states during training and can therefore be more robust when mistakes are made early in a process. In this paper, we apply reinforcement learning to greedy dependency parsing which is known to suffer from error propagation. Reinforcement learning improves accuracy of both labeled and unlabeled dependencies of the Stanford Neural Dependency Parser, a high performance greedy parser, while maintaining its efficiency. We investigate the portion of errors which are the result of error propagation and confirm that reinforcement learning reduces the occurrence of error propagation.