- Topic Modeling
- Natural Language Processing Techniques
- Electoral Systems and Political Participation
- Multimodal Machine Learning Applications
- Political Influence and Corporate Strategies
- European Union Policy and Governance
- Computational and Text Analysis Methods
- European and International Law Studies
- Speech Recognition and Synthesis
- Internet Traffic Analysis and Secure E-voting
- Social Media and Politics
- Flood Risk Assessment and Management
- Political Systems and Governance
- Disaster Management and Resilience
- Speech and dialogue systems
- Software Engineering Research
- Local Government Finance and Decentralization
- Text Readability and Simplification
- Policy Transfer and Learning
- Military History and Strategy
- Advanced Computational Techniques and Applications
- Advanced Queuing Theory Analysis
- Wine Industry and Tourism
- Textile materials and evaluations
- Algorithms and Data Compression
University College Dublin
2014-2023
University of California, Irvine
2022
Microsoft Research (United Kingdom)
2022
Meta (Israel)
2018-2021
Johns Hopkins University
2021
University of Washington
1995-2020
Meta (United States)
2018
Menlo School
2018
Oregon State University
2016
City University of New York
2013
This study analyzes the political agenda of European Parliament (EP) plenary, how it has evolved over time, and manner in which Members (MEPs) have reacted to external internal stimuli when making plenary speeches. To unveil detect latent themes legislative speeches MEP speech content is analyzed using a new dynamic topic modeling method based on two layers Non-negative Matrix Factorization (NMF). applied corpus all English language EP from period 1999 2014. Our findings suggest that...
Parsing accuracy using efficient greedy transition systems has improved dramatically in recent years thanks to neural networks.Despite striking results dependency parsing, however, models have not surpassed stateof-the-art approaches constituency parsing.To remedy this, we introduce a new shiftreduce system whose stack contains merely sentence spans, represented by bare minimum of LSTM features.We also design the first provably optimal dynamic oracle for which runs amortized O(1) time,...
We present a new dataset on decision-making in the European Union (DEUII) that revises and expands previous dataset. Researchers are using this to address range of research questions regarding inputs, processes outputs EU's legislative system. The contains information 331 controversial issues raised by 125 proposals were introduced between 1996 2008. For each these issues, identifies policy alternative favoured most main political actors: Commission; Parliament; member states'...
Recently, neural network approaches for parsing have largely automated the combination of individual features, but still rely on (often a larger number of) atomic features created from human linguistic intuition, and potentially omitting important global context.To further reduce feature engineering to bare minimum, we use bi-directional LSTM sentence representations model parser state with only three positions, which automatically identifies aspects entire sentence.This achieves...
Simultaneous machine translation models start generating a target sequence before they have encoded or read the source sequence. Recent approaches for this task either apply fixed policy on state-of-the art Transformer model, learnable monotonic attention weaker recurrent neural network-based structure. In paper, we propose new mechanism, Monotonic Multihead Attention (MMA), which extends mechanism to multihead attention. We also introduce two novel and interpretable latency control that are...
Neural Machine Translation (NMT) typically leverages monolingual data in training through backtranslation. We investigate an alternative simple method to use for NMT training: combine the scores of a pre-trained and fixed language model (LM) with translation (TM) while TM is trained from scratch. To achieve that, we train predict residual probability added prediction LM. This enables focus its capacity on modeling source sentence since it can rely LM fluency. show that our outperforms...
Much recent effort has been invested in non-autoregressive neural machine translation, which appears to be an efficient alternative state-of-the-art autoregressive translation on modern GPUs. In contrast the latter, where generation is sequential, former allows parallelized across target token positions. Some of latest models have achieved impressive quality-speed tradeoffs compared baselines. this work, we reexamine tradeoff and argue that baselines can substantially sped up without loss...
State-of-the-art neural machine translation models generate a from left to right and every step is conditioned on the previously generated tokens. The sequential nature of this generation process causes fundamental latency in inference since we cannot multiple tokens each sentence parallel. We propose an attention-masking based model, called Disentangled Context (DisCo) transformer, that simultaneously generates all given different contexts. DisCo transformer trained predict output token...
This paper examines member state bargaining success in legislative negotiations the European Union. Bargaining is thought to be determined by factors attributable intervention behaviour, relative policy positions and power. Intervention relates a state’s efforts make its position known over course of negotiations, relate space under negotiation other actors’ positions, power refers size state. New measures for are introduced that account saliency proposals consideration. The results...
Danni Liu, Jan Niehues, James Cross, Francisco Guzmán, Xian Li. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.
This study analyzes political interactions in the European Parliament (EP) by considering how agenda of plenary sessions has evolved over time and manner which Members (MEPs) have reacted to external internal stimuli when making Parliamentary speeches. It does so context speeches are made, content those To detect latent themes legislative time, speech is analyzed using a new dynamic topic modeling method, based on two layers matrix factorization. method applied corpus all English language EP...
Xiang Kong, Adithya Renduchintala, James Cross, Yuqing Tang, Jiatao Gu, Xian Li. Proceedings of the 16th Conference European Chapter Association for Computational Linguistics: Main Volume. 2021.
The ability to amend legislative proposals introduced by the Commission is central process in European Union. Despite this, very few attempts have been made capture and explain such amendments. This study addresses this gap literature considering changes between Commission’s final outcome passed It does so implementing minimum edit distance algorithms measure outcomes. findings suggest that amendments are determined formal informal institutional structures which negotiations take place...
Abstract This article unveils the policy agenda of European Central Bank (ECB) Governing Council as found in speeches that Members gave between 1999 and 2018. Using a dynamic topic‐modeling approached based on non‐negative matrix factorization, we demonstrate how issues discussed by ECB members have evolved over time, general punctuation hypothesis (Jones, B. D. & Baumgartner, F. R. (2005). The politics attention: How government prioritizes problems . University Chicago Press) sheds...
This study examines the evolving role of parliamentary committees in legislative decision-making and their relationships with external actors when amending bills. Drawing on resource dependence theory informational perspectives committees, we analyse impact information provided by business interest organizations, public groups, experts German Bundestag's hearings 2004, 2007, 2011. Using a text-reuse approach to capture amendments, assess influence different types explaining such changes. Our...
Abstract In recent years, transparency (or the lack thereof) has become a central concern of E uropean U nion and its attempts to increase democratic legitimacy legislative decision‐making process. The claim regularly made is that increasing increases potential for holding decision makers account. This study investigates manner in which process affects policy positions taken by negotiators at outset negotiations. findings presented suggest tends lead polarisation negotiations, with taking...
Making interventions during negotiations within the Council of Ministers is primary way in which member states make their policy positions known to one another and attempt influence negotiations. In spite this, relatively little scholarly attention has been paid factors that a state’s decision intervene. This paper seeks address this gap our understanding by analysing new data set specifies are intervening at what level negotiation they doing so. Significant differences between state...
This study examines transparency and censorship in the Council of Ministers European Union from 1999 to 2009. We measure by considering timeliness record release levels applied records when (and if) they are released. show that legislation introduced 2001 (Regulation 1049) triggered a massive shift towards greater transparency, line with its intention. However, we also trend has been interrupted enlargement rounds 2004 2007. attribute this fact inexperience on part new member states...