achine Reading

An NLP research group at the UCL Computer Science department teaching machines how to read.

The amount of published information is growing rapidly. Much of this information comes in the form of unstructured text which cannot easily be searched, mined, visualized or, ultimately, acted upon. The principal goal of our group is to build machines that can read and "understand" this textual information, converting it into interpretable structured knowledge to be leveraged by humans and other machines alike.

To achieve our goal we work in the intersection of Natural Language Processing, Machine Learning, Cognitive Science, and Information Retrieval. We rely heavily on statistical methods of various flavours.

Our group is part of the UCL Computer Science department, affiliated with CSML and based in the London Media Technology Campus. We are organizing the South England Natural Language Processing Meetup. Get in touch if you're interested in attending.

  • Grants from EU-H2020 on Scalable Understanding of Multilingual Media and from Elsevier on mining scientific publications awarded!
  • 3 papers accepted at EMNLP!
  • Paper on Dependency RNN language models at ACL!
  • Slides and code for our tutorial on Matrix and Tensor Factorization Methods for NLP presented at ACL in Beijing!
  • Paper on Injecting Logical Background Knowledge into Embeddings for Relation Extraction accepted for oral presentation at NAACL!
  • Sebastian Riedel Reader

    Sebastian works in NLP and Machine Learning. He is particularly interested in helping machines to read more accurately by leveraging knowledge gathered through reading more accurately.

  • V. Ivan Sanchez 4th year PhD Student

    I am working on learning interpretable models, such as decision trees and Bayesian networks, from Matrix Factorization models. I'm interested in probabilistic graphical models. I'm funded by CONACYT.

  • Marzieh Saeidi 4th year PhD Student

    Marzieh is interested in urban knowledge extraction and in particular spatial analysis of language data extracted from social media.

  • Matko Bosnjak 3rd year PhD Student

    Matko's interest include the theory and application of machine learning (ML) methods, data, graph and text mining, ML in bioinformatics, graph theory, information retrieval and natural language processing.

  • Tim Rocktäschel 3rd year PhD Student

    Tim is interested in representation learning for NLP and automated knowledge base completion, and how such methods can take advantage of symbolic background knowledge.

  • George Spithourakis 3rd year PhD Student

    I am working on Multi-Instance Text Regression and learning weakly supervised word embeddings. I am interested in structured prediction, distributional semantics, neural models and optimisation. My secondary supervisor is Steffen Petersen and I am funded by the Farr Institute of Health Informatics Research.

  • Jason Naradowsky Research Associate

    Jason is working in collaboration with Google on leveraging knowledge bases of semantic relations and the web's vast quantity of unstructured text to guide the learning of latent variable models for NLP. He is interested in joint inference, graphical models, reinforcement learning, and natural language acquisition.

  • Pontus Stenetorp Research Associate

    Pontus works somewhere in the intersection between Natural Language Processing and Machine Learning. He is particularly interested in representation learning and is currently funded by a machine reading grant from the Allen Foundation.

  • Guillaume Bouchard Senior Research Associate

    After a PhD in Machine Learning and ten years as a researcher at Xerox Research Centre Europe, Guillaume recently joined the Machine Reading Group to pursue his long term research direction: teaching machine to understand language rather waiting that the machine learns it by itself.

  • Fact Checking

    Prime minster said: ”Our government has halved youth unemployment!” True or False? Fact checking is one of the main tasks performed by journalists, especially in an era in which information sources abound. In order to automate it, the main challenges are the open domain nature of the task and the importance of context (temporal, geographical, conversational). Andreas is looking at how to perform fact checking on statistical claims using matrix factorisation methods and distributed representations.

  • Multi-Instance Text Regression

    In the text regression problem, given a document we want to learn how to predict a continuous quantity. The challenge is that not all parts of a document will be equally informative regarding the variable of interest. George is taking a multi-instance approach for this problem, where some of the instances are irrelevant and thus should not influence the overall prediction. There are different possible domains of application, such as aspect ratings, sentiment analysis and clinical datasets.

  • Domain-Independent Language Generation (EPSRC project in collaboration with the Heriott-Watt University)

    Currently weather sites host data (temperature, wind speed, humidity) concerning a multitude of cities and areas around the world. However, only a few locations feature textual weather forecasts as well. To automatically generate forecasts from the data itself, Gerasimos aims to develop a domain-independent Natural Language Generation (NLG) framework by imitating generation policies from unaligned corpora. This will be applied to a variety of domains, as well as weather reports and data obtained from the Met Office.

  • wolfe is a framework for building rich machine learning models, based on functional programming, factor graphs, optimization and composition.
  • ucleed is a biomedical event extractor that ranked first in several tracks of the BioNLP 2011 shared task.
  • thebeast is a Markov Logic inference and learning engine.
  • What's Wrong With My NLP? is a visualizer for NLP problems.