Hal Daumé III

I am Hal Daumé III, an Associate Professor in Computer Science (also UMIACS and Linguistics) at the University of Maryland; I was previously in the School of Computing at the University of Utah (CV). Although I'd like to be known for my research in language (computational linguistics and natural language processing) and machine learning (structured prediction, domain adapation and Bayesian methods), I am probably best known for my NLPers blog. I associate myself most with conferences like ACL, ICML, EMNLP and NIPS. At UMD, I'm affiliated with the Computational Linguistics lab, the machine learning reading group, the language science program and the AI group, and interact closely with LINQS and computer vision.

BRAQUE   Braque is a news for researchers site that Percy Liang and I developed to help people stay on top of their research fields. Sign up and try it out!


The best way to reach me is by email at me AT hal3 DOT name, I cannot reply to all emails from prospective students; please read this to ensure that I read your email. For pressing matters, please come visit me in person at AVW 3227, or call my office at 301-405-1073.


Recent publications:

  • Modeling Syntactic and Semantic Structures in Hierarchical Phrase-based TranslationAbstract     Incorporating semantic structure into a linguistics-free translation model is challenging, since semantic structures are closely tied to syntax. In this paper, we propose a two-level approach to exploiting predicate-argument structure reordering in a hierarchical phrase-based translation model. First, we introduce linguistically motivated constraints into a hierarchical model, guiding translation phrase choices in favor of those that respect syntactic boundaries. Second, based on such translation phrases, we propose a predicate-argument structure reordering model that predicts reordering not only between an argument and its predicate, but also between two arguments. Experiments on Chinese-to-English translation demonstrate that both advances significantly improve translation accuracy. [Li+al.] (NAACL 2013)@inproceedings{daume13semanticmt,
       title = {Modeling Syntactic and Semantic Structures in Hierarchical Phrase-based Translation},
       author = {Junhui Li and Philip Resnik and and Hal {Daum\'e III}},
       booktitle = {Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics (NAACL)},
       year = {2013},
       url = {http://hal3.name/docs/#daume13semanticmt},
    }
  • Kernel Regression for Head-Related Transfer Function Interpolation and Spectral Extrema ExtractionAbstract     Head-Related Transfer Function (HRTF) representation and interpolation is an important problem in spatial audio. We present a kernel regression method based on Gaussian process (GP) modeling of the joint spatial-frequency relationship between HRTF measurements and obtain a smooth non-linear representation based on data measured over both arbitrary and structured spherical measurement grids. This representation is further extended to the problem of extracting spectral extrema (notches and peaks). We perform HRTF interpolation and spectral extrema extraction using freely available CIPIC HRTF data. Experimental results are shown. [Luo+al.] (ICASSP 2013)@inproceedings{daume13hrtf,
       title = {Kernel Regression for Head-Related Transfer Function Interpolation and Spectral Extrema Extraction},
       author = {Yuancheng Luo and Dmitry N. Zotkin and me and Ramani Duraiswami},
       booktitle = {Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
       year = {2013},
       url = {http://hal3.name/docs/#daume13hrtf},
    }
  • Measuring Machine Translation Errors in New DomainsAbstract     We develop two techniques for analyzing the effect of porting a machine translation system to a new domain. One is a macro-level analysis that measures how domain shift affects corpus-level evaluation; the second is a micro-level analysis for word-level errors. We apply these methods to understand what happens when a Parliament-trained phrase-based machine translation system is applied in four very different domains: news, medical texts, scientific articles and movie subtitles. We present quantitative and qualitative experiments that highlight opportunities for future research in domain adaptation for machine translation. [Irvine+al.] (TACL 2013)@article{daume13mterrors,
       title = {Measuring Machine Translation Errors in New Domains},
       author = {Ann Irvine and John Morgan and Marine Carpuat and Hal {Daum\'e III} and Dragos Munteanu},
       journal = {Transactions of the Association for Computational Linguistics (TACL)},
       year = {2013},
       url = {http://hal3.name/docs/#daume13mterrors},
    }
  • Discriminatively Enhanced Topic Models [Chaturvedi+al.] (ICDM 2013)@inproceedings{daume13detm,
       title = {Discriminatively Enhanced Topic Models},
       author = {Snigdha Chaturvedi and Hal {Daum\'e III} and Taesun Moon},
       booktitle = {International Conference on Data Mining (ICDM)},
       year = {2013},
       url = {http://hal3.name/docs/#daume13detm},
    }
  • Dynamic Feature Selection for Dependency ParsingAbstract     Feature computation and exhaustive search have significantly restricted the speed of graph-based dependency parsing. We propose a faster framework of dynamic feature selection, where features are added sequentially as needed, edges are pruned early, and decisions are made online for each sentence. We model this as a sequential decision-making problem and solve it by imitation learning techniques. We test our method on 7 languages. Our dynamic parser can achieve accuracies comparable or even superior to parsers using a full set of features, while computing fewer than 30% of the feature templates. [He+al.] (EMNLP 2013)@inproceedings{daume13depfeat,
       title = {Dynamic Feature Selection for Dependency Parsing},
       author = {He He and Hal {Daum\'e III} and Jason Eisner},
       booktitle = {Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP)},
       year = {2013},
       url = {http://hal3.name/docs/#daume13depfeat},
    }
  • A Topical Graph Kernel for Link Prediction in Labeled GraphsAbstract     This paper proposes a solution to the problem of link prediction in labeled graphs with additional text information associated with the nodes. By fitting a topic model on the text corpus and some processing, we compute the topics of interest to a node. We propose a walk based graph kernel which incorporates the node’s interest and thus represents structural as well as textual information. We then make predictions about the existence of unseen links using a kernelized SVM. Our experiments with an author citation network shows that our method is effective and significantly outperforms a network-oriented approach. [Chaturvedi+al.] (MLG 2013)@inproceedings{daume13graphkernel,
       title = {A Topical Graph Kernel for Link Prediction in Labeled Graphs},
       author = {Snigdha Chaturvedi and Hal {Daum\'e III} and Taesun Moon and Shashank Srivastava},
       booktitle = {ICML workshop on Mining and Learning with Graphs (MLG)},
       year = {2013},
       url = {http://hal3.name/docs/#daume13graphkernel},
    }
  • Monolingual Marginal Matching for Translation Model Adaptation [Irvine+al.] (EMNLP 2013)@inproceedings{daume13mm,
       title = {Monolingual Marginal Matching for Translation Model Adaptation},
       author = {Ann Irvine and Chris Quirk and Hal {Daum\'e III}},
       booktitle = {Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP)},
       year = {2013},
       url = {http://hal3.name/docs/#daume13mm},
    }
  • Predictable Dual-View HashingAbstract     We propose a Predictable Dual-View Hashing (PDH) algorithm which embeds proximity of data samples in the original spaces. We create a cross-view hamming space with the ability to compare information from previously incomparable domains with a notion of ‘predictability’. By performing comparative experimental analysis on two large datasets, PASCAL-Sentence and SUN-Attribute, we demonstrate the superiority of our method to the state-of-the-art dual-view binary code learning algorithms. [Rastegari+al.] (ICML 2013)@inproceedings{daume13dvh,
       title = {Predictable Dual-View Hashing},
       author = {Mohammad Rastegari and Jonghyun Choi and Shobeir Fakhraei and Hal {Daum\'{e} III} and Larry S. Davis},
       booktitle = {Proceedings of the International Conference on Machine Learning (ICML)},
       year = {2013},
       url = {http://hal3.name/docs/#daume13dvh},
    }
  • SenseSpotting: Never let your parallel data tie you to an old domainAbstract     Words often gain new senses in new domains. Being able to automatically identify, from a corpus of monolingual text, which word tokens are being used in a previously unseen sense has applications to machine translation and other tasks sensitive to lexical semantics. We define a task, S ENSE S POTTING, in which we build systems to spot tokens that have new senses in new domain text. Instead of difficult and expensive annotation, we build a goldstandard by leveraging cheaply available parallel corpora, targeting our approach to the problem of domain adaptation for machine translation. Our system is able to achieve F-measures of as much as 80%, when applied to word types it has never seen before. Our approach is based on a large set of novel features that capture varied aspects of how words change when used in new domains. [Carpuat+al.] (ACL 2013)@inproceedings{daume13sensespotting,
       title = {{SenseSpotting}: Never let your parallel data tie you to an old domain},
       author = {Marine Carpuat and Hal {Daum\'e III} and Katharine Henry and Ann Irvine and Jagadeesh Jagarlamudi and Rachel Rudinger},
       booktitle = {Proceedings of the Conference of the Association for Computational Linguistics (ACL)},
       year = {2013},
       url = {http://hal3.name/docs/#daume13sensespotting},
    }
  • Prioritized Asynchronous Belief PropagationAbstract     Message scheduling is shown to be very effective in belief propagation (BP) algorithms. However, most existing scheduling algorithms use fixed heuristics regardless of the structure of the graphs or properties of the distribution. On the other hand, designing different scheduling heuristics for all graph structures are not feasible. In this paper, we propose a reinforcement learning based message scheduling framework (RLBP) to learn the heuristics automatically which generalizes to any graph structures and distributions. In the experiments, we show that the learned problem-specific heuristics largely outperform other baselines in speed. [Jiang+al.] (2013)@inproceedings{daume13pabp,
       title = {Prioritized Asynchronous Belief Propagation},
       author = {Jiarong Jiang and Taesun Moon and Hal {Daum\'e III} and Jason Eisner},
       booktitle = {ICML Workshop on Inferning},
       year = {2013},
       url = {http://hal3.name/docs/#daume13pabp},
    }

Recent teaching:


Advisees:

Prospective students:
  • Read this and email me after taking machine learning and/or NLP about potential research.

Current advisees:

Past advisees:

  • Adam Teichert (MS 2009 at Utah, now PhD student at JHU)
  • Scott Alfeld (BS 2008 at Utah, now PhD student at USC)

Upcoming Conferences

(bold = plan to attend):

LocationDue DateNotificationConference Dates
AISTATS 12Canary IslandsPastPast21-23 Apr
ACL 12Jeju, KoreaPast11 Mar09-11 Jul
EMNLP 12Jeju, Korea28 Mar18 May12-14 Jul
CVPR 12Providence, RIPast02 Mar18-20 Jun
ICML 12Edinburgh, Scotland24 Feb30 Apr26 Jun-01 Jul
AAAI 12Toronto, CanadaPast28 Mar22-26 Jul
KDD 12Beijing, ChinaPast04 May12-16 Aug
UAI 12Catalina Island, CA30 Mar01 Jun15-17 Aug
NIPS 12Reno, NV??????

last updated on six may, two thousand fourteen; contact me AT hal3 DOT name