Machine Translation of Arabic Dialects
Chris Callison-Burch

Translating Low Resource Indic languages
Matt Post

Preparations for Cross-Lingual Speech Retrieval without Recognition or Translation
Adam Lopez

Compression Models for Language and Dialect Identification
Paul McNamee

Building Resources for Multi-lingual Sentiment Analysis
Theresa Wilson

Stylometric Analysis of Scientific Articles
Shane Bergsma

Social Power and Its Expression in Written Dialog
Owen Rambo

What Good is Bad ASR?
Mark Dredze

Project Halo: Towards a Knowledgeable Biology Textbook

As part of Project Halo at Vulcan Inc, we are creating an (iPadhosted) “knowledgeable biology textbook”, called Inquire. Inquire allows the user to not only read and browse the textbook, but also to ask questions and get reasoned or retrieved answers back, explore the material through semantic connections, and receive suggestions of useful questions to ask. While the core of Inquire is a formal, hand-crafted knowledge base encoding some of the book’s content, this year we are extending Inquire with capabilities for question-answering directly from the book text itself, integrating technologies for natural language processing, textual entailment, and paraphrasing (including paraphrasing technology from Johns Hopkins), and exploring the extent to which the system can “reason” with biology knowledge directly at the textual level. In this talk I will overview the project, and then describe the textual question-answering component in detail, in particular how the various NLP technologies are being integrated. I will also discuss the interplay being developed between the hand-built knowledge and automatic text-extracted knowledge, and how each can potentially support the other, with the formal knowledge providing inference support for textual reasoning and the text-extracted knowledge being a potential source of new material for the formal knowledge-base.

“Acoustic Cues for Perceptual Speaker Recognition”

While much research has been devoted to the conditional performance of human listeners in speaker recognition tasks, determining how these decisions are made has received less attention.  Automatic speaker recognition has received even more attention from researchers, but progress in this field has focused more on improvements in machine learning rather than increasing perceptual understanding.  In this talk, I will discuss our recent research intended to shine a light on some of these unexplored, fundamental aspects of perceptual speaker recognition.  The conversation will focus around preliminary results from perceptual experiments designed to isolate the effects of individual acoustic cues on human subject performance by using morphed stimuli with those acoustic parameters normalized across all voices.  In the course of this discussion, I will also present previous research on audio database organization with diffusion maps as well as convex demodulation, an algorithm for decomposing an audio signal into its modulator and carrier components within a convex optimization framework.

Bio: Dr. Gregory Sell received his B.A. in Music, Science and Technology in 2005 from Stanford University.  He then stayed on at the school to complete an M.S. in Electrical Engineering (2007) and a Ph.D. in Computer-based Music Theory and Acoustics (2010) as well.  His dissertation work organized audio and musical elements using diffusion mapping, a non-linear mathematical analysis algorithm.  In his time at Stanford, he also developed convex demodulation, a framework for extracting modulators from arbitrary acoustic carriers.  His other research interests include human auditory perception, music information retrieval, and data sonification. He is now an Intelligence Community Postdoctoral Fellow at the University of Maryland, College Park, in the Institute for Systems Research, where he is conducting perceptual experiments to analyze the acoustic cues of speaker recognition.

“Trust in Online Communities”

The scope of this research addresses the role of interpersonal trust between individuals in online communities. Several streams of research to include organizational trust models, trust in virtual settings, speech act theory, identity theory, common bond theory, natural language processing are leveraged to create a novel trust model that is used to determine antecedents for trust and new algorithms for automatic detection of trust in online communities. Online communities continue to grow in the internet and vary from grass roots organizations to communities facilitated by large corporations. Examples of increased use of social networks include seeking health and financial advice. Topics such as these stress the importance of trust between individuals in online communities. Although trust has been widely studied in the literature, the question of how trust evolves in online communities remains as a
research gap. Online communities continue to grow and involve the daily exchange between individuals with no face-to-face interaction. This research aims to model the evolution of trust in online communities leading to the creation of new tools to assist online community managers in moderating communities.

A trust model based on organizational research is defined and tested via an empirical study using a financial investing community. These results are used to create a formal model using social exchange and speech act theories. With a formal model, speech act profiling using N-gram language models and a Hidden Markov Model classifier is investigated for automating the discovery of interpersonal trust relationships.

Bio:
John Piorkowski is PhD student in Information Systems at University of Maryland Baltimore County who is examining trust in online communities.
His research focuses on trust in virtual settings, speech act and identity theories, and natural language processing. In addition, John serves as the
Chief Engineer in the Asymmetric Operations Department at the Johns Hopkins University Applied Physics Laboratory.

“Automatic Speaker Recognition: 8x improvement in 10 years! Nothing that a 6dB SNR cannot undo”

Automatic speaker recognition in uncontrolled environments is a very challenging task due to channel distortions, additive noise and reverberation. To address these issues, speaker recognition systems have evolved from a simple Gaussian Mixture Model architecture to the current state-of-the-art based on i-vector representations. This has led to an 8x performance improvement measured by the latest NIST evaluation data.

In the first part of the talk, I will present an overview of how the systems have evolved over the past ten years and emphasize two important contributions from my work. Namely, the reformulation of Joint Factor Analysis in terms of over complete dictionaries, and the transformation of i-vectors by length normalization. The first contribution results in computational savings and ease of explanation. The second one has impacted the direction followed by the speaker recognition community due to its high performance and analytical tractability.

In the second part, I will analyze the performance degradation of an i-vector system in the presence of additive noise and propose a robust architecture. In particular, I will explore the use of multi-condition training in a multi-classifier setup. This architecture produces excellent results in anticipated conditions and generalizes well to unseen conditions.

Bio:
Daniel Garcia-Romero is a PhD candidate in the Electrical Engineering Department at the University of Maryland, College Park. His research interests are in the broad areas of speech forensics and machine learning. Most of his contributions are in automatic speaker recognition where he has developed systems to participate in 5 NIST evaluations. He has received the best student paper award at the 159^th meeting of the Acoustical Society of America (2010), and the best poster award at the 4^th International Conference on Audio- and Video-Based Biometric Person Authentication, AVBPA (2003).

Language models that model phenomena beyond the narrow window of words used by n-gram models have a long history in natural language processing, but relatively little success.  The reasons for these failures are not perfectly understood, but are related to difficult tradeoffs between expressiveness, the computational complexity of inference, and the typical setting which integrates language models into the search process of the generation task, where their role is to make fine-grained discriminations among similar hypotheses.  At the same time, the kind of knowledge contained in such models seems necessary to address the severe ungrammaticality of the output of many natural language generation tasks, such as speech recognition and machine translation.

In light of these difficulties, we take a step back from integrated approaches to focus on coarser discriminative tasks.  We demonstrate the effectiveness of syntactic features, showing state of the art classification results on a range of tasks that increasingly approximate the original machine translation setting.  Of particular note is the consistent effectiveness of out-of-domain tree fragments with an extended domain of locality.  With lessons from these coarser tasks in mind, we end with a proposal for an integrated model of syntax-based language modeling within a string-to-tree machine translation decoder.

BIO:

Matt Post is a postdoctoral researcher at the Human Language Technology Center of Excellence at Johns Hopkins University.  He earned his Ph.D. from the University of Rochester in March of 2011, where he worked with Dan Gildea on machine translation, language modeling, and grammar learning.  His main research interests are in machine translation and text analytics.

Variational Bayesian Methods for Unsupervised Latent Factor Models of Text and Audio

In this talk, I will discuss variational strategies for fitting two Bayesian models that explain high-dimensional media data in terms of sets of latent factors.

The first model, Latent Dirichlet Allocation (LDA), is a popular model of text corpora that learns to represent documents as mixtures of latent “topic” distributions. We develop an online variational Bayes (VB) algorithm for LDA. Online LDA is based on online stochastic optimization with a natural gradient step, which we show converges to a local optimum of the VB objective function. It can handily analyze massive document collections, including those arriving in a stream. We study the performance of online LDA in several ways, including by fitting a 100-topic topic model to 3.3M articles from Wikipedia in a single pass. We demonstrate that online LDA finds topic models as good as or better than those found with batch VB, and in a fraction of the time.

The second model, Gamma Process Non negative Matrix Factorization (GaP-NMF), is a new Bayesian nonparametric model of audio spectrograms that addresses the problem of latent source discovery and separation in audio recordings. GaP-NMF allows us to discover what sounds (e.g. bass drums, guitar chords, etc.) are present in a recording and to isolate or suppress individual sources. Crucially, this model is able to decide how many latent sources are necessary to model the data. This feature is particularly valuable in this application, since it is impossible to guess a priori how many sounds will appear in a given recording. Although the GaP-NMF model lacks the conditional conjugacy enjoyed by models such as LDA, we are nonetheless able to efficiently fit it to data using a novel variational algorithm.

Bio:

Matthew D. Hoffman is a postdoctoral researcher in the Department of Statistics at Columbia University. He received his Ph.D. in Computer Science from Princeton University in 2010. His research interests are in machine learning, statistical modeling, audio signal processing, content-based music information retrieval, and the associated computational issues.

“Generating Descriptions of Visible Objects”

Abstract:  What do people describe when they look at objects?  Can we model what they say?  (Why does this matter?)  This talk will characterize what makes up a visual description and define some of the methods necessary to automatically generate such language.  Taking this a bit further, I describe an end-to-end prototype system that reads in computer vision output and generates natural language descriptions.  Time permitting, I argue that improving visual descriptions can also improve computer vision, and working on the interaction between the two may lead to advances in both computer vision and natural language generation.  My prototype vision-to-language system, largely developed during the Hopkins summer workshop 2011 in collaboration with vision researchers at Stony Brook and language researchers at U. Maryland, is available at:  http://recognition.cs.stonybrook.edu:8080/~mitchema/midge/

BIO:

Meg Mitchell graduated with a Bachelor’s in Linguistics from Reed College in 2005.  Since then, she has worked at the Center for Spoken Language Understanding at Oregon Health and Science University, helping to automatically diagnose neurological disorders by analyzing the syntactic and phonetic characteristics of spoken language.  She has also received a Master’s in Computational Linguistics from the University of Washington and is working towards a Ph.D in Computing Science at the University of Aberdeen.

“Statistical Modeling and Learning for Machine Translation”

Recent years have seen a flurry of research in translation modeling for statistical machine translation. Widely-used approaches include (1) models based on flat phrase-to-phrase mappings, and (2) models that use syntactic structure. I will present an approach that combines the strengths of these two in a single model. Experiments show that it leads to improved translation quality over state-of-the-art systems. When supervised syntactic parsers are not available, I will show that unsupervised parsers can be substituted with minimal loss in translation quality. I will also discuss the problem of tuning model parameters. An abundance of techniques has been developed by the statistical machine learning community, but the machine translation problem is fundamentally different from other structure prediction tasks in key ways. These differences have caused well-known learning algorithms to lose theoretical guarantees when applied to machine translation. I will propose a novel algorithm that does offer theoretical guarantees and shows empirical advantage over state-of-the-art approaches.

BIO:

Kevin Gimpel is a PhD student in the Language Technologies Institute at Carnegie Mellon University where he is advised by Noah Smith. His research focuses on machine translation, with supporting interests in natural language processing and machine learning. He is also interested in new tasks involving emerging data sources, including social media, movie reviews, and restaurant menus. He interned with the machine translation team at Google during summer 2009 and has been a Sandia National Laboratories Excellence in Science and Technology Fellow since 2010.

“Learning to Efficiently Rank”

Technological advances have led to increases in the types and amounts of data, and there is a great need for developing methods to manage and find relevant information from such data to satisfy user’s information needs. Learning to rank is an emerging discipline at the intersection of machine learning, data mining, and information retrieval. It develops principled machine learning algorithms to construct ranking (i.e., retrieval) models, for finding and ranking relevant information to user queries over large amounts of data. Although learning to rank approaches are capable of learning highly effective ranking functions, they have mostly ignored the important issue of model efficiency (i.e., model speed). Given that efficiency and effectiveness are competing forces that often counteract each other, models that are optimized for effectiveness alone may not meet the strict efficiency requirements when dealing with real-world large-scale data-sets.

My PhD thesis introduces the Learning to Efficiently Rank framework for learning large-scale ranking models that facilitate fast and effective retrieval, by exploiting and optimizing the tradeoffs between model complexity (i.e., speed) and accuracy.  At a basic level, this framework learns ranking models whose speed and accuracy can be explicitly controlled.  I proposed and designed solutions for three problems within this framework: 1) learning large-scale ranking models according to a desired tradeoff between model speed and accuracy; 2) constructing temporally-constrained models capable of returning results under time budgets; 3) breaking through the speed/accuracy tradeoff barrier by developing a novel cascade ranking model, and learning the cascade model structure and parameters with a novel boosting-based learning algorithm. My research extends the conventional effectiveness-centric approach in model learning and takes an efficiency-minded look at building effective retrieval models.  Results show that models learned this way significantly outperform traditional machine-learned models in terms of speed without sacrificing result effectiveness. Moreover, the new models work particularly well when users impose stringent time requirements for ranked retrieval on very large data-sets.

Bio:

Lidan Wang is a PhD candidate in the Computer Science Department at the University of Maryland, College Park. She received her Master’s degree from the Department of Computer Science at the University of Wisconsin, Madison, and Bachelor’s degree from the Department of Computer Science at the University of Florida. Lidan’s research interests lie at the intersection of machine learning, information retrieval, and text and data mining. Lidan’s work focuses on designing large-scale machine learning and information retrieval techniques for learning, mining, and retrieving information at scale. Her PhD dissertation research led to a recent NSF research grant (IIS-1144034), which she co-authored.