<?xml version="1.0" encoding="ISO-8859-1" ?>
<rss version="2.0">
 <channel>
  <title>URCS Seminars &amp; Talks</title>
  <link>http://www.cs.rochester.edu/dept/seminar</link>
  <description>Upcoming Seminars &amp; Talks from the Computer Science Department at the University of Rochester</description>
  <language>en-us</language>
  <image>
   <url>http://www.cs.rochester.edu/images/monalogo.jpg</url>
   <title>Computer Science Department, University of Rochester</title>
   <link>http://www.cs.rochester.edu/dept/seminar</link>
  </image>
  <lastBuildDate>Thu, 13 Nov 2008 22:21:27 -0500</lastBuildDate>
  
   <item>
    <title>Sam Zhao: Paper: &quot;Dependency Parsing by Belief Propagation&quot;</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/379</link>
    <description>[Friday, November 14, 2008 at  2:00 PM in CSB 632] Sam will be leading discussion of the following paper:

&quot;Dependency Parsing by Belief Propagation&quot;
David Smith and Jason Eisner

We formulate dependency parsing as a graphical model 
with the novel ingredient of global constraints. We show 
how to apply loopy belief propagation (BP), a simple and 
effective tool for approximate learning and inference. As 
a parsing algorithm, BP is both asymptotically and em- 
pirically ef&amp;#64257;cient. Even with second-order features or la- 
tent variables, which would make exact parsing consider- 
ably slower or NP-hard, BP needs only O(n3 ) time with 
a small constant factor. Furthermore, such features sig- 
ni&amp;#64257;cantly improve parse accuracy over exact &amp;#64257;rst-order 
methods. Incorporating additional features would in- 
crease the runtime additively rather than multiplicatively. 

http://www.aclweb.org/anthology-new/D/D08/D08-1016.pdf</description>
    <pubDate>Thu, 13 Nov 2008 22:21:27 -0500</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/379</guid>
   </item>
  
   <item>
    <title>Matt Post: AMTA 2008 review</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/374</link>
    <description>[Friday, November 07, 2008 at  2:00 PM in CSB 632] I&#39;ll talk about a few papers presented at AMTA 2008 last month.</description>
    <pubDate>Thu, 06 Nov 2008 16:10:32 -0500</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/374</guid>
   </item>
  
   <item>
    <title>Dan Gildea: Paper: &quot;Attacking Decipherment Problems Optimally with Low-Order N-gram Models&quot; </title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/372</link>
    <description>[Friday, October 31, 2008 at  2:00 PM in CSB 632] Dan will lead discussion of the following paper:

&quot;Attacking Decipherment Problems Optimally with Low-Order N-gram Models&quot; 
Sujith Ravi and Kevin Knight 
In Proceedings of Conference on Empirical Methods in Natural Language Processing (EMNLP), 2008.

http://www.isi.edu/~sravi/pubs/emnlp08_decipher-letter-sub.pdf</description>
    <pubDate>Thu, 30 Oct 2008 22:42:35 -0400</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/372</guid>
   </item>
  
   <item>
    <title>Sam Zhao: Finding massive and accurate translation pairs from the web</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/362</link>
    <description>[Friday, October 03, 2008 at  2:00 PM in CSB 632] Statistical machine translation is a supervised learning problem, requiring a huge amount of training data to make the system produce good
quality translations. Given that high quality bilingual text is
expensive to get, we propose a method to mine massive and accurate
translation pairs from the web. Adding these translation pairs to a translation
system shows significant improvement in terms of translation quality.
We further propose a novel evaluation method using Wikipedia.</description>
    <pubDate>Wed, 01 Oct 2008 22:13:25 -0400</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/362</guid>
   </item>
  
   <item>
    <title>Matt Post: Parsers as language models for statistical machine translation</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/361</link>
    <description>[Friday, October 17, 2008 at  2:00 PM in CSB 632] Practice talk for AMTA 2008 presentation.</description>
    <pubDate>Tue, 30 Sep 2008 09:50:20 -0400</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/361</guid>
   </item>
  
   <item>
    <title>Aravind Joshi: Flexible Composition, Multiple Adjoning and Word Order Variation </title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/359</link>
    <description>[Monday, October 06, 2008 at 11:00 AM in CSB 703] Tree-Local Multi-Component TAGs (called hereafter just MC-TAG for
short) are known to be weakly equivalent to standard TAGs,
however, they can describe structures not derivable in the
standard TAG. There are other variants of MC-TAG, such as MC-TAG
with (a) flexible composition and (b) multiple adjoining of
modifier (non-predicative) auxiliary trees that are also weakly
equivalent to TAGs, but can describe structures not derivable
with MC-TAG.  Our main goal in this paper is to determine the
word order patterns that can be generated in these MC-TAG
variants while respecting semantic dependencies in the grammar
and derivation. We use some word order phenomena such as
scrambling and clitic climbing to illustrate our approach. This
is not a study of scrambling or clitic climbing per se. We do not
claim that the patterns of dependencies that are derivable are
all equally acceptable. Other considerations such as processing
will also come into play.  However, patterns that are not
derivable are predicted to be clearly unacceptable.</description>
    <pubDate>Mon, 29 Sep 2008 10:39:11 -0400</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/359</guid>
   </item>
  
   <item>
    <title>Matt Post: Paper: Online large-margin training of syntactic and structural translation features</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/358</link>
    <description>[Friday, September 26, 2008 at  2:00 PM in CSB 632] Since there is no talk scheduled this week, I&#39;ll instead discuss this upcoming paper from Chiang, Marton, and Resnik.  &lt;a href=&quot;http://www.isi.edu/~chiang/papers/mira.pdf&quot;&gt;Download paper&lt;/a&gt;.</description>
    <pubDate>Fri, 26 Sep 2008 07:56:21 -0400</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/358</guid>
   </item>
  
   <item>
    <title>Ben Van Durme: Open Knowledge Extraction </title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/354</link>
    <description>[Friday, September 19, 2008 at 12:30 PM in CSB 632] We present results for a system designed to perform \emph{Open
Knowledge Extraction}, based on a tradition of compositional language
processing, as applied to a large collection of text derived from the
Web. Evaluation through manual assessment shows that well-formed
propositions of reasonable quality, representing general world
knowledge, given in a logical form potentially useable for inference,
may be extracted in high volume from arbitrary input  sentences. We
compare these results with those obtained in recent work on Open
\emph{Information} Extraction, indicating with some examples the quite
different kinds of output obtained by the two approaches. Finally, we
observe that portions of the extracted knowledge are comparable to
results of recent work on \emph{class attribute} extraction.</description>
    <pubDate>Wed, 17 Sep 2008 08:20:56 -0400</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/354</guid>
   </item>
  
   <item>
    <title>Ding Liu: Improved Tree-to-string Transducer for Machine Translation</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/341</link>
    <description>[Friday, May 23, 2008 at 12:30 PM in CSB 703] We propose three enhancements to the tree-to-string (TTS) transducer for
machine translation:
first-level expansion-based normalization
for TTS templates, a syntactic alignment
framework integrating the insertion of
unaligned target words, and a subtree-based ngram
model addressing the tree decomposition
probability. Empirical results show that
these methods improve the performance of a
TTS transducer based on the standard BLEU-4 metric. We also experiment with
semantic
labels in a TTS transducer, and achieve improvement
over our baseline system.
</description>
    <pubDate>Tue, 20 May 2008 10:52:33 -0400</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/341</guid>
   </item>
  
   <item>
    <title>Carlos Gómez Gallo: Syntactic Production across Clauses</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/338</link>
    <description>[Friday, May 09, 2008 at 12:30 PM in CSB 209] Language production research has focused on planning at the
lexical and syntactic levels. Consequently, little is known
about speakers&#39; strategies for distributing information across
multiple utterances. We present evidence that (a) the overall
amount of information in an intended message affects how
speakers distribute that information across clauses, and (b)
speakers have relatively early access to (at least an estimate of)
message complexity, specifically before word retrieval. The
observed effect is unexpected for competing theories of sentence production based on availability or macro-propositional
accounts. We discuss these findings in an information theoretic
context coupled with a sketch of a model of limited resources
at the level of macro-propositional planning.
</description>
    <pubDate>Tue, 06 May 2008 14:06:01 -0400</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/338</guid>
   </item>
  
   <item>
    <title>Liang Huang: Forest Reranking: Chart Parsing with Non-Local Features</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/337</link>
    <description>[Monday, May 05, 2008 at 11:00 AM in ECE 426] [Note that this talk will take place on Monday and in the ECE department]

Conventional n-best reranking techniques often suffer from the limited scope of the n-best list, which rules out many potentially good alternatives. We instead propose forest reranking, a method that reranks the packed forest of exponentially many parses. Although exact inference is intractable with non-local features, we present an approximate algorithm that makes discriminative training practical over the whole Treebank. Our &amp;#64257;nal result, an F-score of 91.7, outperforms both 50-best and 100-best reranking baselines, and is better than any previously reported systems trained on the Treebank. This method can be viewed as an integration of discriminative reranking with traditional chart parsing based on dynamic programming.

I will also discuss some on-going work on applying forest reranking to machine translation.

--------------------

References:

Liang Huang. Forest Reranking: Discriminative Parsing with Non-Local Features.
To Appear in Proceedings of ACL 2008.
http://www.cis.upenn.edu/~lhuang3/forest-rerank.pdf
</description>
    <pubDate>Mon, 21 Apr 2008 14:31:26 -0400</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/337</guid>
   </item>
  
   <item>
    <title>Benjamin Van Durme: Finding Cars, Goddesses and Enzymes &amp; Open Knowledge Acquisition</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/327</link>
    <description>[Friday, April 25, 2008 at 12:30 PM in CSB 703] I will discuss a simple, TFxIDF like method for acquiring large
numbers of pairs of instances and class labels.  For example,
&quot;go-karting&quot; is-a &quot;outdoor activity&quot;,  or &quot;alsike clover&quot; is-a &quot;forage&quot;.
This method allows for the combination of clusters built using
distributional similarity, plus imperfect lists of labelings derived
via Hearst-like patterns, in a manner allowing for easy trade-off
between precision and recall.

This research was done with Marius Pasca, while an intern at Google.
Papers on this and related work will appear at this year&#39;s ACL and AAAI.

If there is time, I will follow with an overview of current work
showing that so-called &quot;old school&quot; methods of extraction based on
traditional NLP are competitive with state of the art systems recently
developed within the community (where competing systems are built
using shallow methods, giving results not useable for symbolic
inference). In addition, we&#39;ve we shown that our system, KNEXT, can
produce the same sorts of &quot;class attributes&quot; that have thus far been
best acquired through the use of commercial search engine query logs,
a propriety resource currently unavailable to the research community.

This later research was done with Lenhart Schubert, and Ting Qian,
with assistance in result evaluation provided by Daphne Liu and Matt
Post.</description>
    <pubDate>Mon, 21 Apr 2008 10:25:23 -0400</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/327</guid>
   </item>
  
   <item>
    <title>Suzanne Stevenson: Bridging the Gap between Syntax and the Lexicon: Computational Models of Acquiring Multiword Lexemes</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/324</link>
    <description>[Friday, April 18, 2008 at 12:00 PM in CSB 601] (Note change of time and location)

As the study of language has shifted over the past 20 years from a focus on abstract knowledge of language to a focus on actual language use, there&#39;s been increasing awareness that much of what people say does not fall neatly into the traditional model of grammar that separates syntax and the lexicon. In response, there&#39;s been significant work on elements of language that used to be relegated to &quot;the periphery&quot;, including multiword lexemes (collocations, idioms, multiword verbs of various kinds) that exhibit properties of both single words and syntactic phrases. Moreover, theories of child language learning have explored the importance of multiword constructions in guiding the acquisition process. However, computational modeling of child language acquisition has continued to focus largely on word learning and grammar learning. In this talk, I will present our computational linguistic research on identifying and classifying multiword verbs, and discuss preliminary work relating these techniques to the process of child language acquisition.

This is joint work with Afsaneh Fazly, University of Toronto.</description>
    <pubDate>Sun, 13 Apr 2008 18:26:07 -0400</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/324</guid>
   </item>
  
   <item>
    <title>Hao Zhang: Efficient Multi-pass Decoding for Synchronous Context Free Grammars</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/303</link>
    <description>[Friday, February 22, 2008 at 12:30 PM in CSB 703] We take a multi-pass approach to machine
translation decoding when using synchronous
context-free grammars as the translation
model and n-gram language models:
the first pass uses a bigram language model,
and the resulting parse forest is used in the
second pass to guide search with a trigram language
model. The trigram pass closes most
of the performance gap between a bigram decoder
and a much slower trigram decoder, but
takes time that is insignificant in comparison
to the bigram pass. An additional fast decoding
pass maximizing the expected count
of correct translation hypotheses increases the
BLEU score significantly.</description>
    <pubDate>Mon, 18 Feb 2008 19:13:28 -0500</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/303</guid>
   </item>
  
   <item>
    <title>Naushad UzZaman: Bangla Language and Research on Bangla Language Processing: Its Motivation and Impact!</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/302</link>
    <description>[Friday, February 15, 2008 at 12:30 PM in CSB 703] Bangla, also known as Bengali, is the language of more than 250 million people, the majority of whom live in Bangladesh and the Indian State of West Bengal, making it the 5th/6th most widely spoken language in the world. Research on Bangla Language Processing (BLP) possibly had started 10-15 years back, but progress in the past has been difficult because resources were not readily available.  A few years ago, the Center for Research on Bangla Language Processing (CRBLP) of BRAC University was established under the leadership of Professor Mumit Khan.

In this talk, I will start by giving background of Bangla Language, then I will show some projects of CRBLP, including some of my previous work. I will conclude my presentation with the motivation for working on BLP and the impact of research on BLP.  As a developing country with only 41% literacy rate and 35% of the population living under $1 per day, Bangladesh could benefit enormously from the availability of advanced NLP tools.

I also hope to draw attention to attention to the tenth International Mother Language Day, celebrated on February 21.</description>
    <pubDate>Thu, 14 Feb 2008 21:19:44 -0500</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/302</guid>
   </item>
  
   <item>
    <title>James Allen: Current progress on the PLOW project</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/301</link>
    <description>[Friday, February 01, 2008 at 12:30 PM in CSB 703] </description>
    <pubDate>Fri, 01 Feb 2008 11:54:25 -0500</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/301</guid>
   </item>
  
   <item>
    <title>Matt Post: Syntax-based language models for machine translation</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/297</link>
    <description>[Friday, January 25, 2008 at 12:30 PM in CSB 703] </description>
    <pubDate>Mon, 21 Jan 2008 11:05:30 -0500</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/297</guid>
   </item>
  
   <item>
    <title>Charles Parker: Structured Gradient Boosting</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/274</link>
    <description>[Friday, October 05, 2007 at 10:00 AM in 703] The goal of many machine learning problems can be formalized as the creation of a function that can properly classify an input vector, given a set of examples of that function. While this formalism has produced a number of success stories, there are notable situations in which it fails. One such situation arises when the class labels are composed of multiple variables, each of which may be correlated with all or part of the input or output vectors. Such problems, known as structured prediction problems, are common in the fields of information retrieval, computational linguistics, and computer vision, among others. In this talk, I will discuss structured prediction problems and some of the previous approaches to solving them. I will then present a new algorithm, structured gradient boosting, that combines strong points of previous approaches while retaining their generality. More specifically, the algorithm will combine some of the notions of margin maximization present in support vector methods with the speed and flexibility of the structured perceptron algorithm. Finally, I will show a number of novel ways in which this algorithm can be applied effectively, highlighting applications in learning by demonstration and music information retrieval.</description>
    <pubDate>Tue, 06 Nov 2007 19:50:24 -0500</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/274</guid>
   </item>
  
   <item>
    <title>Hao Zhang: Learning Minimum Translation Units with Synchronous Parsing</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/286</link>
    <description>[Friday, November 02, 2007 at 10:00 AM in CSB 703] Multi-word translation pairs are the building blocks of machine translation systems. Most systems utilize a set of heuristics to extract such translation units from word-aligned bilingual corpora. We present a Bayesian learning approach within a structured space to learn basic multi-word translation units unsupervisedly. We show the effectiveness of Variational Bayes (VB) which biases towards sparser distributions of translation pairs. We also show that structural constraints that restrict the learning space to a subspace that is consistent with reliable single-word translation pairs is important and when combined with VB yields superior translation results.

This is a joint work with Chris Quirk at Microsoft Research at Redmond.
</description>
    <pubDate>Tue, 06 Nov 2007 19:49:24 -0500</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/286</guid>
   </item>
  
   <item>
    <title>Mehdi Manshadi: Learning a Probabilistic Model of Event Sequences From Internet Weblog Stories</title>
    <link>http://www.cs.rochester.edu/dept/seminar/view/285</link>
    <description>[Friday, November 09, 2007 at 10:00 AM in CSB 703] One of the central problems in building broad-coverage story understanding systems is generating expectations about event sequences, i.e. predicting what happens next given some arbitrary narrative context. In this talk, I describe how a large corpus of stories extracted from Internet weblogs was used to learn a probabilistic model of event sequences using statistical language modeling techniques. Our approach was to encode weblog stories as sequences of events, one per sentence in the story, where each event was represented as a pair of descriptive key words extracted from the sentence. We then applied statistical language modeling techniques to each of the event sequences in the corpus. We evaluated the utility of the resulting model for the tasks of narrative event ordering and event prediction.</description>
    <pubDate>Tue, 06 Nov 2007 19:48:17 -0500</pubDate>
    <guid>http://www.cs.rochester.edu/dept/seminar/view/285</guid>
   </item>
  
 </channel>
</rss>
