Indiana University

Previous month Previous week Next week Next month
See by year See by month See by week See Today Search Jump to month
Events for the week :
March 03, 2013 - March 09, 2013
Sunday
March 03
Monday
March 04
  • Understanding Ambiguous Sentences and the Winograd Schema Challenge

    Time: 04:00pm - 05:00pm 

    Place: Psychology 128

     

    David Bender

    The last eight years have seen several challenges put forward in the domain of "commonsense reasoning" -- a term mostly used by AI researchers interested in how people understand everyday situations and solve common problems. Recognizing Textual Entailment (RTE), which ran from 2005 to 2011, was followed by Choice of Plausible Alternatives (COPA). Recently, in 2011, Levesque, Davis and Morgenstern proposed the Winograd Schema Challenge (WSC), and while similar in some ways to RTE and COPA, it has a different focus. The Winograd Schema Challenge concerns sentences that contain ambiguous pronoun references. Here's an example:

    Lilly interrupted Donna, breaking her concentration. Whose concentration was broken?
    Lilly interrupted Donna, breaking her silence. Whose silence was broken?

    Average English speakers understand sentences like this seemingly without effort. And yet the task appears to be extremely difficult to model computationally, for several reasons: it draws on immense amounts of contextual knowledge, simple cues that often resolve pronoun ambiguity (e.g. gender disagreement) are forbidden, and the form is designed to resist techniques that aggregate statistics over large corpora to approximate word meaning.

    I'll talk about how humans perform on Winograd Schema sentences, how a state of the art co-reference resolver performs, and I'll discuss one attempt at a computational model for solving them (this from NLP researchers). I'd like to discuss the types of approaches people have generally taken in trying to make progress in the domain of commonsense reasoning, including various systems of logic, attempts at building comprehensive knowledge bases (like Cyc), statistical models, etc. Finally, Levesque et al. intended the Winograd Schema Challenge to replace the Turing Test. Depending on the interests of those present, I'd like to discuss some methodological issues, originally raised against the Turing Test, that are relevant.

     

    In category: Morphosyntax and semantics

     

Tuesday
March 05
  • Circumlocution, Oral Proficiency, and Pedagogical Application

    Time: 04:30pm - 06:00pm 

    Place: IMU Persimmon Room

     

    Yea-Fen Chen (University of Wisconsin-Milwaukee)

    Circumlocution, a component of strategic competence, is a very important strategy for second/foreign language learners. Circumlocution can be a critical skill when students are trying to express themselves but lack the exact vocabulary, especially when in the target culture. Training in this particular strategy will help students build up their confidence and encourage them to make creative use of the L2 knowledge they have acquired. Hence, it will also help students develop their language fluency and obtain more input appropriate for their current proficiency.

    Although it has been mentioned repeatedly in the ACTFL Oral Proficiency Guidelines, circumlocution is seldom taught and even more rarely mastered in a foreign language classroom. This talk will begin with a review of prior studies on circumlocution to set up a framework for circumlocution strategy training. Then suggestions will be made for instruction about and practice in the use of circumlocution in Chinese language classes.

     

    In category: Second language acquisition

     

Wednesday
March 06
  • Learner perception and use of task-based interactional feedback in face-to-face and computer-mediated communication

    Time: 12:15pm - 01:00pm 

    Place: Ballantine 004

     

    Laura Gurzynski-Weiss and Melissa Baralt (Florida International University)

    Theoretical claims about the benefits of corrective feedback have been largely premised on learners' noticing of feedback (e.g., Gass & Mackey, 2006; Long, 1996; Schmidt, 1990, 1995; Swain, 1995), and findings have demonstrated that both the feedback target (Mackey, Gass, & McDonough, 2000) and the mode of provision (Lai & Zhao, 2006) can affect learners' accurate perception of feedback. The current study extended this research by investigating learners' perception and use of feedback provided in task-based interaction in both computer-mediated synchronous chat (CMC) and face-to-face (FTF) modes. Utilizing stimulated recall, the study examined if 26 intermediate-level learners of Spanish as a foreign language accurately noticed feedback as feedback, if they noticed the feedback target, and if the environment in which they interacted (CMC versus FTF) made a difference in their accuracy. The study also investigated if modality affected opportunities for modified output immediately following feedback and if learners used those opportunities differently according to mode. Results demonstrate that overall learners did notice feedback as feedback in both modes. Contrary to expectations, there were no statistical differences between modes in feedback perception accuracy. Significant differences were found, however, in learners' opportunity for and use of feedback depending on the interaction environment and the type of error being addressed.


     

    In category: Second language acquisition

     

  • The distribution of preverbal en in (West) Flemish: Syntactic and interpretive properties

    Time: 01:00pm - 02:15pm 

    Place: Ballantine Hall 003

     

    Liliane Haegeman (Ghent University)

     

    In category: Morphosyntax and semantics

     

  • An intervention account for the distribution of Main Clause Phenomena

    Time: 05:30pm - 06:30pm 

    Place: Oak Room, Indiana Memorial Union

     

    Liliane Haegeman (Ghent University)

    It is well known that certain fronting phenomena in English are by and large restricted to main clause domains. For instance, object fronting is by and large excluded from temporal adverbial clauses:

    (i) *When this news I heard, I was very upset

    Hooper and Thompson (1973) and many accounts inspired by them (Bianchi and Frascarelli 2010) offer a mainly semantic/pragmatic account, claiming that Main Clause Phenomena (MCP) depend on ‘assertion’. They also challenge any attempts to provide a syntactic analysis for the ungrammaticality of (i).

    In my presentation I will argue that a syntactic account for the distribution of (a subset of) MCP is possible, and I will explore two syntactic accounts for the distribution of Main Clause Phenomena. The ‘truncation’ accounts explore H&T’s own intuition that adverbial clauses – and other clauses that resist MCP - are structurally ‘reduced’ and display a deficient left periphery in which the landing site for the argument fronting is missing. Such accounts postulate a specialized landing site for CLLD which survives truncation in Romance. An alternative approach is to interpret the unavailability of MCP as the result of locality restrictions on movement. According to these accounts, adverbial when-clauses, for instance, are derived by operator movement (Geis 1970). Assuming a movement derivation of an adverbial when-clause, the unavailability of argument fronting is like the unavailability of argument fronting in interrogative when-clauses. Likewise, the availability of CLLD in Romance adverbial clauses is parallel to that in wh-clauses. The movement account can be extended to clauses embedded under factive verbs which are also incompatible with MCP.

     

    In category: Morphosyntax and semantics

     

Thursday
March 07
Friday
March 08
  • Natural Language Toolkit

    Time: 09:00am - 11:00am 

    Place: Wells Library Information Commons Cluster 1

     

    Markus Dickinson

    This workshop will cover the Natural Language Toolkit (NLTK), a Python-based platform for natural language processing (NLP) tools (http://nltk.org/). NLTK is useful because it's a "one stop shop" for a variety of applications: storing data, searching data, applying part-of-speech taggers, applying syntactic parsers, classifying documents, and so forth. It does, however, require some understanding of the Python programming language (http://www.python.org), which we'll go over the basics of during the workshop. For those unfamiliar with programming, it may seem like a lot of initial cost to learn some Python, but your efforts will be rewarded by being able to do so many different things.

    In this workshop, we'll cover the basics of text & corpus handling in NLTK, as well as a little bit on POS tagging (time permitting). The goal is to give people enough familiarity with NLTK to bootstrap themselves into more NLTK knowledge.

    If you want to follow along during the tutorial on your own laptop, you are strongly encouraged to install NLTK ahead of time: http://nltk.org/install.html (After that, it won't hurt to install the NLTK data: http://nltk.org/data.html) If you do not have Python on your computer, you will need that, too: http://www.python.org/download/

     

    In category: Computational linguistics

     

  • Perceptual category structure of nonnative speech

    Time: 01:30pm - 03:00pm 

    Place: Psychology 128 (conference room)

     

    Eriko Atagi

    Listeners encode representations of talker-specific features, which facilitate speech processing (e.g., Samuel & Kraljic, 2009). For nonnative speech, such facilitative representations could include talkers’ native language backgrounds. Bradlow and Bent (2008) demonstrated that perceptual adaptation to one nonnative accent generalizes to novel talkers with the same accent, but not to talkers with a novel accent (cf. Baese-Berk, Bradlow, & Wright, 2013). This accent-dependent adaptation suggests that listeners may be creating native language categories as they encounter nonnative speech. To directly investigate the perceptual category structure of nonnative speech and its relationship to intelligibility, 50 native listeners completed a six-alternative forced-choice native language categorization task and a transcription task. The stimuli consisted of English sentences produced by 24 talkers from six native language backgrounds (French, German, Japanese, Korean, Mandarin, Spanish). Listeners were above chance, but not highly accurate, at categorizing the talkers into language background categories (28% mean accuracy). Additionally, no correlation was found between listeners’ categorization and transcription abilities. Data were also modeled using the Similarity Choice Model (Luce, 1963; Shepard, 1957) and additive clustering (Sattath & Tversky, 1977) to examine biases and perceptual similarities. These models showed that the three Asian language backgrounds were perceptually highly similar to each other and very distinct from the other three language backgrounds. Furthermore, when biases were calculated for each language background by talker gender, the direction of biases differed by gender, such that there were higher biases for the female Asian categories and the male non-Asian categories. These results suggest that listeners may not have categorical representations of talkers’ native languages, but are sensitive to talker gender and sociolinguistic similarities of the native languages.

     

    In category: Phonetics and phonology

     

  • Non-canonical agreement in English-Arabic Interlanguage Development

    Time: 02:30pm - 04:00pm 

    Place: Ballantine Hall 205

     

    Boshra El-Ghazoly and Rex A. Sprouse

     

    In category: Second language acquisition

     

Saturday
March 09



JEvents v3.0.9 Stable   Copyright © 2006-2013