Autumn School "Methods in Language Comprehension"

The four-day autumn-school aims to expose students to new methods for analyzing language data collected with behavioral and electrophysiological methods. The school’s instructors are experts in corpus-based analysis, EEG, and ERP methods as applied to complex language inputs. Each will present a theoretical module and supervise hands-on sessions where their analyses methods will be implemented on sample datasets. The target audience is advanced M.A. students, Ph.D. students and post-docs involved in studying language comprehension using corpus-based tools, electrophysiology, fMRI or combinations of these techniques. Courses will be held in English.

Venue and dates

The Autumn School will take place in Rovereto, from November 13 to November 16, 2014.

Faculty

•    Harald Baayen (University of Tübingen) - Naive discriminative learning
•    Marcel Bastiaansen (Breda University) - The role of theta, beta, and gamma-band synchronization in sentence processing
•    Falk Huettig (Max Planck Institute for Psycholinguistics) - Using visual world eye-tracking for language research
•    Emmanuel Keuleers (Ghent University) - Megastudies as tools for psycholinguistic investigation

Program

Marcel Bastiaansen
The role of theta, beta, and gamma-band synchronization in sentence processing
During language comprehension, different parts of the brain's language network have to cooperate with millisecond precision in order to derive a message-level understanding from the language input. Oscillatory neuronal dynamics observed in the EEG or MEG provide a window into the network dynamics associated with this extremely complex task. Event-related changes in EEG/MEG power and coherence are thought to capture changes in local and long-range neuronal interactions, respectively, and thus provide the necessary tools to study the dynamics of functional network formation in the brain (Bastiaansen & Hagoort, 2006; Bastiaansen, Mazaheri & Jensen, 2013). In the theoretical module I will outline this analytical framework, and review studies that have addressed the role of theta, beta and gamma-band neuronal synchronization during sentence-level language comprehension. This will set the stage for the hands-on module, in which students will practice with the different data analysis techniques.

Falk Huettig
Using visual world eye-tracking for language research
Language processing tends to be remarkably fast, efficient, and accurate. This is at least partly due to that developing and mature language users anticipate upcoming language input. Visual world eye-tracking can be used to investigate anticipatory language processing. I will discuss visual world research which has focused on the cues used for prediction (e.g., what types of information are used to anticipate upcoming words), the contents of prediction (e.g., what types of representations are activated), as well as mechanisms and mediating factors of predictive language processing. This will set the stage for the hands-on module, during which students will get to know the pros and cons of different approaches to analyzing visual world data. We will then focus on using the 'new statistics' approach, which advocates moving away from null-hypothesis significance testing towards magnitude estimation using 95% confidence intervals and measures of effect size (cf. Cumming, 2014, Psychological Science).

Emmanuel Keuleers
Megastudies as tools for psycholinguistic investigation
The single purpose factorial experiment has long been the staple research method of psycholinguistics. Instead of this single experiment designs, the megastudy approach proposes to collect data on a large number of stimuli without a priori hypothesis. The data can then be infinitely re-used. The data collection effort has been particularly concentrated in the area of visual word recognition, where lexical decision times have been collected for tens of thousands of stimuli in different languages (mostly English, Dutch, and French). Megastudy data from naming and eye-tracking already exist in English, and are becoming available for other languages.  This module will focus on the methods needed to process megastudy data. We will concentrate on three methods. First, we will learn how to set up virtual experiments, allowing us to verify results from the literature by comparing results on the megastudy stimuli with the identical stimuli from existing experiments. In addition, new virtual factorial experiments can be set up using megastudy data to test new hypotheses. Second, we will learn how the wealth of megastudy data can be leveraged using regression designs, which are a statistically more powerful alternative to factorial experiments where continuous predictors are available. Next to statistical significance, the regression approach also emphasises explained variance. Finally, we will show how to implement Monte-Carlo methods, allowing us to make stronger claims about the validity of findings by repeating virtual experiments with small variations.

R. Harald Baayen
Implicit morphology
The standard view of language takes grammar to be a calculus, a formal system comprising an alphabet of elementary symbols, and a set of rules defining how symbols can be combined to form new well-formed symbol sequences.  Each of these rules is in turn paired with a semantic rule, which describes the compositional change in interpretation that comes with changes in form.  For instance, the form rule prefixing un- in English is paired with a semantic rule specifying negation (e.g., happy/unhappy).  
Although many foundational scientists of the 20th century were dissatisfied with the standard view (see, e.g., Turing, 1948; Wittgenstein, 1953; Shannon,1956; Von Neumann, 1958),  the tradition of grammar as logical calculus, initiated by Frege and Russell, has remained dominant in linguistics.
Linguists typically compare different forms and the different meanings of these forms. They then devise rules, schemata, inheritance hierarchies, or neural networks that convert forms into other forms, or meanings into other meanings.
Implicit morphology is a computational theory that rejects the axiom of grammar as a formal calculus.  Implicit morphology goes back to Shannon's theory ofinformation, and starts from the assumption that linguistic form is a signal that interlocutors emit and interpret on the basis of a shared code for "encrypting" and "decrypting" the signal.  Crucially, the experiences communicated through the signal are much richer than the signal, and hence, it makes no sense to assume that experiences would be encoded in a compositionalway.  The encoding and decoding must be discriminative, such that the experience encoded by the speaker can be selected from the listener's rich repertoire of experiences. The first axiom of implicit morphology is that, possibly through a process of unsupervised learning (see, e.g., Kohonen, 1982), minimal units of form come into existence, thanks to exposure to speech input.  The second axiom of implicit morphology is that, thanks to interaction with the world, minimal units of experience come into being as well.  The third axiom of implicit morphology is that the learning of the relation between form and experience is governed by the discriminative learning equations proposed by Rescorla and Wagner (1972).  The fourth and last axiom is that, due to the late maturation of the prefrontal cortex and its communication with the anterior cingulated cortex (O'Hare et al. 2008; Yeung et al., 2004), which begins only in the fourth year of life and continues into early adulthood, the implicit learning system gradually comes to be supplemented (but never replaced) with cognitive skills allowing logical reflection on conflicting alternatives (Ramscar & Gitcho, 2007; Ramscar et al., 2013).
In my first lecture, after introducing the central ideas underlying the theory of implicit morphology, I will discuss the computational implementations for language comprehension, starting with the naive discriminative learner presented in Baayen et al. (2011), and proceeding to ongoing work on comprehension in language acquisition.  In the first lab session, the ndl package for R will be introduced, and participants will be taken through the steps for building an ndl comprehension model and evaluating the model against empirical data.(Bring your laptop.)
In my second lecture, I will first introduce a computational model for speech production, which proceeds from experience to signal without requiring (or allowing) any representations for words, stems, morphemes, or exponents.  Two case studies will be presented, one for Dutch, and one for Estonian, a language with rich inflectional morphology.   I will also discuss the value for understanding experimentally gauged costs of lexical processing of lexical distributional predictors such as frequency, surprisal, and relative entropy.I will present evidence that these predictors present unidimensional, often somewhat distorted, and unavoidably incomplete, perspectives on the knowledge that speakers accumulate over their lifetime.  In the second lab session, I will introduce the R code for the production model, and illustrate its use forsimplified illustrative data sets from English, Latin, Finnish, and Hebrew.

Target audience, eligibility, and evaluation criteria

Participants must be enrolled in a Ph.D. program or be in a post-doctoral position having earned the doctoral degree no more than 5 years before the application deadline.  Students in their last year of Master’s studies may also apply. A very good knowledge of English is required. Prior knowledge or experience with qualitative research in behavioral or biological sciences is necessary. Participants will receive a certificate for their attendance. Participants will be selected on the basis of the strength of their CV, motivation statement and supervisor’s support letter.

Important dates

Applications are closed
Notification of acceptance: September 21
Registration: September 22 – October 10

To apply:

Candidates should email the documents listed below at the following email address:
autumn.langschool [at] unitn.it

1.    Application form downloadable here.
2.    Academic Curriculum vitae (max 1000 words in no more than 2 pages) including any relevant publications or conference presentations;
3.    Motivation letter which specifies in which way this school may benefit their research aims (max 400 words).
4.    Support letter from supervisor (max 400 words)

Participation Fee

The school is financially supported by the University of Trento. The registration fee is 250 Euros and covers 5 hotel nights, as well as lunch for each of the school days. The registration fee is to be paid only after notification of acceptance. The registration fee will be waived for the  strongest applicant.
Payments can be made online with credit card or through a bank transfer.

From Italy:
Banca Popolare di Sondrio - Piazza Centa, 14 - 38122 Trento
IT03Y0569601800000003110X62

From abroad:
Banca Popolare di Sondrio - Piazza Centa, 14 - 38122 Trento
IBAN Code:   IT03Y0569601800000003110X62
BIC Code: POSOIT22XXX

In the “reason of payment” field (or “causale” in Italian) write “Autumn School Methods in Language Comprehension + Last name First Name”.

IMPORTANT: In case of withdrawal, payments will not be refunded.

Organizing Committee
(from the Department of Psychology and Cognitive Sciences and the Center for Mind/Brain Sciences: CIMeC, of the University of Trento):

• Giovanna Egidi
• Uri Hasson
• Remo Job
• Marco Marelli
• Francesco Vespignani

For further information, feel free to contact any of the organizing committee members. Their email is: firstname.lastname [at] unitn.it

Event organized by: