Child Language & Eyetracking: Analyses and Rationale
Exchange forum for researchers working with eye tracking. See Aim of CLEAR below.
9:30 – 16:00
CLEAR MenuProgramme Registration Venue Aim of CLEAR Contact
There will be several presentations on a range of topics and we plan to have plenty of time for questions and discussions.
A cafeteria is located right next to the conference centre. Four different dishes (one vegetarian) are served for lunch at prices between € 3,50 and € 5. Some hot and cold beverages will be provided during CLEAR.
Click on the coloured presentations to view the abstract.
09.30h Tom Fritzsche: Welcome & Introduction
Jacolien van Rij-Tange (University of Groningen, The Netherlands)
The pros & cons of using eye-tracking glasses for linguistic research
Eye tracking glasses (ETG) are a relatively new type of eye tracker that incorporate an eye tracker in the frame of glasses. This type of eye tracker is designed to be used in real world situations and is not limited to experimental computer tasks. This is very useful for testing children, and for running more naturalistic experiments outside the lab. However, the eye tracking glasses yield different data than the conventional eye trackers. As a result, we often cannot use the standard methods for preprocessing and analysis.
In this presentation, I will first introduce the ETG and it's implications for experimental design, measures, preprocessing, and analyses. In addition, I will share about our experiences with the ETG, discuss some issues and preliminary solutions. (Abby Toth's talk – next – will discuss the results of an actual ETG experiment where some of these issues were present.)
Abigail Toth (University of Groningen, The Netherlands)
What do visual world eye movements tell us in a continuous discourse?
Since its development, the visual world eye-tracking paradigm (VWP; Cooper, 1974; Tanenhaus, Spivey-Knowlton, Eberhard, & Sedivy, 1995) has been used to investigate the online processing of various linguistic phenomena, including the online processing of reference (e.g. online pronoun resolution). Typically in these studies, participants view scenes with two (or more) referents while listening to passages that contain an ambiguous pronoun (e.g., Arnold, Eisenband, Brown- Schmidt, & Trueswell, 2000). Whichever referent participants fixate on following the pronoun onset is assumed to be the referent that they take the pronoun as referring to. However, the experimental items used in these studies were usually no more than 2-4 sentences long and thus lacked any sort of elaborate discourse context, which in turn may not accurately reflect naturalistic language processing. Very few studies have applied the VWP in the context of a naturalistic continuous discourse.
In a preliminary study (Toth, 2018; Toth, Charest, van Rij, & Järvikivi, 2018), we investigated the online processing of reference in a naturalistic language setting. Children and adults listened to a 5-minute long (electronic) storybook wearing mobile eye-tracking glasses. The gaze data were analyzed relative to the onset of referring expressions (both full noun phrases and pronouns) that were mentioned throughout the story. In line with previous studies, it was found that following the mention of a referring expression there was an increase in the proportion of looks to the intended referent for both children and adults. However, this effect was only present early on in the story. As the story progressed, the likelihood that participants directed their eye gaze towards the intended referent decreased. These findings suggest that visual world eye movements are largely influenced by an unfolding discourse and that there may not be one to one mapping between linguistic input and corresponding eye movements.
Currently, we are running a more controlled experiment to further investigate the influence of a continuous narration on eye movements. In this follow-up study, participants hear the same experimental items either in the context of a continuous narrative or completely randomized without a narrative context. We will present the findings from both of studies, which will provide further insight into the mapping linguistic input and corresponding eye movements.
Iyad Aldaqre (SR LABS, Milan, Italy)
The use of relative eye tracking measures
Researchers have been using eye tracking in developmental psychology research for almost half a century (e.g. Brickner, 1970). Since then, eye tracking data has been analyzed in different ways, mainly focusing on categorizing different types of eye movements - or the lack thereof - into saccades or fixations (Duchowski, 2007).
In order to identify these eye movements, researchers use algorithms, or filters, which vary in complexity and might have many parameters to set in an arbitrary fashion (Salvucci & Goldberg, 2000). This procedure might lead to diverging results, even when manipulated for the same dataset, posing a threat to data replication.
Attempts have been made to establish guidelines for employing and reporting eye tracking methods (Oaks, 2010; Gredebäck et al 2010). Such attempts surely help clarifying technical aspects, like data processing, especially to new researchers. Here I am trying to contribute to these efforts by focusing on the nature of measures used when dealing with eye tracking data, specifically whether and when to use relative instead of absolute measures.
The use of relative measures is common practice in psychology (e.g. percentages and difference scores) and it can help researchers control for technical discrepancies between one study and another. For example, when activating the option to group nearby fixations in filter settings, one might end up with fewer, longer fixations, possibly leading to different results. However, when relative measures are used, results would be resistant to such differences.
In this talk, different eye tracking measures will be shown from a word learning paradigm (Aldaqre et al., 2015) and a spatial cueing paradigm (Aldaqre et al., 2016). A comparison between absolute and relative measures will be presented from data collected for these studies.
Sudha Arunachalam (New York University, USA)
Eye tracking of children's online comprehension of their own parent's speech
To combine elements of naturalistic studies of parent input with experimental studies of language processing, I've recently been using eye tracking to study children's real-time language comprehension of their own parent's unscripted speech. The parent and child play a game together involving finding pictures on the screen, or the parent tells a story following a picture book displayed on the screen, and the child's eye gaze is evaluated as a measure of their comprehension. This method allows us to study how closely parents' language is tailored (or not) to their own child's language comprehension abilities.
Antonija Blaži, Ana Matić & Jelena Kuvač Kraljević (University of Zagreb, Croatia)
Methodological issues in studying online processing of anaphora in LI children
Reading comprehension can be defined as complex process of integration of lexical features (word forms and meanings), word recognition skills, reading fluency, syntactic processing, inference generation, comprehension monitoring, etc. (Cain & Oakhill, 2006; Perfetti, 2007; Language and Reading Research Consortium & Logan, 2017). To build a meaningful representation of a written word, sentence or a text, component skills of lower- and higher-level language processing must be activated. It is commonly thought that these processing components are defined depending on the level of processing on which they are activated during text comprehension: word, sentence or a discourse level (Daneman, 1996; Perfetti, 2001; Rayner & Reichle, 2010). Having in mind this complex nature of language comprehension, it is not surprising that there is a population of children with language impairment (LI) who are struggling with extracting information given in the written form, integrating it with previous knowledge and understanding what they read (Yuill & Oakhill, 1991; Oakhill, 1994; Bishop & Snowling, 2004). With respect to the previous studies that provided data about individual differences between skilled and less skilled comprehenders, there is still a need for specifying the exact source of these differences.
Naturally, text paragraphs or larger discourse units are reduced of repetition in order to increase text cohesion. Therefore, understanding sentences commonly depends on interpretation of the previous ones. Anaphora is a major linguistic tool used to maintain cohesion within and between sentences (Ehrlich & Remond, 1997) and processing such complex, yet common syntactic structures may be demanding. By studying the processes that underlie resolving anaphoric structures, researchers could gain insights into the interactions of various anaphoric markers and comprehension skills, and disentangle the potential source of specific processing difficulties.
Most researchers in the field of written text comprehension processes have relied mostly on off-line experimental methods. However, they do not give us specific data about when the comprehension failure happens and what is in its background. On the other hand, on-line experimental methods, specifically eye-tracking, stand out as objective, naturalistic, highly informative methods that provide valuable and reliable data about multi-level comprehension processes (Staub & Rayner, 2007). As is the case with any experimental method, the use of eye-tracking and the gathered eye movement data open up many questions related to the conceptual, empirical and practical issues in this research field (Rayner & Carroll, 1984; Rayner et al., 2006). In order to use eye tracking with the purpose of gathering reliable and valid data about specific difficulties in anaphora processing in children with LI, several methodological issues have to be discussed. These will be briefly presented during CLEAR.
Alan Langus (University of Potsdam, Germany)
Frequency tagging and entrainment to auditory rhythm in pupillary data
Regularly occurring stimuli are known to elicit a brain response that closely follows the temporal structure of the rhythmic stimuli. Whether these responses are purely stimulus-driven or also reflect entrainment to rhythm, is the topic of considerable research. Recent advances in eye-tracking suggest that also the human pupil may rapidly change in size in synchrony to rhythmic stimuli, providing a simpler and less time-consuming method for studying rhythm perception than electrophysiological methods. This talk will show how a simple transformation of pupillary data from the time domain to the frequency domain – e.g., frequency tagging – can be used to measure this response by simply looking at the amplitude of the pupillary changes at rhythm frequency. How to use this frequency response in pupillary data to study entrainment to rhythm will be discussed.
Samuel Forbes (University of East Anglia, UK)
PupillometryR – An R package for preparing and analysing pupillometry data
In recent years, the analysis of pupil size data has emerged as a powerful alternative to traditional eye-tracking methods. Pupillometry gives a direct window into the cognitive state of the participant, but pre-processing, cleaning, visualising, and analysing the data can be complex. In R, a popular statistical language, there are few established pipelines for handling pupil data, although there are a small number of options available to users familiar with other languages. PupillometryR allows the user to follow an integrated pipeline from setting up and cleaning the data, to preprocessing, analysing, and even visualising results. The package is also designed to keep pupillometry as familiar as possible for the regular R user, handling data in formats and functions that will be accessible for most users. In this presentation, I will demonstrate analysis of pupil data using the PupillometryR pipeline, and introduce the basic concepts researchers need to analyse pupil data.
16.00h Closing & Goodbye
Registration for CLEAR is independent from the WILD registration. Participation at CLEAR is free of charge.
Registration & survey form
Please go to the registration desk that day.
CHANGED! CLEAR will take place at the Fraunhofer Conference Center in Potsdam-Golm.
The new CLEAR venue is outside of the city centre in the Science Park Potsdam-Golm.
The best way to reach the Conference center is by local train or bus to »Potsdam, Golm Bhf«.
- Local train lines RB20, RB21, RB22 (about 10 min from Potsdam Central Station)
- Bus 605 or 606 (25 min from Central Station)
From the station walk westwards straight across the roundabout (see orange marked route on the map below), along the little pond to the conference centre (black building). There will be signs guiding the way from the station. The walk takes about 5 min.
Start location: e.g. »Potsdam Hauptbahnhof«
Destination: »Potsdam, Golm Bhf«
Fare (buses and local trains): Potsdam AB Single (2.10 €), Potsdam AB 4-trip ticket (7.60 €)
Am Mühlenberg 12
Aim of CLEAR
This one-day workshop is intended to be an exchange forum for people (beginner, advanced and expert) doing research on linguistic abilities in children using the eye-tracking methodology.
Previous CLEAR workshops have been held in Tours (2009), Groningen (2010), Potsdam (2011), Nijmegen (2012) and Potsdam (2016).
With eye tracking being such a versatile instrument and given the constraints that we face in testing infants and children (with or without SLI, ASD, …) there are a lot of decisions that have to made before, while, and after running a study. The aim of this workshop is to discuss these methodological issues pertaining to:
- Optimal timing, repetition of items
- Fillers/attention getters/pauses
- Influences of instructions and/or additional tasks
- Technical issues
- High- and low-cost equipment: Tobii/SMI/EyeLink/GazePoint
- Mobile and stationary settings
- 3rd party presentation software (PyGaze, Matlab, E-Prime, etc.)
- Proportion of looks
- Saccades, first looks, etc.
- Units of analysis (raw data, fixations)
- Inclusion/exclusion criteria & missing data
- Influence of the size of analysis windows (time intervals) on results
- Statistical procedures: ANOVA, permutation analysis, mixed models, growth-curve analysis, GAMMs, etc.
- Statistical power/sample size
Thus, the focus of the meeting is less on the what (i.e. research questions and results) and more (but not solely) on the how.
CLEAR 2019 is organised by Tom Fritzsche
Stay connected beyond CLEAR!
Join the Eye-Tracking Mailing List at Google Groups.
University of Potsdam
Linguistics / H14 / R140
Email, Phone & Fax
Phone: +49 331 977 2296
Fax: +49 331 977 2095