Prelexical language cues in bilingual visual word recognition

Project description

The research project aims at clarifying the role of sublexical language cues in the word recognition process of bilinguals. For example, if the decision is between Norwegian and English, the "ø" in "brød" indicates that the word is Norwegian. In this study, these language cues are used to investigate an issue that currently receives a lot of international attention, namely whether bilingual readers are able to modify their word identification process on the basis of their expectations or the characteristics of the words they read. Currently available studies with respect to this issue are ambiguous to whether this is the case.

Focussing on Norwegian-English bilinguals, one would expect that if the reading process of bilinguals is sensitive to language-specific markers in the input, the presence of a letter such as 'å', which only occurs in Norwegian, might help Norwegian-English bilinguals to speed up their recognition process by limiting their word search to Norwegian words. The presence of bigrams that are normal in Norwegian, but very infrequent in English, such as 'hv' in 'hvit', might have the same effect.
There is some evidence that bilinguals indeed identify words with a language-specific orthographic structure more quickly than words that are language neutral. However, it is not clear at all if the language cues are used to direct search to the language the cues point to. Because Norwegian not only uses language-specific letters, but also the more standard alphabet, it provides an optimal possiblility to test the different theoretical views.

The concrete proposal for the Bergen project was designing and running two experiments; a visual language decision task and a lexical decision task with Norwegian and Englishs words, performed by Norwegian-English late bilinguals. In a language decision task, participants decide to which language a word belongs, whereas in a lexical decision task they decide if a given letter string is a word or not. The main question is if the presence of language-specific letters and/or bigrams influences reaction times and if this effect is caused by modifications to the word recognition process. Moreover, the number of neighbors will be varied to be able to determine if the cues are used to restrict search to one language. Afterwards, simulations of the experiments in the BIA or BIA+ models (Dijkstra & van Heuven, 1998) can provide additional insight.
The BATMULT training site is a good facility for this because, first of all, the Training Site and the AKSIS institute have the required expertise in linguistics required for the word selection. Secondly, the experiment requires English-Norwegian bilinguals which are not easily found outside of Norway. Lastly, the project and the training site share the focus on multilinguality.

Activities

• For English words, the CELEX database (Baayen, Piepenbrock, & van Rijn, 1993) provided almost all the required information for the stimulus selection. For Bokmål such a database did not exist. Data was gathered from various sources (SCARRIE project (De Smedt, 1999), SCARRIE frequency counts (Knut Hofland), Nynorsk Frekvensordbok (Per Vestbøstad), Bigram frequencies (Gisle Andersen)) and combined, using the Python scripting language (http://www.python.org), into a basic lexical database for Bokmål.
• From the lexical databases, three sets of 120 stimuli were selected (for Bokmål, English, and Bokmål derived nonwords). The stimuli varied in their orthographical specificity (language unique, language marked, or language neutral) and cross-language neighborhood size, and were matched on frequency and within-language neighborhood size. Gisle Andersen helped verifying that the selected words were not confusing in any way for Norwegian speakers.
• Three experiments were set up: an English-Bokmål language decision, a Bokmål lexical decision with English words and nonwords, and a Bokmål lexical decision with only nonwords. The items used in the experiments were subsets of the previously selected stimuli. Setting up the experiment furthermore included preparing the experimentation room, writing detailed instructions for the participants, creating a questionnaire on their knowledge of Norwegian and English, and reqruiting the participants. AKSIS was very helpful in providing financial compensation for the participants, which made recruiting much easier.
• The three experiments were run (20 people participated in each experiment) resulting in reaction time, accuracy data, and the filled out questionnaires.
• The results were analyzed using SPSS.
• The progress was presented at the MULTILINGUA/BATMULT seminar and the AKSIS Language Technology seminar.
• Chapters on stimulus selection and the experiments were written as parts of the thesis.

The obtained skills and knowledge are best described as experience in designing, running, and analyzing experiments. I was able to go through the entire process from general idea to analysis. Two important specific things I have learned are how to make decisions when selecting stimuli and how to pick the method for analysis. On the side, Python was added to the collection of programming languages I have used.
Another great experience was the opportunity to be in an environment with many people, including a relatively large number of Marie Curie students, many of which had a different background. This provided an inspiring environment and many ideas for further research and links to other areas to explore.

List of results

1. Bokmål Lexical database: A database of all 3 to 5 letter Bokmål lemmas was created including information on the frequency of occurence (in Bokmål and Nynorsk), syntactical category, homograph status (in Nynorsk and English), number of orthographical neighbors (in Bokmål, Nynorsk, and English), and bigram frequencies (in Bokmål, Nynorsk, and English).
2. Stimuli: Three sets of 120 stimuli were selected (for Bokmål, English, and Bokmål derived nonwords). The stimuli varied in their orthographical specificity (unique, marked, or neutral) and cross-language neighborhood size, and were matched on frequency and within-language neighborhood size. The lexical decision experiments all used a subset of these stimuli. Other subsets can be used for future experiments. Furthermore, a different task (for example progressive demasking) or population (for example English or Norwegian monolinguals) could be used.
3. Experimental data:
The three experiments performed resulted in reaction times and error rates of the participants. Furthermore, each participant filled out a questionnaire on their use of Norwegian and English.
4. Analysis: An analysis of the experimental data was conducted, showing which of the factors (orthographical specificity and cross-language neighborhood size) significantly influenced the participants' performance.
5. Parts of the thesis concerning the finished activities: Chapters of the thesis dealing with the stimulus selection and the experiments have been written.
6. Presentation at the MULTILINGUA/BATMULT seminar in Mjølfell (October 7-9, 2005)
7. Presentation at the AKSIS Language Technology seminar (December 16, 2005)
8. Ongoing work: Discussion has started on the interpretation of the experimental results and the writing of a scientific paper

References