O n f e r e n c e

Short paper 
I explo
re how VanPatten’s (2004) principles of input processing can be implemented in 
ICALL environments to help select optimal sentences for textual enhancement or 
automatic exercise generation. Natural Language Processing technology has been 
implemented in CALL environments to help learners to study authentic texts, including 
WERTi/VIEW (Meurers et al. 2010), the REAP Project (e.g. Collins-Thompson & Callan 
2004), and ATA/Text Adaptor (Burstein et al. 2007). For example, Meurers et al. (2010) 
developed a webapp and FireFox extension that allows a user to automatically embed 
grammar exercises within texts found on the Internet (English, German, or Spanish). 
Authentic text ICALL systems sometimes encounter texts that contain many tokens of the 
target language structure (e.g. the grammar principle under focus).

---

-293- 
2014 CALL Conference 
LINGUAPOLIS
www.antwerpcall.be 
If the ICALL application enhances too many tokens, it can be overwhelming for the 
learner. Previously, this problem has almost always been either ignored by enhancing all of 
the tokens in the text, or by enhancing only a randomly selected number of tokens. The 
present research project attempts to identify pedagogically well-motivated methods for 
selecting optimal tokens to bring to the learner’s attention. I refer to this task as optimal 
token identification (OTI). 
Introduction 
In existing research, OTI has been conceptualized as a task of matching textual properties 
with learner abilities and goals. Regarding the learner’s goals, a text should obviously be in 
the learner’s target language, and should contain 
tokens of the target language structure. 
Resolving these questions is usually trivial for both humans and computers. The remaining 
task of matching text readability to learners’ abilities has proven to be much more difficult, 
and research connected to OTI has mostly been focused on this part of the task. 
Readability analysis can trace its roots back to 1893 (DuBay 2006), and until recently, it 
was dominated by formulas based on easily countable features, such as syllables per word, 
or words per sentence. However, research in computational linguistics has made it possible 
to automatically extract more complicated features from a text, both lexical and structural. 
At the same time, the recent rise in machine learning techniques has made it possible to 
build m
odels to estimate text readability. In spite of this recently renewed interest in 
readability, only a small number of studies have looked at readability with any relation to 
second language learning (Vajjala & Meurers 2012, Pil
n et al. 2013, François & Fairon 
2012). 
While conceptualizing OTI as a readability categorization task appears to be a promising 
approach, it is strange to think that OTI could be completely successful in a second 
language learning domain without regard for the unique processing inherent to second 
language acquisition. For example, the Read-X/Toreador project (Miltsakaki & Troutt 2008) 
is an application that aims to improve L1 English literacy by evaluating expected reading 
difficulty of online texts and matching it with students’ projected reading abilities and 
interests. This approach – designed for L1 readers – is effectually identical to the 
approaches of Pil
n et al. 2013 and François & Fairon 2012, which were designed for L2 
readers. However, since L1 literacy acquisition and second language acquisition are 
qualitatively different processes, we should expect approaches to OTI to reflect that 
difference.
6
The present study aims to address this difference. 
Readability analysis has negative predictive value for learner uptake, i.e. it rules out 
sentences or entire texts that do not lead to uptake. If the meaning of a text is beyond a 
learner’s ability, then the potential for learning is severely limited. By ruling out unsuitable 
tokens, readability analysis identifies which sentences are viable or workable targets for 
learner uptake. It does not, however, identify which tokens are 
optimal
for learner uptake. 
This next step for OTI is a pedagogical question, one that has already received significant 
attention from second language acquisition researchers. In the following sections, I show 
how the results of one particular vein of second language acquisition research can be 
applied to the OTI task.