Mobile Edition
Print
Get the Mag
Weekly UpdatesUsing computers to help students practice and learn grammatical constructions goes back to the earliest days of computer-assisted language learning (CALL). With the coming of the Internet age, CALL began to focus more heavily on the new capabilities of group connectivity and computer-mediated communication. More recently, a gathering consensus has emerged that for adult learners, at least, an awareness of forms and rules is a vital component of online language learning (Skehan, 2003). Compared to the nature of earlier grammar-oriented applications, however, there is recognition today that a focus on form should not be an isolated, stand-alone activity but rather should be integrated into a communication-centered, networked language learning environment. Contemporaneously, it has become clear that grammar exercises need to require more than single word or phrase answers. The older exercise formats, such as multiple choice and fill in the blanks, should be supplemented by new and engaging interactions with real communicative goals. Something more than just canned feedback should accompany the exercises. Feedback should be informative, contextual, and, whenever possible, individualized. The expectation today is that programs will guide students to pay attention to forms and structures, to gain facility in their use, and to extend their language knowledge/usage to more complex constructions. Stated succinctly, grammar exercises need to be integrated, intelligent, and innovative.
Fortunately, technology developments and creative language professionals are moving us down this path. Recent trends in intelligent language tutors (ILT) are quite promising, despite the multiple challenges of natural language processing (NLP). Having over-promised and underachieved in the past, developers of ILT (also known as ICALL, intelligent CALL, or parser-based CALL) have mostly narrowed their ambitions and scope, resulting in actually deployed systems rather than just research prototypes. Advances in collecting and processing language corpora have helped in that process. At the same time, Web developments offer new approaches to exercise design and distribution.
INTEGRATING GRAMMAR INTO TASK-BASED ACTIVITIES
Much of the language instruction done today, either in the classroom or on the computer, revolves around specific tasks learners must carry out, either individually or in groups. Tasks lead learners to focus on communicating within concrete, realistic contexts, negotiating meaning towards reaching a specific goal, and gaining confidence in using the target language to achieve results. On a computer, that process can begin in the task planning stage. Building on the idea of pre-task vocabulary brainstorming, a possible partner activity could be a brief discussion of any constructions that learners anticipate using in the task, such as the subjunctive for hypothetical situations in Spanish. This would likely be most applicable to intermediate/advanced level students. The partner interaction could be done through a chat window integrated into the exercise environment, as suggested by Shaoqun Wu and Ian Witten (Wu & Witten, 2006). This provides the possibility of groups sharing ideas with one another. Integrating collaborative tools makes grammar exercises less isolating and more communicative. The kind of pre-task help suggested here is common in writing assignments (suggested sentence starters, possible sentence connectors, etc.), but is less frequent in the context of grammar exercises. Having some discussion of lexico-grammatical issues before the task begins may alert learners of their importance and possibly result in the use of more complex language.
Post-task activities can also be integrated into a task to encourage focus on form. If students know, for example, that a product of their efforts will be shared with others (through a forum, blog, or wiki), they are likely to be more deliberate in their use of language. The notion of posting students' work could also be extended from individual efforts to class exchanges or school collaborations. A necessary condition for greater focus on form is a level of comfort on the part of the students with the task content. If the topic is too unfamiliar or challenging, it is less likely that learners will focus on form. It is also unlikely in that case that they will consider using more complex language constructions. Whether it is a task to be evaluated by the teacher or processed by a computer, there needs to be recognition of efforts by students to stretch their skills beyond familiar collocations and simple sentence formations. This is one of the most challenging tasks for software, even if some form of artificial intelligence is used. It is easy to check for expected and typical responses; dealing with creativity is much harder.
Post-task activities provide opportunities for students to expand and internalize what they have learned. There should be a record kept of task results and new information learned, either by the learner or automatically by the program. Note should be taken of remediation that needs to be done. Ideally, this information should be used to help determine the next steps in the student's language learning. As we shall see, this is a crucial aspect of best practices in the design of ILT's. Follow-ups to assigned tasks can have students write a short summary of what they learned (shared electronically) or construct sentences or dialogs using the practiced/learned structures and vocabulary in different contexts. Another follow-up could be a game or contest in which correct and rapid use of practiced forms is required. An additional post-task exercise could be the consultation of a concordance in which learners find and analyze occurrences of phrases or structures. This brings into play the use of language corpora, another important element in the creation of intelligent language tutors.
INTELLIGENT LANGUAGE TUTORS
For language learners to be successful, tasks should not be disconnected activities but integrated into a learning environment that tracks and analyzes their progress, offers help as needed, suggests and provides logical next steps, and gives appropriate feedback. An intelligent computer program, like a good teacher, should be able to treat students as distinct individuals, provide learner-specific guidance, and design a customized learning path. This is a tall order, and not just from a technical standpoint, as knowledge is required in fields such as learning psychology, discipline-specific pedagogy, and logic. This is on top of expertise in software engineering and human/computer interaction. To create an advanced language learning application, one needs to add to the mix of expertise listed above knowledge in areas such as second language acquisition, computational linguistics, formal grammar theory, and natural language processing. Creating an ILT is not a project to be undertaken lightly, and it is not surprising that, even with teams of experts, many projects never reach production status. Marina Dodigovic's recent monograph (2005) offers a glimpse into the many considerations and efforts that go into just the preparatory stage of designing an ILT. Similarly, Markus Dickinson and Joshua Herring (Ohio State University) show how complex the process can be, even in a restricted task such as analyzing Russian verb conjugation.
The ILT projects most likely to see the light of day are those that restrict the learner input to be processed and evaluated. This restriction might be through the exercise type, the length/complexity of the learner input, or aspects of the input which are excluded from analysis. For many areas of discrete grammar knowledge, little machine intelligence is required. To determine, for example, whether a German adjective ending is correct, it suffices to compare the student response to the known (single) correct answer. This kind of string matching is easy and fast but only provides responses of right or wrong to the learner. "Intelligence" lies in enabling the computer program to provide more nuanced and useful feedback to the learner. For German adjective endings, the system would need to know about variations based on gender, number, case, and context, and be able to evaluate a student error based on that information. Even such a small slice of language involves a large number of variables. Consider how much more complex a challenge it is to parse and evaluate even a short sentence. The system needs to analyze why a given response is incorrect in the given context, then provide options and help for learners to improve their response. It is little wonder that in the past ILT's were plagued by false positives (correct input marked incorrect) and missed incorrect constructions.
In recent years, a number of ILT's have become available which represent considerable advances over earlier systems in a number of ways, including processing responsiveness, evaluative accuracy, and meaningful feedback. Among recent ILT projects which have attracted generally positive receptions are E-Tutor (formally German Tutor) from Trude Heift, Robo-Sensei (formerly Banzai, for learning Japanese) from Noriko Nagata, Spanish for Business Professionals from L. Kirk Hagen, TAGARELA (for beginning Portuguese) from Luiz Amaral and Detmar Meurers, and imPRESSions from Joan-Tomas Pujola (for understanding English language media). Taken together as the current state of art in ILT's, there are a few best practices that can be gathered from these systems:
Reusability. The design of the project should whenever possible enable (and encourage) re-use. There is so much effort and expense involved in creating an ILT that building on previous projects makes more sense here than in any other area of CALL. One design factor which aids reuse is modularity. This involves separating the processing of student input into separate programs or modules which can be run sequentially (as in E-Tutor) or selectively on demand (as in TAGARELA). This approach to programming has allowed for creation of programs like Boltun, re-using TAGARELA for Russian, and a Greek ILT building on E-Tutor. The modular design of E-Tutor allows particular parts of the system, such as the punctuation checker, to be turned on or off as desired. This also allows updating of only one section of the system at a time, or adding custom modules in the future. It also builds in the option of sharing only particular modules. Of course, grammar models and analysis processes will vary according to the target language, which makes some parts of the system non-transferrable. There may be as well technical hurdles to moving modules from one system to another. These are typically complex systems, which may use a variety of languages, most often Prolog and Lisp for NLP and Java, Perl, or Python for Web delivery. This makes it important that possible compatibility be built in to a new system from the beginning. The ultimate re-use of these systems would be the ability for non-programmers to add or edit content. In that sense, the evolution of ILT's seems to point towards the need for creation of user-friendly authoring interfaces.
FEED