• No results found

Sentence parsing is the process of assigning a grammatical structure to an input sen- tence, as specified by a declarative grammar. As Heift and Schulze (2007) explain, sentence parsing is one of the NLP techniques which have been used in developing CALL systems. Heift and Schulze define a CALL system that incorporates a grammar and a sentence parser as a parser-based CALL system. Parser-based CALL systems may or may not be capable of handling dialogues. Among the four dialogue-based CALL systems discussed earlier in §2.3, only the Kaitito system is considered as a parser-based CALL system. Although L2tutor has a parser to detect erroneous utter- ances, the parser is only activated after a dialogue session with learners is completed. In the following, I describe four existing parser-based CALL systems: German Tutor, ICICLE, Arabic ICALL, and BANZAI.

2.4.1

German Tutor

German Tutor (Heift and Nicholson, 2001) is an Intelligent Language Tutoring Sys- tem (ILTS) for learning German language.7 The system is developed to help students

practise German grammar. The German grammar formalism is represented in HPSG, similar to Kaitito. When a learner enters an input utterance, it is parsed, and the result- ing grammatical representations are used to perform a series of checking mechanisms relating to missing word detection, word order and grammatical errors. If an error is found, then the system provides an error feedback utterance based on the learner’s level of language proficiency.

A learner’s language proficiency level is recorded in the learner’s profile. The profile is stored in a student model.8 The proficiency level is either beginner, intermediate

or advanced. In order to respond to an erroneous sentence, a detailed error feedback which consists of an exact error location and the type of error, is provided to a beginner learner. For an intermediate level, only the type of error is given in the error feedback.

7ILTS and parser-based CALL systems are two similar things as mentioned in Heift and Schulze

(2007, pg. 2).

For an advanced learner, the error feedback is a hint as to where an error is located. As an example in Heift and McFetridge (1999, pg. 60), a learner enters an ill-formed sentence as in (7) below,

(7) Sie tr¨aumt von einen Urlaub. (She is dreaming of a vacation.)

One of the following error feedback is provided based on the learner’s proficiency level. 1. There is a mistake with the article einen of the prepositional phrase.

2. There is a mistake in case with the article einen of the prepositional phrase. 3. This is not the correct case for the article einen of the prepositional phrase. Von

assigns the dative case.

The feedback (1) is the most general and provided to expert learners. It provides a hint where an error is located (prepositional phrase) in the sentence. Intermediate learners are provided with the feedback (2) in which the feedback is more detailed than (1), providing additional information on the type of error (case). The most detailed feedback is the feedback (3), provided to beginner learners. The feedback response does not only pinpoint the location of and the type of error but also refers to the exact location of the error (dative preposition).

2.4.2

ICICLE

The name ICICLE is an acronym for “Interactive Computer Identification and Correc- tion of Language Errors”. It is a parser-based CALL system that helps deaf students with grammatical components of their written English (Michaud, 2002). ICICLE also employs linguistic features such as detecting errors on sentences and generating feed- back on the errors. ICICLE consists of two modules: the identification module and the response generation module (Michaud, McCoy, and Pennington, 2000). The identi- fication module analyses each sentence, and if errors occur, the response generation module generates error feedback utterances.

The system’s work begins when a learner’s written text is sent to ICICLE, either by directly typing into the system or by loading a text file. The text is analysed for its

grammatical structure. During the analyses, a chart parser is used to parse the sen- tence. There are two kinds of English grammar referred to during the parsing process. First is the grammar that represents a correct structure of English sentences. This grammar is used for analysing a grammatical input. As for analysing the grammati- cal errors, the second grammar that consists of structure of ungrammatical sentences, known as an error grammar, is used. More about error grammars is explained in §2.7.1.1. In order to address any errors occurred, the response generation module pro- vides feedback pertaining to the error. The feedback is generated from a list of canned explanations for the error which occurred.

2.4.3

Arabic ICALL

Arabic ICALL is also another example of a parser-based CALL system (Shaalan, 2005). The word ICALL is an abbreviation of “Intelligent Computer Assisted Language Learn- ing”. The word “intelligent” is used because of the use of NLP features on CALL. Obviously the system is developed for learning Arabic.

Arabic ICALL consists of four components: a user interface, course materials, a sentence analyser and feedback provider. The user interface acts as a medium of com- munication between users and the system. The course material consists of teaching content, which includes a database of test questions, a tool to generate test scripts, and a tool to maintain lessons and test items.

The sentence analyser has a morphological analyser, a parser, grammar rules and a lexical database. The Arabic grammar is written in a definite-clause grammar (DCG) formalism. The analyser parses a learner’s sentence with the grammar. In order to handle an ungrammatical answer, similar to ICICLE, Arabic ICALL is also equipped by an Arabic error grammar. Then, the analyser passes the analysis form of the learner’s answer to feedback provider.

For each question given to a learner, its corresponding answer is provided to Arabic ICALL. The answer is also analysed by the sentence analyser in order to produce an analysis form of the answer. The feedback provider compares the analysis of learner’s answers with the analysis of the correct answer. A positive message will be issued to

the learner if it matches. Otherwise, error feedback is given based on information in the Arabic error grammar. Refer to Shaalan (2005, pg. 100–105) for further information about how learner’s ill-formed answers are handled.

2.4.4

BANZAI

BANZAI is a parser-based CALL system that helps learners to learn Japanese (Na- gata, 1997). BANZAI employs a simple parser based on a finite state grammar (Gazdar and Mellish, 1989). The system is fully implemented in Lisp and described in Nagata (1997). BANZAI parses not only grammtical sentences but also a variety of erroneous sentences. The target errors are Japanese particles. When the system detects that a sentence contains errors, BANZAI diagnoses what types of errors have occurred and provides feedback. Two types of feedback are provided: deductive and inductive feed- back. Besides a parsing facility, BANZAI provides a simple authoring tool which helps a teacher to provide exercises and their corresponding answers. Deductive feedback utterances give detailed linguistic information on particle errors and besides, the cor- rect particle is also given. Inductive feedback is deductive feedback with the addition of two or three examples of correct sentences that use the particles.