Abstract:

Wouldn’t it be nice to supplement your writing lesson with your elementary school class with some sentence combining or lexical choice tasks done by each pupil individually in an e-learning system in her/his own speed—presupposing PCs, tablets or cell phones are available in the classroom or at home? First of all, the e-learning system helps with oral support (via headset) in case of reading difficulties. Moreover, the system gives exact, personalized feedback to any construction by a pupil. Based on the e-learning experiences, the class decides on the preferred improvements for the original (poor) story as a result of a so-called writing conference (often used in German class rooms).

In our talk (+demo), we describe how simple the dialogue of the teacher mode is to set up a new story for the Sentence Fairy system. The Sentence Fairy System implements a “virtual writing conference” for elementary-school children, with German as the target language. It deploys natural-language generation technology to evaluate and improve the grammatical quality of learner output. Based on an abstract representation of the story under construction, all paraphrases of simple and combined clauses are generated fully automatically. From this source, the system produces exercises enabling the learners to improve their sentences. Up to now, the abstract representation to run the Sentence Fairy had to be written by skilled linguists. By way of the teacher mode, users with elementary linguistic knowledge about constituents of a simple clause can enter their own stories to be translated into e-learning exercises fully automatically.

In order to avoid any ambiguity in clauses, which would spoil the automatic exercise generation—especially, exact feedback generation—by the Sentence Fairy system, the teacher mode restricts natural language parsing to constituents instead of whole clauses. The system provides a drop-down menu with available grammatical functions of a chosen verb, i.e, subject and direct object. All grammatical functions provide mouse-over tips showing definitions for the linguistic terms to avoid confusion about names such as accusative object vs. direct object. Some additional information has to be given as answers to dedicated questions—e.g., gender of new proper nouns and co-referentiality of identical objects all over the story to create pronominalization and relative clause exercises properly.

Information on rhetorical relations, e.g. temporal or causal relations, to become the basis for a variety of sentence combination exercises can be specified by setting marks in a system-generated table providing all sentence pairs. Alternatively, a subordinate conjunction can be typed in. The system translates the choice into the abstract relation in Rhetorical Structure Theory (RST); e.g. “Peter combed his hair before Peter left the house” will become EARLIER(a,b) and can result in “Peter left the house after he combed his hair”.

Moreover, the teacher mode allows the specification of lexical choice alternatives (e.g. run, hasten, hustle). The system automatically generates a suggestion list by selecting all synonyms, hyponyms and hyperonyms in WORDNET for all words of the story. As the German variant requires the translation of the German terms to English WORDNET and back again, the teacher can erase wrong translations for unintended meanings of the German term as well as (s)he can add further alternatives not provided in the suggestion list.

Keywords:

e-learning, virtual writing conference, Sentence combining, lexical choice.