Interactive grammar analysis of historical texts: Adaptive annotation approach to reconstruct the grammatical elaboration of Middle Low German (InterGramm)

我们的研究项目调查了中古低地德语（MLG）从13世纪到书面语言转变（16/17世纪）的（结构）阐述。在这一时期（16/17世纪），随着早期新高地德语（ENHG）的影响越来越大，MLG失去了作为一种超区域书面语言的主导地位。该研究对书面语文语法发展的重建做出了重要贡献，但迄今为止，对书面语文语法发展的研究还只是在一定程度上进行的。由于几个原因，本研究的重点是城市法律法规：我们假设可以研究语言阐述的过程，特别是在法律文本中。他们必须在不依赖上下文信息的情况下解释复杂的（法律）问题，因此必须发展出能够进行这种解释的复杂语言结构。这些法律问题是特别有条件的关系；因此，我们能够检查在调查期间有关条件的语言结构的变化。此外，法律法规是可定位和可追溯的，其结果是，制定过程的发展动态可以在时空上重建。现有的计算语言学和语料库语言学的解析和标注系统需要静态的（先验定义的）语法和语法类别，这不能处理语法的历史动态。为了解决现有历史文本注释工具的一个最核心问题，我们计划开发一种结合机器学习和专家反馈的交互式计算工具。使用基于规则的文本分析技术和机器学习方法来“发现”不断发展的动态语法，从而以一种基于证据的方式重建语言变化，这是一种新颖的方法。由于这以历史语言知识和计算语言学和计算机科学技能为前提，因此该项目需要在整个资助期间各学科的密切合作。我们的实证基础包括13至17世纪可确定日期和可定位的法律文本，建立了一个语料库，该语料库分为以下子语料库：子语料库MLG（主语料库）由1227年至1650年的中低地德语文本组成，包括约120万字。亚语料库ENHG包括在书面语言转变后低地德语地区产生的第一批早期新高地德语文本的选择。这个子语料库包含400,000个单词。我们的目的是验证这样的假设，即这些早期新高地德语文本实例化了中低地德语语法/语法——尽管使用了ENHG词汇。

Our research project investigates the (structural) elaboration of Middle Low German (MLG) from the 13th century to the written language shift (16/17th century). During this period (16/17th century) MLG lost its dominant position as a supraregional written language accompanied with the growing influence of (written) Early New High German (ENHG). The study makes an important contribution to the reconstruction of grammatical developments in written MLG, which are - until now - examined only to some extent. For several reasons the study focuses on urban legal statutes: We assume that processes of language elaboration can be investigated especially in legal texts. They have to construe complex (legal) issues understandable independent of contextual information, so that elaborated linguistic structures capable of such a construal must be developed. These legal issues are particularly conditional relations; consequently, we are able to examine changes concerning the linguistic construction of conditionality during the investigation period. Furthermore, legal statutes are locatable and dateable, with the result that developmental dynamics of elaboration processes can be spatio-temporally reconstructed. Existing parsing and tagging systems of computational and corpus linguistics require static (a priori defined) grammars and grammatical categories, which does not cope with the historical dynamics of grammar. To solve one of the most central problems of existing annotation tools for historical texts, we plan to develop an interactive computational tool that combines machine learning and expert feedback. Using rule-based text analysis techniques and machine learning methods for "discovering" an evolving, dynamic grammar, and thereby reconstructing language change in an evidence-based way, is a novelty. Since this presupposes both historical language knowledge and skills in computational linguistics and computer sciences, the project requires a close cooperation of disciplines over the entire period of funding. Our empirical base consists of dateable and locatable legal texts from the 13th to the 17th century, building a corpus that is divided into the following subcorpora: The subcorpus MLG (main corpus) consists of Middle Low German texts from 1227 to 1650 and includes about 1.2 million words. The subcorpus ENHG consists of a selection of the first Early New High German texts arising in the Low German language area after the written language shift. This subcorpus includes 400,000 words. Our aim is to verify the assumption that these Early New High German texts instantiate - despite an ENHG lexis - a Middle Low German syntax/grammar.