Specialised POS Tagged Syriac Corpus for State Morphology

Overview

A total of twelve .TXT files each representing a Syriac text that has been transcribed and tagged for part-of-speech (POS). This corpus forms part of a PhD research project on the historical syntax of Aramaic (Syriac) at The Australian National University (2020—current) in Canberra, Australia. This research project is interested in noun state morphology, among other topics, which is reflected in the POS scheme for this corpus.

Method

A detailed summary of this methodology is provided in El-Khaissi (data paper in review with the Journal of Open Data Humanities).

Transcriptions are sourced from Digital Syriac Corpus.

POS tags are based on word matches using SEDRA IV API (v1.0.0).

Selection of Syriac texts was optimised to minimise external influence on Syriac grammar and maximise full coverage of key periods of the Syriac language from 2nd—13th century AD.

POS Format & Abbreviations 

POS tags in the text files follow the following format:

<syntax-category>-<state>_<syriac_word>

Thus, an underscore '_' marks the beginning of a tag sequence while tag values are separated by hyphen(s) '-'. For example (noting text directionality constraints):

ܒܘܪܟܬܐ_EMP-N

The following abbreviation lists the definition of all POS tags, which are based on the parameters available in SEDRA IV API (v1.0.0).

Absolute state noun (indeterminate relic)
ABS

Emphatic state noun (new indeterminate)
EMP

Construct state noun (bound noun)
CNS

State not applicable
X

particle
PTCL

pronoun
PRO

preposition
PREP

verb
V

denominative
DEN

noun
N

numeral
NUM

substantive
SBV

adjective
ADJ

proper noun
PN

adverb
ADV

demonym
DNM

participle adjective
PTCPADJ

adverb
ADV

idiom
IDM

See Quality Control & Limitations below
DUP

Quality Control & Limitations

On average per manuscript, the POS-tagging process achieved a 63.13% saturation of texts. The POS tagging process was based on an exact-match process, which does not take into account syntactic or semantic context. Syriac words which exhibit homonymy are thus tagged with the value 'DUP' and should be assessed manually based on its original context. Among all 297,981 words in the corpus with an available POS tag, approximately 73,188 (24.56%) of tags reflected some kind of homonymy involving a word with various semantic and/or syntactic interpretations.

Since this dataset was created as part of a research project investigating noun state morphology, additional tags were created targetting various state values. Grammatical elements, like number and gender, were not required as part of this investigation and therefore excluded from the POS-tagging process.

Contact

For any questions, please contact Charbel El-Khaissi <[e-mail]>.

概况

本数据集包含12个.TXT格式文件，每个文件对应一篇已完成转录并标注词性（Part-of-Speech，简称POS）的叙利亚语文本。该语料库为澳大利亚堪培拉澳大利亚国立大学一项针对阿拉姆语（叙利亚语）历史句法的博士研究项目（2020年至今）的组成部分。本研究项目聚焦名词状态形态等议题，这一点也体现在本次语料库的POS标注体系中。

标注方法

该方法的详细概述详见El-Khaissi的研究论文（目前正在《开放数据人文期刊》（Journal of Open Data Humanities）审稿中）。
转录文本源自《叙利亚语数字语料库》（Digital Syriac Corpus）。
词性标注基于使用SEDRA IV API（v1.0.0）实现的词汇匹配算法。
叙利亚语文本的遴选经过优化，以尽可能降低外来因素对叙利亚语语法的影响，并最大限度覆盖公元2至13世纪叙利亚语的关键发展阶段。

POS标注格式与缩写说明

文本文件中的POS标注遵循以下格式：
<句法类别>-<状态>_<叙利亚语词汇>
因此，下划线「_」用于标记标注序列的起始位置，标注值之间以连字符「-」分隔。示例如下（需注意文本方向性限制）：
ܒܘܪܟܬܐ_EMP-N

以下缩写列表给出了所有POS标签的定义，其依据为SEDRA IV API（v1.0.0）中可用的参数：
- 绝对状态名词（不定残留形式）：ABS
- 强调状态名词（新不定形式）：EMP
- 黏附状态名词（绑定名词）：CNS
- 不适用状态：X
- 小品词：PTCL
- 代词：PRO
- 介词：PREP
- 动词：V
- 名源派生词：DEN
- 名词：N
- 体词：SBV
- 形容词：ADJ
- 专有名词：PN
- 副词：ADV
- 民族称谓词：DNM
- 分词形容词：PTCPADJ
- 习语：IDM
- 需结合上下文人工研判：DUP（详见下文「质量控制与局限性」）

注：原文中「adverb」出现两次，对应缩写均为ADV，保留原文对应关系。

质量控制与局限性

平均而言，每份手稿的POS标注流程完成了63.13%的文本覆盖度。本次POS标注流程基于完全匹配规则，未考虑句法或语义语境。因此，存在同音异义现象的叙利亚语词汇将被标注为「DUP」，需结合原始语境进行人工研判。在该语料库中所有已完成POS标注的297,981个词汇中，约73,188个（占比24.56%）的标注涉及存在多重语义和/或句法解释的同音异义词。
由于本数据集是针对名词状态形态的研究项目的一部分，因此专门针对各类状态值增设了标注项。而本研究无需涉及数、性等语法要素，故未将其纳入POS标注流程。

联系方式

如有任何疑问，请联系Charbel El-Khaissi，邮箱：[e-mail]。