Chinese-SNACS

This corpus annotates adposition supersenses for the Mandarin Chinese translation of the Little Prince (小王子). Adposition supersense annotations are based on English supersense guidelines v2.5 (https://arxiv.org/abs/1704.02134) and source Chinese-English bitext is obtained from the AMR project (https://amr.isi.edu/download/amr-bank-struct-v1.6.txt).

The corpus contains all 27 chapters of The Little Prince, with a total of 1562 sentences, 19799 tokens, and 937 adposition targets.

Data are stored in the following files:

• snacs_annotated.txt: plain text file with TOKEN:SCENE~FUNCTION annotations
• snacs_stanza.conllulex: 20-column extended conllu file with stanza-parsed and udapy converted dependencies and supersense annotations in the 14th, 15th columns
• cf. conllulex validation (https://github.com/nert-nlp/conllulex)

How to obtain the data:

Step 1: stanza parse using customized models

cd scripts

python stanza_parse_ud1.py

• stanza-parsed UDv1 output file: snacs_stanza.conllu1
• pretrained Chinese OntoNotes stanza models can be downloaded here: https://drive.google.com/drive/folders/14439AS7xeKAInpk3XDfZdqLgHDRIHI15?usp=sharing

Step 2: use udapi to convert UD v1 to UD v2

if udapi not installed

pip install udapi

back to Chinese_SNACS main directory

cd ../

udapy -s ud.Convert1to2 < snacs_stanza.conllu1 > snacs_stanza.conllu2

Step 3: generate conllulex file by incorporating supersense annotations, adposition alignments and postag corrections (to pass validation)

cd scripts/

python ud2_to_conllulex.py

Chinese to English adposition alignments

• For each adposition in Chinese, two annotators search across adpositions in the parallel English sentence and choose the aligned adposition if any.
• Adposition alignment annotations appear in column 9, e.g. AlignedTokId=6|AlignedAdposition=on
• Multi-word adpositions do not occur in Chinese, but appear in English. For example, Chinese adpositions are aligned to English MWE such as because_of, as_if, out_of, etc; AlignedTokId=2,3|AlignedAdposition=ahead_of.
• Adposition alignments are not yet annotated for Chapters 1, 4 and 5.

Chinese Lexical Categories (lexcat)

• ADJ: adjective, e.g. 新
• ADV: adverb, e.g. 也
• AUX: aspect marker, e.g. 着
• BA: ba-construction, e.g. 把
• CCONJ: coordinating conjunction, e.g. 和
• DE: de-construction, e.g. 的
• DET: determiner, e.g. 这
• INTJ: interjection, e.g. 哈哈
• LB: long bei construction, e.g. 被
• LC: localizer, e.g. 里
• N: noun, e.g. 中国
• NUM: number, e.g. 一百
• P: Prepositions (excluding 把 and 被), e.g. 在
• PART: particle, e.g. 吗
• PRON: pronoun, e.g. 我
• PUNCT: punctuation, e.g. 。
• SCONJ: subordinating conjunction, e.g. 虽然
• X: other part-of-speech, e.g. mathematical sign, foreign words

Annotation Contributors:

• Adposition annotations and analyses: Yang (Janet) Liu, Yilun Zhu, Siyao (Logan) Peng, Yushi Zhao, Austin Blodgett; supervised by Nathan Schneider
• Chinese-English adposition alignments: Wai Ching Leung, Yifu Mu; coordinated by Siyao (Logan) Peng and supervised by Nathan Schneider

References

Please cite:

@inproceedings{peng-etal-2020-corpus,
    title = "A Corpus of Adpositional Supersenses for {M}andarin {C}hinese",
    author = "Peng, Siyao  and
      Liu, Yang  and
      Zhu, Yilun  and
      Blodgett, Austin  and
      Zhao, Yushi  and
      Schneider, Nathan",
    booktitle = "Proceedings of the 12th Language Resources and Evaluation Conference",
    month = may,
    year = "2020",
    address = "Marseille, France",
    publisher = "European Language Resources Association",
    url = "https://aclanthology.org/2020.lrec-1.733",
    pages = "5986--5994",
    abstract = "Adpositions are frequent markers of semantic relations, but they are highly ambiguous and vary significantly from language to language. Moreover, there is a dearth of annotated corpora for investigating the cross-linguistic variation of adposition semantics, or for building multilingual disambiguation systems. This paper presents a corpus in which all adpositions have been semantically annotated in Mandarin Chinese; to the best of our knowledge, this is the first Chinese corpus to be broadly annotated with adposition semantics. Our approach adapts a framework that defined a general set of supersenses according to ostensibly language-independent semantic criteria, though its development focused primarily on English prepositions (Schneider et al., 2018). We find that the supersense categories are well-suited to Chinese adpositions despite syntactic differences from English. On a Mandarin translation of The Little Prince, we achieve high inter-annotator agreement and analyze semantic correspondences of adposition tokens in bitext.",
    language = "English",
    ISBN = "979-10-95546-34-4",
}