The ENP-China Named Entity Annotation Guidelines are designed to provide guidance in the annotation campaign that the ENP-China project is implementing on historical documents extracted from the Shenbao newspaper and from Who’s who directories in February 2023. It is part of series of annotation initiatives that aim to create reliable resources for Natural Language Processing of Chinese historical sources produced in the 1850-1949 period.

1 Preamble

Named-entity recognition (NER) is a subtask of information extraction that seeks to identify and classify named entities mentioned in unstructured text into pre-defined categories such as person names, organizations, locations, time expressions, quantities, monetary values, percentages, etc.

Annotation is a common task in machine learning. Annotation aims to provide the machine with correct examples (often termed “ground truth” or “gold standard”) that originate from human reading and interpretation. More specifically, entity annotation aims to teach natural language processing (NLP) models how to identify parts of speech, named entities and key phrases within a text. In this task, annotators read the text thoroughly, locate the target entities, highlight them on the annotation platform and choose from a predetermined list of labels. To help NLP models learn about named entities further, entity annotation in this campaign will be paired with entity linking.

This annotation campaign has four major objectives:

  1. To provide a ground truth for what can be called “transitional” Chinese (1872-1949)
  2. To integrate new genres of sources (biographies)
  3. To integrate new categories of entities (events, positions)
  4. To add a set of complex tasks such as entity linking, relations between entities (biographies), co-references, discontinuous entities.

The core objective is to extract information from historical newspaper articles and other historical sources to support the querying, filtering and analysis of large collection of newspaper archives, and to build a historical knowledge base that will eventually connect to knowledge bases (e.g. Wikidata, MCBD). The explicit goal is to create an annotated dataset specifically for texts written in Chinese for the 1870-1949 period, which we call “transitional” because the Chinese language of this period actually includes different stages of language development. This campaign aims at filling a gap between the datasets that exist for classical Chinese (prior to the Song dynasty, 960-1279) and contemporary Chinese (post 1949).
Another objective is to participate in the next HIPE campaign(s)

These guidelines are derived from the Impresso Guidelines.

The campaign includes two major tasks and some specific subtasks (depending on the genre of sources):

  1. Named entity recognition (NER): the main task is to identify not just regular named entities, but also complex named entities, including names with multiple nested components and to identify new categories of entities such as events and positions.
  2. Named entity linking (NEL): the main challenge is to link entities not only to mainstream knowledge bases (e.g. Wikidata) but also to knowledge bases specific to the history of modern China (MCBD, Baidu).
  3. Co-reference and relations: this is about identifying incomplete named entities (surname only, pronoun, etc.) and to link them to their complete reference in the text ; relations aims at linking entities that are interrelated. See further instructions below.

1.1 Methodology

These guidelines were built as follows:

  1. Selection of high level types of interest by historians and Natural Language Processing (NLP) specialists;
  2. Constitution of “mini-reference” corpora from enp-china sources for Chinese and English;
  3. Annotation of mini-references to test and validate our guidelines
  4. Curation of mini-references, selection of exemplary difficult cases;
  5. Final validation of guidelines.

We use INCEpTION as annotation tool, without the visualization of image segments (which are not available for Chinese texts). For a quick introduction to the ENP-China interface of InCepTION, see our Quick Guide

2 General definition

The objective is to annotate all named entities in the documents described above. Named entities are linguistic units that must include a proper name, or a definite description having the status of a proper name. We do not specify further the definition of proper names, but instead rely on the linguistic intuition/awareness of annotators, who should always keep in mind our objective of extracting ‘historical’ information typically conveyed via referential entities. There will be borderline cases, which we ask annotators to report in a separate file for further discussion.

Examples:

Phrases such as

are not annotated because they do not contain proper names.

Phrases such as


汪       org.adm name pers.ind 偽 政 府

are annotated.

2.1 Entity types and subtypes

The tables below describe the types and subtypes of entities that need to be annotated during the campaign:

Type Subtypes Subtypes2 Description
pers pers.ind A single person (王曉賴)
pers.coll A named group of people
pers.ind.articleauthor A single person who is the author of an article
org org.adm Organization that plays a mainly administrative role (財政部)
org.asso Professional or students’ associations, clubs…
org.busi Business enterprises, private companies, banks…
org.edu Educational institutions (school, college, university, academy…)
org.ent Other type of undefined organizations.
org.ent org.ent.pressagency Special type related to newspaper to spot press agencies (e.g., 申報館)
prod prod.media Newspapers, magazines… etc. (e.g., 申報, 東方雜誌…)
prod.doctr Political, philosophical, religious, sectarian doctrines.
prod.creation Creative works (book, theatrical play, movie, etc.)
loc loc.adm loc.adm.dist City or town district (e.g. 南市 in Shanghai)
loc.adm.town Village, town, city (上海, 上海市)
loc.adm.reg Regions, provinces (江蘇)
loc.adm.nat Countries
loc.adm.sup World regions, continent
loc.phys loc.phys.geo Mountains, plains, plateaus, caves, volcanoes, canyons
loc.phys.hydro Oceans, seas, rivers, streams, ponds, marshes
loc.phys.astro Planets, stars, galaxies and their parts
loc.oro Refers to roads, highways, streets, avenues, squares, etc.
loc.fac Refers to buildings and other facilities
loc.add loc.add.phys Refers to physical addresses (street name, street number, etc)
loc.add.elec Refers to electronic contact information (not applicable in most cases)
loc.unk Type used when it is not possible to choose among other location types
time time.abs time.abs.year Absolute datation: year (e.g., 1931, 一九三一年)
time.abs.month Absolute datation: month (e.g., 十一月)
time.abs.day Absolute datation: day (e.g., 三日, 星期三)
time.rel time.rel.ref Relative datation: referent (imperial reigns or founding of the Republic)
time.rel.year Relative datation: year
time.rel.month Relative datation: month
time.rel.day Relative datation: day
event event.type Event (meeting, conference, etc)


2.2 Components & Nested Entities

A named entity mention consists of one or more components, as well as parts without explicit component markup. The ENP-China Guidelines partially follow Impresso guidelines and consider the following components:

Type Components Description
All (except time) name is the only transversal component and is applied to any class except time
pers comp.name The component includes first, middle and last names as well as nickname
and initials of individuals
comp.title Title or designator of a person.
comp.qualifier A qualifier specifies a person in the form of a qualifying adjective.
comp.func A function or job of a named person.
comp.demonym The geographical origin of a person
comp.degree Academic degree obtained by a person (M.A., PhD…)
org comp.lev Sublevel in an organization (department, bureau…)

In the next sections, we provide examples of annotated entities with components and nested entities for various types and subtypes of entities. Nested entities are annotated for the types PERS, LOC, ORG, without limit of nesting level during the annotation phase.

2.3 Examples

2.3.1 Persons






Name Ambigious case Title Ambigious case
孫 name pers.ind 中 山
吳 name pers.ind 德 君
何    pers.ind name 君 title
何    pers.ind name 君    title 少 name 山
張    pers.ind name 邑 title 尊
周 pers.ind name 芝 田 太 title 守










Demonym Function (position) Ambigious case
宜 pers.ind demonym town 興 人 王 name 南 祥
吳 name pers.ind 市 func 長 鐵 成
上 pers.ind town 海 市 市 func 長 吳 name 鐵 成
孫 pers.ind name 蕚 齋 司 func 馬
孫 name pers.ind 司 func 馬 蕚 齋











Function (position) Ambigious case
吳 name pers.ind 市 func 長 鐵 成
上 pers.ind town 海 市 市 func 長 吳 name 鐵 成
孫 pers.ind name 蕚 齋 司 func 馬
孫 name pers.ind 司 func 馬 蕚 齋









Liminal cases
孫 name pers.ind 司 func 馬 蕚 齋
台 pers.ind reg name 南 縣 境 袁    name 縣 func 長
包    pers.ind name 探 func 員





2.3.2 Organizations













中 org.edu 央 研 究 院
上 org.busi town 海 银 行
中 org.busi name 国 银 行 上 town 海 分 行
中 org.edu name 央 研 究 院 歷 lev 史 研 究 所
行 org.adm name 政 院 善 lev 後 救 濟 總 署 上 town 海 分 署 善 lev 後 工 作 計 劃 委 員 會
中 org.asso name 大 畢 業 同 學 會 上 town 海 分 會
Ambiguous cases: Organization or Facility?
海 org.adm 防 廳
displaCy
海 loc.fac adm 防 廳 署













There will be cases when it is difficult to categorizean entity as an organization or as a facility (e.g., movie theater, restaurant, dancing hall…). In such cases, the context will help determine which category to adopt. For example:

  • I spent the night in the Cathay Hotel(facility).
  • The Cathay Hotelhas acquired a large plot of land to expand (organization).
  • The manager of the Cathay Hotelhas announced that he will resign next week (organization).
Other examples Not to be annotated
中央銀行貼放委員會工業貸款審査委員會
中央黨部經濟政策研究委會
雜糧商業同業公會籌備委員會
偽郵政總局駐滬辦事處
剿共建國萬元勞軍大會籌備會宣傳處
大夏大學會計系
天津運輸株式會社接收委員會
四十二軍
廿一軍四三五旅五團輸送連
中西門大街保甲局
上海北市業會舘籌賑公所
上海同仁轉元堂
同仁輔元堂分局
吳淞四明公所
美商同茂洋行
 美公司
英公司
芝蘭室
花烟館
英京倫敦電報
英京電信
英京電報
西報
代表會
代表團
俄軍隊
保甲處
財政廳
海關

2.3.3 Locations






















胡 loc.adm.town 家 宅
錢 loc.adm.town 家 塘
南 loc.add.phys dist 市 徽 oro name 寗 路
漢 loc.add.phys town 口 雲 oro name 南 路
南 loc.add.phys town name 華 新 邨 十 address-number 七 號
寗 loc.add.phys oro name 波 路 五 address-number 六 號
吳 loc.oro name 家 弄
恒 loc.oro name 豐 路 橋
西 loc.oro 孫 name 家 橋
復 loc.phys.hydro name 興 島 漁 管 區
木 loc.phys.hydro name 齊 河
蘇 loc.adm.reg 北
蘇 loc.adm.town reg 北 東 name 台
Ambiguous cases
上 loc.adm.town name 海 法 租 界

Is annotated (The French Concession
in Shanghai is an administrative entity)
上 loc.adm.town 海       租界

The concession(s) in Shanghai is
a generic name, and therefore not annotated.
余公館

Names of private residences (Mr. Yu’s residence) are
not considered as named entities, and therefore are not annotated.
Locations or persons?
上 loc.adm.town name 海 市      敎育局長
兩 loc.adm.reg 江      總督
江 loc.adm.reg 蘇      海運委員






















2.3.4 Time

2.3.4.1 Absolute datation

Absolute dates are dates that appear in the following format: YYYY or YYYY-MM or YYYY-MM-DD. Examples:





一 time.abs year 九 三 一 年
一 time.abs year 九 三 一 年 三 month 月
一 time.abs year 九 三 一 年 十 month 一 月 三 day 日





2.3.4.2 Relative datation

Relative dates are dates that can be calculated in reference to a specific point in time (foundation of the Republic, imperial reigns). The point of reference may or may not be explicit in the text. Examples:
























民 time.rel ref 廿 year 二 年
三 time.rel year 十 四 年
中 time.rel ref 華 民 國 二 year 十 六 年 五 month 月 三 day 十 日
二 time.rel year 十 七 年 四 month 月
一 time.rel month 月 七 day 日
上 time.rel ref 年 三 month 月
上 time.rel ref 年 十 month 二 月 六 day 日
For relative dates that do not refer to a explicit referent in the text itself,
the whole string of character needs to be annotated with < time.rel >
without any further detail.
上 time.rel 月 廿 五 日
上 time.rel 週 三
下 time.rel 星 期 三
下 time.rel 月 三 日
昨 time.rel 三 十 日
昨 time.rel 三 十 日
昨 time.rel （ 二 十 日 ）
昨 time.rel 晚
本 time.rel 年 一 月 二 十 日
本 time.rel 星 期 五
本 time.rel 月 一 十 五 日
displaCy
次 time.rel 年 七 月
明 time.rel 日
三 time.rel year 十 四 年 度 八 month 月 至 十 month 二 月 份

Not to be annotated
一個半月 一個月 一個月前, 一月半前
一兩年 一兩日 建國時期
建隆年間 抗戰八年 抗戰後二年
敵僞時期

























2.3.5 Events

Events refer tonamed events such as wars, political movements, meetings of organizations, etc. Examples:












Events
五 event.type 一 勞 動 節
俄 event.type nat 國 革 命
中 event.type asso 華 海 員 工 會 第 一 次 全 國 代 表 大 會
Simple events Nested events
北伐
七七事變
七五事件
七屆全運會
全國地政檢討會議
八一三事變
 三中全會第五次大會
二屆大東亞青少年指導會議
申新九廠罷工流血事件
Events or Organizations?
Events Organizations
國民參政會駐會委員會第四次會議
國立北平研究院學術會議二次大會
國防最高委員會第二二五常會
浙江省第六屆省運會
立法院第二會期第十六次會議
陸軍軍官學校成立廿二週年紀念火會
 市小校長聯誼會
浙江麗水旅滬同鄕會













2.3.6 Additional Layers

2.3.6.1 Discontinuity

The nature of the Chinese language makes it possible to detach the names of various entities from the entity they refer to. For example, one can find a list of names of companies that refer to the term 公司 at the end of the sentence.

2.3.6.2 Coreference

Coreference is the task of finding all expressions that refer to the same entity in a text. In this campaign, we will link the named entities that are mentioned in a text with all their alternative forms in the same text. For this specific task, you need to select the “co-reference” layer. Then highlight the expression you want to link and draw an arrow toward the named entity it refers to.

2.3.6.3 Relations

This task applies exclusively to biographies. The objective is to connect entities that are interrelated, such as a position (comp.func) that a person held in an organization (org.) at a given time (time) and place (loc).
For this specific task, first select the “NER” layer. Then select the mention of the position and link it to the related entities (e.g., organization, date, location). In case a position refers to the same organization but with different dates or places, each link needs to be numbered.

Example Chinese

Example English

2.4 Entity Linking

Entity linking consists in linking the named entity found in the text with its avatar defined by a unique identifier in an external knowledge base, e.g. wikidata. This is a crucial task for disambiguating names. A given entity can appear under different names (e.g., 孔祥熙, 孔庸之). Conversely, different persons can share the same names (e.g., 王大吉, John Smith…).

In this campaign, entity mentions are linked against Wikidata, and against the Modern China Biographical database (MCBD). We link the following entity types: Persons (pers.ind), Organizations (org.) , Locations (loc.), and Events (events). In case the entity to which the mention refers to is not present in Wikidata , the mention is marked as referring to a NIL entity.

Entity components and nested entities are excluded from the linking. For example, in the entities below, components (in red) are excluded from linking.

中國紅字會第一醫院

中央銀行貼放委員會工業貸款審査委員會

In the case of a long entity, the full string needs to be annotated if it can be found as such or its historical avatar in the knowledge base.

For historical entities for which it is difficult to determine the exact corresponding referent, or for which there is no exact historical referent but a contemporary one in Wikidata, one has to pick the contemporary entity (for example 漢口, 漢陽, 武昌 in late imperial China are not the same as the contemporary 武漢 referred to in Wikidata, but in the absence of historical entity one takes the contemporary 武漢 Wikidata entry).