1 of 22

Lost in Translation: Investigating Systematic Discrepancies between Parallel English and Chinese Names of American Chinese Restaurants (ACRs)

Hang Jiang, Nanxi Liu, Cassandra Overney, Rukun Zhang, Artemisia Luk, Diyi Yang,

Jad Kabbara, Deb Roy

MIT Center for Constructive Communication, Stanford University

2 of 22

Background

3 of 22

It begins with a story…

Generated by ChatGPT-4o

4 of 22

Szechuan vs. Hangzhou

Translation:

Intoxicated Hangzhou

5 of 22

More Mismatched Names

Lingnan Restaurant

Taste of Sea

Jiangnan Mansion

Szechuan Mountain + Pangolin (pun)

The process of immigrant groups adapting their language, values, and behaviors to a new society is commonly known as cultural assimilation in sociology and ethnic studies. Among the diverse immigrant groups in the US, colloquially hailed as “a melting pot”, those of Chinese ethnicity (referred to as “Chinese”) hold a significant place, because their path to assimilation is not in a “straight-line” fashion [1, 4] but somehow “bumpy” and “segmented” [2, 3]. Over time, Chinese immigrants have established themselves through the restaurant business, which is central to the dining scene in the United States. Different from other ethnic restaurants in the US, Chinese restaurants usually have parallel names in English and Chinese, which sometimes do not match in meaning (Figure 1).

6 of 22

Why Does It Matter?

For these Chinese immigrants, the parallel names of their restaurants reflect both an embrace of cultural assimilation and a preservation of their home identity.

7 of 22

Prior Work

Prior work has primarily focused on common English ACR names – “panda”, “china”, “wok”, “great wall” (Chen, 2018).

No one has studied the parallel English and Chinese ACR names.

8 of 22

Research Questions (RQs)

RQ1: What is lost in translation between Chinese and English ACR names?
RQ2: How are socio-contextual factors correlated with naming discrepancies?
RQ3: What are the underlying reasons to explain such discrepancies?

9 of 22

Dataset & Method

10 of 22

Name Transcription

Sampled and annotated 3166 ACRs (10% in U.S.)

Chinese Name: 快乐人

Translation: Happy Guy

English Name: Uncle Lou

To study name discrepancies, we first need to have parallel English and Chinese names. Go to a google maps page, there is usually only English name.
We first curated a comprehensive list of restaurants tagged as “Chinese restaurant” in the United States through the Outscrapper1 platform, which allows us to get the public information of ACRs via the Google Maps Places API. Afterwards, we filtered out chain or fast food restaurants since their names tend to repeat many times within the list. We obtained 31,548 unique ACRs in the United States and used a state-based stratified sampling strategy to sample 3,166 ACRs (10% of the total) from different states.
To obtain accurate transcriptions of their Chinese names, we first obtained IRB exemption and hired two Chinese-speaking crowd workers to identify and transcribed the Chinese name of each restaurant in simplified Chinese characters by browsing its photos from Google Maps (see Figure 1). If there is any disagreement between two annotators, one author of the paper will review the annotations and make a decision. We observed a decent inter-rater agreement score 0.673 in Cohen’s Kappa. After transcribing Chinese names of the 3,166 ACRs, we identified 1,720 ACRs with both English and Chinese names and 1,446 with only English names.

11 of 22

8 Name Frames

Adapted from Chen (2018)

12 of 22

Frame Annotation + Model Evaluation

Model Evaluation

Models (e.g., Rule-based, ChatGPT)

1

Model Comparison

Validation

Please decide if a restaurant name contains any personal name.

Definition

Personal names are usually, but not necessarily, surnames or first names.

Here are some common examples:

- Surnames with a possessive: Qing’s Kitchen, Hoy’s Wok

- Names without a possessive: China Lee, Hunan Mao, House of Louie

If the name of the restaurant ''Uncle Lou'' contains any personal name, return 1; if not, return 0.

Codebook → Instructions

Human Annotation

1

0

1

3 annotators

Rounds of discussion

Decision

13 of 22

Quantifying Naming Discrepancies

1. Frame-based Jaccard Distance (JD)

2. Embedding-based Cosine Distance (CD)

14 of 22

Interviews in Chinatown

15 of 22

Analysis

16 of 22

RQ1: What Is Lost in Translation?

Positivity and Ambiance are more emphasized in Chinese names
Specialty and Location are more emphasized in English names

17 of 22

RQ2: Location vs. Naming Discrepancies

ACRs in urban areas show higher discrepancies than those in rural and suburb areas

18 of 22

RQ2: Chinese Percentage vs. Naming Discrepancies

ACRs in areas with higher Chinese percentage show higher discrepancies

19 of 22

RQ3: Explanations of Phenomena

1. Customer base

2. Neighborhood

3. Personal experience

R7 from Chinatown: “… unique to stand out so that they can have business …”

R9 without Chinese name : “There are no Chinese customers here …”

R3 with personal name: “I want to show love to my daughter …”

20 of 22

Takeaways

Present a novel study that examines the cultural assimilation of Chinese in the US and uncovers systematic semantic discrepancies between English and Chinese names of ACRs
Adapt 8-dimension name frames to capture underlying meanings of names and create the Name Frames dataset on 884 pairs of ACR names
Quantify semantic discrepancies with two metrics, showing that for ACRs in urban areas or those in counties with a higher percent of Chinese, the discrepancies are significantly larger
Conduct ten qualitative interviews to complement the computational analysis, which helps us better understand their motivations and experiences behind their naming practices.

21 of 22

Thank you!�hjian42@mit.edu

22 of 22

Performance