Language of Bargaining Dataset

Dataset Summary

This repo contains the Natural Language (NL) and Alternating Offer (AO) negotation transcript data as described in https://aclanthology.org/2023.acl-long.735.pdf.

The data includes the "Bargaining Act" annotations for the NL setting.

Dataset Information

Within the negotiations-data.zip compressed folder you will find two folders containing the AO and NL data.

Alternating Offers (AO)

The AO data is found in the folder ao. The folder contains 230 files, each one representing a single negotation conducted in the AO setting. They are in JSONL format where each line adheres to the following schema:

{
  "id": string,
  "role": string seller|buyer,
  "message": float,
  "created_at": datetime,
  "status": string
}

There are multiple ways you could load this data, here is an example using the pandas Python package for loading a single negotiation transcript, e.g. ao/299.jsonl.

import pandas as pd

ao_file = 'ao/299.jsonl'
negotiation_df = pd.read_json(ao_file, orient='records', lines=True)

Natural language (NL)

The NL data is found in the folder nl. The folder contains 178 files, each one representing a single negotation conducted in the NL setting. They are in JSON format and adhere to the following schema:

{
  "duration_min": float,
  "turns": [
    [
      {
        "ID": string,
        "Role": string seller|buyer,
        "Word": string,
        "Span": string,
        "Spoken Numeric": string,
        "Numeric": string,
        "Category": string,
        "Firm or Soft": string,
        "External Incentive": string
      },
      ...
      {
        ...
      },
    ]
  ]
}

There are multiple ways you could load this data, here is an example using the json and pandas Python package for loading a single negotiation transcript, e.g. nl/100.json.

import pandas as pd
import json

nl_file = 'nl/100.json'
with open(nl_file, 'r') as f:
  data = json.load(f)
  # data contains the data, in json, of the entire negotiation
for utt in data['turns']:
  turn_df = pd.DataFrame.from_records(utt)
  # turn_df contains a single utterance, as a dataframe, where each row corresponds to one transcribed word.
  # Process and do something with turn_df here...

Annotation Instructions

Below is a relevant portion of the instructions provided to annotators. It shoud help clarify the data:

Span:
- x on all tokens included in the actual offer
- Be inclusive - main idea is to exclude entirely different sentences
Spoken Numeric
- t: if thousands place is spoken aloud (234,000), else blank
Numeric (i.e. offer amount)
- Single number (plain numeric, thousands)
  - 220
- Range or Bounds
  - [220, 225] [,225] [220,]
Category
- n: Numeric Offer (common case) - used for new numeric offers (numbers that haven’t been mentioned yet)
- p: Push (request the other person move in my direction, mainly needs to reference offer made by other party)
- a: Allowance (offer to move in the other persons’ direction without explicit number, needs to reference offer made by other party)
- c: Comparisons to prices of external stuff (price of comparable house, price of imaginary existing offers)
- r: Repetition of previous offer, non-committal
- e: End of negotiation via offer acceptance entering mutual common ground (e.g., offer given followed by, “That works for me, let’s do it.”) - explicitly only happens once.
Firm or Soft
- s: Soft number (e.g., 220ish)
  - Suffixes: “-ish”, prefixes: “around”, etc.
- f: Firm number
External Incentive
- y: speaker incorporates as a part of the offer non-monetary incentives (landscaping, sale/offer timing, cash payment of amount vs mortgage), needs to reference an offer.

Anonymizing

We replaced all occurences of participant names with the token [PERSON].

Citation

If you use this dataset in your research or publication, please cite it as:

@inproceedings{heddaya-etal-2023-language,
    title = "Language of Bargaining",
    author = "Heddaya, Mourad  and
      Dworkin, Solomon  and
      Tan, Chenhao  and
      Voigt, Rob  and
      Zentefis, Alexander",
    booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2023",
    address = "Toronto, Canada",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.acl-long.735",
    pages = "13161--13185",
}

Datasets:

ChicagoHAI
/

language-of-bargaining

You need to agree to share your contact information to access this dataset

Language of Bargaining Data License Agreement