Bntqa Mbart

🚀 BnTQA-mBart: Low-Resource Table Question Answering

BnTQA-mBart is a model designed for low-resource table question answering, especially for Bengali and other Indic languages. It addresses the lack of data and models in this area by introducing a large - scale data generation process.

🚀 Quick Start

This section provides a quick guide on how to use the BnTQA-mBart model for table question answering.

📦 Installation

The installation steps are not provided in the original README. If you need to install relevant libraries, you can use the following general commands:

pip install pandas datasets transformers

💻 Usage Examples

Basic Usage

import pandas as pd
from datasets import load_dataset
from transformers import MBartForConditionalGeneration
model = MBartForConditionalGeneration.from_pretrained("vaishali/BnTQA-mBart")
tokenizer = AutoTokenizer.from_pretrained(args.pretrained_model_name, src_lang="bn_IN", tgt_lang="bn_IN")
forced_bos_id = forced_bos_token_id = tokenizer.lang_code_to_id["bn_IN"]


# linearize table
def process_header(headers: List):
  return "<কলাম> " + " | ".join(headers)

def process_row(row: List, row_index: int):
  en2bnDigits = {'0': '০',  '1': '১', '2': '২', '3': '৩', '4': '৪', '5': '৫', '6': '৬', '7': '৭', '8': '৮', '9': '৯', '.': '.'}
  row_str = ""
  row_cell_values = []
  for cell_value in row:
      if isinstance(cell_value, int) or isinstance(cell_value, float):
          cell_value = convert_engDigit_to_bengali(str(cell_value))
          row_cell_values.append(str(cell_value))
      else:
          row_cell_values.append(cell_value)
  row_str += " | ".join(row_cell_values)
  bn_row_index = []
  for c in str(row_index):
      bn_row_index.append(en2bnDigits[c])
  return "<রো " + "".join(bn_row_index) + "> " + row_str

def process_table(table_content: Dict):
  table_str = process_header(table_content["header"]) + " "
  for i, row_example in enumerate(table_content["rows"]):
      table_str += process_row(row_example, row_index=i + 1) + " "
  return table_str.strip()

# load the dataset
banglatableQA = load_dataset("vaishali/banglaTabQA")

for sample in banglatableQA['train']:
  question = sample['question']
  input_table = pd.read_json(sample['table'], orient='split')
  answer = pd.read_json(sample['answer'], orient='split')

  # create the input sequence: query + linearized input table
  table_content = {"header": list(input_table.columns)[1:], "rows": [list(row.values)[1:] for i, row in input_table.iterrows()]}
  linearized_inp_table = process_table(table_content)
  linearized_output_table = process_table({"name": None, "header": [translate_column(col) for col in list(answer.columns)], 
                             "rows": [list(row.values) for i, row in answer.iterrows()]})
  source = query + " " + linearized_inp_table
  target = linearized_output_table
  input = tokenizer(source,
                    return_tensors="pt",
                    padding="max_length",
                    truncation="longest_first",
                    max_length=1024,
                    add_special_tokens=True)

  with tokenizer.as_target_tokenizer():
    labels = tokenizer(target,
               return_tensors="pt",
               padding="max_length",
               truncation="longest_first",
               max_length=1024,
               add_special_tokens=True).input_ids

  # inference
  out = model.generate(input["input_ids"].to("cuda"), num_beams=5, return_dict_in_generate=True,
                                output_scores=True, max_length=1024)

📚 Documentation

The model is based on facebook/mbart-large-50 and is trained on the vaishali/banglaTabQA dataset. It is designed for low - resource table question answering, especially for Bengali and other Indic languages.

Model Information

📄 License

This project is licensed under the MIT License.

📚 BibTeX entry and citation info

@inproceedings{pal-etal-2024-table,
    title = "Table Question Answering for Low-resourced {I}ndic Languages",
    author = "Pal, Vaishali  and
      Kanoulas, Evangelos  and
      Yates, Andrew  and
      de Rijke, Maarten",
    editor = "Al-Onaizan, Yaser  and
      Bansal, Mohit  and
      Chen, Yun-Nung",
    booktitle = "Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2024",
    address = "Miami, Florida, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.emnlp-main.5",
    pages = "75--92",
    abstract = "TableQA is the task of answering questions over tables of structured information, returning individual cells or tables as output. TableQA research has focused primarily on high-resource languages, leaving medium- and low-resource languages with little progress due to scarcity of annotated data and neural models. We address this gap by introducing a fully automatic large-scale tableQA data generation process for low-resource languages with limited budget. We incorporate our data generation method on two Indic languages, Bengali and Hindi, which have no tableQA datasets or models. TableQA models trained on our large-scale datasets outperform state-of-the-art LLMs. We further study the trained models on different aspects, including mathematical reasoning capabilities and zero-shot cross-lingual transfer. Our work is the first on low-resource tableQA focusing on scalable data generation and evaluation procedures. Our proposed data generation method can be applied to any low-resource language with a web presence. We release datasets, models, and code (https://github.com/kolk/Low-Resource-TableQA-Indic-languages).",
}