crfm-helm 0.5.2__py3-none-any.whl → 0.5.4__py3-none-any.whl

helm/benchmark/scenarios/bhasa_scenario.py

@@ -0,0 +1,1942 @@
+import datasets
+import os
+import random
+from typing import List, Dict
+
+import pandas as pd
+
+from helm.benchmark.scenarios.scenario import (
+    Input,
+    Instance,
+    Output,
+    PassageQuestionInput,
+    Reference,
+    Scenario,
+    CORRECT_TAG,
+    TEST_SPLIT,
+    TRAIN_SPLIT,
+)
+from helm.common.general import ensure_file_downloaded
+from helm.common.hierarchical_logger import hlog
+
+# BHASA Scenarios
+#   A. Natural Language Understanding
+#   B. Natural Language Generation
+#   C. Natural Language Reasoning
+#   D. Linguistic Diagnostics
+
+# A. Natural Language Understanding
+#   1. Question Answering
+#   2. Sentiment Analysis
+#   3. Toxicity Detection/Classification
+
+
+# 1. Question Answering
+# 1.1 Indonesian: TyDiQA
+class TyDiQAScenario(Scenario):
+    """
+    TyDiQA is is an open-book question answering scenario for 11 typologically-diverse languages.
+    The questions are written by people who want to know the answer, but do not know the answer yet,
+    and the data is collected directly in each language without the use of translation.
+
+    This scenario only uses the Indonesian subset of the data, and uses the Gold Passage (GoldP) task,
+    which requires the tested system to extract a span from the given passage to answer a given question.
+    There are no unanswerable questions.
+
+    The models are prompted using the following format:
+
+        Anda akan diberikan sebuah paragraf dan sebuah pertanyaan. Jawablah pertanyaannya dengan mengekstrak jawaban
+        dari paragraf tersebut.
+
+        Paragraf: <text>
+        Pertanyaan: <question>
+        Jawaban: <answer>
+
+        ...
+
+        Paragraf: <text>
+        Pertanyaan: <question>
+        Jawaban:
+
+
+    Target completion:
+        <answer>
+
+    @article{clark-etal-2020-tydi,
+        title = "{T}y{D}i {QA}: A Benchmark for Information-Seeking Question Answering in Typologically
+        Diverse Languages",
+        author = "Clark, Jonathan H.  and
+        Choi, Eunsol  and
+        Collins, Michael  and
+        Garrette, Dan  and
+        Kwiatkowski, Tom  and
+        Nikolaev, Vitaly  and
+        Palomaki, Jennimaria",
+        editor = "Johnson, Mark  and
+        Roark, Brian  and
+        Nenkova, Ani",
+        journal = "Transactions of the Association for Computational Linguistics",
+        volume = "8",
+        year = "2020",
+        address = "Cambridge, MA",
+        publisher = "MIT Press",
+        url = "https://aclanthology.org/2020.tacl-1.30",
+        doi = "10.1162/tacl_a_00317",
+        pages = "454--470",
+    }
+    """
+
+    name = "tydiqa"
+    description = "Indonesian Open-book Question Answering task"
+    tags = ["question_answering"]
+
+    def __init__(self):
+        super().__init__()
+        self.splits = {"train": TRAIN_SPLIT, "validation": TEST_SPLIT}
+
+    def get_instances(self, output_path) -> List[Instance]:
+        dataset = datasets.load_dataset("khalidalt/tydiqa-goldp", "indonesian")
+
+        outputs = []
+        for split in self.splits.keys():
+            df = dataset[split].to_pandas()
+
+            if split == "train":
+                # Select only bottom 20th percentile by length for in-context examples as examples are very long
+                data = df[df["passage_text"].apply(len) < df["passage_text"].apply(len).quantile(0.2)]
+            else:
+                data = df
+
+            for _, row in data.iterrows():
+                passage = row["passage_text"].strip()
+                question = row["question_text"].strip()
+                input = PassageQuestionInput(
+                    passage=passage,
+                    question=question,
+                    passage_prefix="Paragraf: ",
+                    question_prefix="Pertanyaan: ",
+                )
+                references = []
+                for answer in row["answers"]["text"]:
+                    output = Output(text=answer.strip())
+                    references.append(Reference(output, tags=[CORRECT_TAG]))
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# 1.2 Vietnamese & Thai: XQuAD
+class XQuADScenario(Scenario):
+    """
+    XQuAD is an open-book question answering scenario that is parallel across 10 languages.
+    The dataset consists of a subset of 240 paragraphs and 1190 question-answer pairs from the
+    development set of SQuAD v1.1 (Rajpurkar et al., 2016) together with their professional translations.
+
+    This scenario only uses the Vietnamese and Thai subsets of the data and there are no
+    unanswerable questions.
+
+    The models are prompted using the following general format:
+
+        You will be given a paragraph and a question. Answer the question by extracting the answer from the paragraph.
+
+        Paragraph: <text>
+        Question: <question>
+        Answer: <answer>
+
+        ...
+
+        Paragraph: <text>
+        Question: <question>
+        Answer:
+
+    Target completion:
+        <answer>
+
+    @article{Artetxe:etal:2019,
+      author    = {Mikel Artetxe and Sebastian Ruder and Dani Yogatama},
+      title     = {On the cross-lingual transferability of monolingual representations},
+      journal   = {CoRR},
+      volume    = {abs/1910.11856},
+      year      = {2019},
+      archivePrefix = {arXiv},
+      eprint    = {1910.11856}
+    }
+    """
+
+    name = "xquad"
+    description = "Vietnamese and Thai Open-book Question Answering task"
+    tags = ["question_answering"]
+
+    def __init__(self, language: str):
+        super().__init__()
+        self.language = language
+        self.splits = {"train": TRAIN_SPLIT, "test": TEST_SPLIT}
+        self.language_to_prompt_components = {
+            "th": {
+                "passage_prefix": "ข้อความ: ",
+                "question_prefix": "คำถาม: ",
+                "random_state": 4520,
+            },
+            "vi": {
+                "passage_prefix": "Đoạn văn: ",
+                "question_prefix": "Câu hỏi: ",
+                "random_state": 4502,
+            },
+        }
+        if self.language not in self.language_to_prompt_components.keys():
+            raise Exception(
+                f"{self.language} not supported. Supported languages are {self.language_to_prompt_components.keys()}."
+            )
+        else:
+            self.prompt_components = self.language_to_prompt_components[self.language]
+
+    def get_instances(self, output_path) -> List[Instance]:
+        dataset = datasets.load_dataset("xquad", f"xquad.{self.language}", split="validation")
+        df = dataset.to_pandas()
+
+        # Sample 1000 examples for test
+        df_test = df.sample(n=1000, random_state=self.prompt_components["random_state"])
+
+        # In-context examples to be drawn from remaining examples (since there is no train data)
+        df_train = df[~df.index.isin(df_test.index)]
+
+        # Select only bottom 20th percentile by length for in-context examples as examples are very long
+        df_train = df_train[df_train["context"].apply(len) < df_train["context"].apply(len).quantile(0.2)]
+        dataset = {
+            "train": df_train,
+            "test": df_test,
+        }
+
+        outputs = []
+        for split in self.splits.keys():
+            data = dataset[split]
+            for _, row in data.iterrows():
+                passage = row["context"].strip()
+                question = row["question"].strip()
+                input = PassageQuestionInput(
+                    passage=passage,
+                    question=question,
+                    passage_prefix=str(self.prompt_components["passage_prefix"]),
+                    question_prefix=str(self.prompt_components["question_prefix"]),
+                )
+                references = []
+                for answer in row["answers"]["text"]:
+                    output = Output(text=answer.strip())
+                    references.append(Reference(output, tags=[CORRECT_TAG]))
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# 1.3 Tamil: IndicQA
+class IndicQAScenario(Scenario):
+    """
+    IndicQA is an open-book question answering scenario for 11 Indic languages.
+    Answers to questions are to be extracted from the text provided. The data is taken from
+    Wikipedia articles across various domains and questions and answers were manually created
+    by native speakers.
+
+    This scenario only uses the Tamil subset of the data and unanswerable questions
+    are removed from the dataset in order to be consistent with the question answering
+    scenarios for Indonesian, Vietnamese and Thai.
+
+    The models are prompted using the following format:
+
+        உங்களுக்கு ஒரு பத்தியும் ஒரு கேள்வியும் தரப்படும். தரப்பட்ட பத்தியிலிருந்து கேள்விக்கான பதிலைக் கண்டறியவும்.
+
+        பத்தி: <text>
+        கேள்வி: <question>
+        பதில்: <answer>
+
+        ...
+
+        பத்தி: <text>
+        கேள்வி: <question>
+        பதில்:
+
+    Target completion:
+        <answer>
+
+    @inproceedings{doddapaneni-etal-2023-towards,
+        title = "Towards Leaving No {I}ndic Language Behind: Building Monolingual Corpora, Benchmark and Models for
+            {I}ndic Languages",
+        author = "Doddapaneni, Sumanth  and
+            Aralikatte, Rahul  and
+            Ramesh, Gowtham  and
+            Goyal, Shreya  and
+            Khapra, Mitesh M.  and
+            Kunchukuttan, Anoop  and
+            Kumar, Pratyush",
+        editor = "Rogers, Anna  and
+            Boyd-Graber, Jordan  and
+            Okazaki, Naoaki",
+        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.693",
+        doi = "10.18653/v1/2023.acl-long.693",
+        pages = "12402--12426",
+    }
+    """
+
+    name = "indicqa"
+    description = "Tamil Open-book Question Answering task"
+    tags = ["question_answering"]
+
+    def __init__(self):
+        super().__init__()
+        self.splits = {"train": TRAIN_SPLIT, "test": TEST_SPLIT}
+
+    def get_instances(self, output_path) -> List[Instance]:
+        dataset = datasets.load_dataset(
+            "ai4bharat/IndicQA",
+            "indicqa.ta",
+            split="test",
+            revision="78ee8d58e880c72f324e176c989dfefa55427af4",
+            trust_remote_code=True,
+        )
+        df = dataset.to_pandas()
+
+        # Remove unanswerable questions (answer is an empty string)
+        df = df[df["answers"].apply(lambda x: len(x["text"][0].strip()) > 0)]
+
+        # Sample 1000 examples for test
+        df_test = df.sample(n=1000, random_state=7900)
+
+        # In-context examples to be drawn from remaining examples (since there is no train/dev data)
+        df_train = df[~df.index.isin(df_test.index)]
+
+        # Select only bottom 20th percentile by length for in-context examples as examples are very long
+        df_train = df_train[df_train["context"].apply(len) < df_train["context"].apply(len).quantile(0.2)]
+        dataset = {
+            "train": df_train,
+            "test": df_test,
+        }
+
+        outputs = []
+        for split in self.splits.keys():
+            data = dataset[split]
+            for _, row in data.iterrows():
+                passage = row["context"].strip()
+                question = row["question"].strip()
+                input = PassageQuestionInput(
+                    passage=passage,
+                    question=question,
+                    passage_prefix="பத்தி: ",
+                    question_prefix="கேள்வி: ",
+                )
+                references = []
+                for answer in row["answers"]["text"]:
+                    output = Output(text=answer.strip())
+                    references.append(Reference(output, tags=[CORRECT_TAG]))
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# 2. Sentiment Analysis
+# 2.1 Indonesian: NusaX Sentiment
+class NusaXScenario(Scenario):
+    """
+    NusaX is a sentiment analysis scenario for 11 Indonesian languages.
+    The data is derived from a subset of SmSA (Purwarianti and Crisdayanti, 2019) and manually translated
+    from Indonesian to 10 other local languages, such as Acehnese and Toba Batak.
+    It consists of comments and reviews from various online platforms.
+
+    Only the Indonesian subset of the data is used for this scenario, and the labels are
+    positive, negative or neutral.
+
+    The models are prompted using the following format:
+
+        Apa sentimen dari kalimat berikut ini?
+        Jawablah dengan satu kata saja:
+        - Positif
+        - Negatif
+        - Netral
+
+        Kalimat: <text>
+        Jawaban: <sentiment>
+
+        ...
+
+        Kalimat: <text>
+        Jawaban:
+
+    Target completion:
+        <sentiment>
+
+    @inproceedings{winata-etal-2023-nusax,
+        title = "{N}usa{X}: Multilingual Parallel Sentiment Dataset for 10 {I}ndonesian Local Languages",
+        author = "Winata, Genta Indra  and
+        Aji, Alham Fikri  and
+        Cahyawijaya, Samuel  and
+        Mahendra, Rahmad  and
+        Koto, Fajri  and
+        Romadhony, Ade  and
+        Kurniawan, Kemal  and
+        Moeljadi, David  and
+        Prasojo, Radityo Eko  and
+        Fung, Pascale  and
+        Baldwin, Timothy  and
+        Lau, Jey Han  and
+        Sennrich, Rico  and
+        Ruder, Sebastian",
+        editor = "Vlachos, Andreas  and
+        Augenstein, Isabelle",
+        booktitle = "Proceedings of the 17th Conference of the European Chapter of the Association for
+            Computational Linguistics",
+        month = may,
+        year = "2023",
+        address = "Dubrovnik, Croatia",
+        publisher = "Association for Computational Linguistics",
+        url = "https://aclanthology.org/2023.eacl-main.57",
+        doi = "10.18653/v1/2023.eacl-main.57",
+        pages = "815--834",
+    }
+    """
+
+    name = "nusax"
+    description = "Indonesian NusaX-Senti Sentiment Analysis dataset"
+    tags = ["sentiment_analysis"]
+
+    def __init__(self):
+        super().__init__()
+        self.splits = {"train": TRAIN_SPLIT, "test": TEST_SPLIT}
+        self.sentiment2label = {
+            "positive": "Positif",
+            "negative": "Negatif",
+            "neutral": "Netral",
+        }
+
+    def download_dataset(self, output_path: str):
+        URLS = {
+            "test": "https://raw.githubusercontent.com/IndoNLP/nusax/main/datasets/sentiment/indonesian/test.csv",
+            "train": "https://raw.githubusercontent.com/IndoNLP/nusax/main/datasets/sentiment/indonesian/train.csv",
+        }
+
+        dataset: Dict[str, pd.DataFrame] = {}
+        for split in self.splits.keys():
+            target_path_file = os.path.join(output_path, split)
+            ensure_file_downloaded(source_url=URLS[split], target_path=target_path_file)
+            data = pd.read_csv(target_path_file)
+            dataset[split] = data
+        return dataset
+
+    def get_instances(self, output_path) -> List[Instance]:
+        dataset = self.download_dataset(output_path)
+        outputs = []
+        for split in self.splits.keys():
+            data = dataset[split]
+            for _, row in data.iterrows():
+                input = Input(row["text"].strip())
+                output = Output(text=self.sentiment2label[row["label"]])
+                references = [
+                    Reference(output, tags=[CORRECT_TAG]),
+                ]
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# 2.2 Vietnamese: UIT-VSFC
+class UITVSFCScenario(Scenario):
+    """
+    UIT-VSFC is a Vietnamese sentiment analysis scenario. The data consists of student feedback obtained from
+    end-of-semester surveys at a Vietnamese university. Feedback is labeled as one of three sentiment
+    polarities: positive, negative or neutral.
+
+    The models are prompted using the following format:
+
+        Sắc thái của câu sau đây là gì?
+        Trả lời với một từ duy nhất:
+        - Tích cực
+        - Tiêu cực
+        - Trung lập
+
+        Câu văn: <text>
+        Câu trả lời: <sentiment>
+
+        ...
+
+        Câu văn: <text>
+        Câu trả lời:
+
+    Target completion:
+        <sentiment>
+
+    @inproceedings{van2018uit,
+        title={UIT-VSFC: Vietnamese students’ feedback corpus for sentiment analysis},
+        author={Van Nguyen, Kiet and Nguyen, Vu Duc and Nguyen, Phu XV and Truong, Tham TH and Nguyen, Ngan Luu-Thuy},
+        booktitle={2018 10th international conference on knowledge and systems engineering (KSE)},
+        pages={19--24},
+        year={2018},
+        organization={IEEE},
+        url={https://ieeexplore.ieee.org/document/8573337},
+    }
+    """
+
+    name = "uitvsfc"
+    description = "Vietnamese Students' Feedback Corpus sentiment analysis task"
+    tags = ["sentiment_analysis"]
+
+    def __init__(self):
+        super().__init__()
+        self.splits = {"train": TRAIN_SPLIT, "test": TEST_SPLIT}
+        self.id2label = {
+            0: "Tiêu cực",
+            1: "Trung lập",
+            2: "Tích cực",
+        }
+
+    def download_dataset(self, output_path: str):
+        URLS = {
+            "train": {
+                "sentences": "https://drive.google.com/uc?id=1nzak5OkrheRV1ltOGCXkT671bmjODLhP&export=download",
+                "sentiments": "https://drive.google.com/uc?id=1ye-gOZIBqXdKOoi_YxvpT6FeRNmViPPv&export=download",
+            },
+            "test": {
+                "sentences": "https://drive.google.com/uc?id=1aNMOeZZbNwSRkjyCWAGtNCMa3YrshR-n&export=download",
+                "sentiments": "https://drive.google.com/uc?id=1vkQS5gI0is4ACU58-AbWusnemw7KZNfO&export=download",
+            },
+        }
+
+        dataset: Dict[str, pd.DataFrame] = {}
+        for split in list(URLS.keys()):
+            file_lines: Dict[str, List[str]] = {}
+            for file in list(URLS[split].keys()):
+                file_lines[file] = []
+                target_path_file = os.path.join(output_path, split, file)
+                ensure_file_downloaded(source_url=URLS[split][file], target_path=target_path_file)
+                with open(target_path_file, "r") as f:
+                    lines = f.readlines()
+                    for line in lines:
+                        file_lines[file].append(str(line).strip())
+            df = pd.DataFrame({"text": file_lines["sentences"], "label": file_lines["sentiments"]})
+            if split == "test":
+                dataset[split] = df.groupby("label", group_keys=False).apply(
+                    lambda x: x.sample(frac=1000 / len(df), random_state=4156)
+                )
+            else:
+                dataset[split] = df
+        return dataset
+
+    def get_instances(self, output_path) -> List[Instance]:
+        dataset = self.download_dataset(output_path)
+        outputs = []
+        for split in self.splits.keys():
+            data = dataset[split]
+            for _, row in data.iterrows():
+                input = Input(row["text"])
+                output = Output(text=self.id2label[int(row["label"])])
+                references = [
+                    Reference(output, tags=[CORRECT_TAG]),
+                ]
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# 2.3 Thai: Wisesight Sentiment
+class WisesightScenario(Scenario):
+    """
+    Wisesight Sentiment is a Thai sentiment analysis scenario. The data consists of social media messages
+    regarding consumer products and services.
+
+    The dataset originally included the label "question" for instances that were questions. These instances
+    made up only a small subset of the data and were dropped in order to make the task more consistent
+    with those of other languages. Labels are therefore only positive, negative or neutral.
+
+    The models are prompted using the following format:
+
+        อารมณ์ความรู้สึกของข้อความต่อไปนี้เป็นอย่างไร?
+        กรุณาตอบโดยใช้คำเดียวเท่านั้น:
+        - แง่บวก
+        - แง่ลบ
+        - เฉยๆ
+
+        ข้อความ: <text>
+        คำตอบ: <sentiment>
+
+        ...
+
+        ข้อความ: <text>
+        คำตอบ:
+
+    Target completion:
+        <sentiment>
+
+    @software{bact_2019_3457447,
+        author       = {Suriyawongkul, Arthit and
+                        Chuangsuwanich, Ekapol and
+                        Chormai, Pattarawat and
+                        Polpanumas, Charin},
+        title        = {PyThaiNLP/wisesight-sentiment: First release},
+        month        = sep,
+        year         = 2019,
+        publisher    = {Zenodo},
+        version      = {v1.0},
+        doi          = {10.5281/zenodo.3457447},
+        url          = {https://doi.org/10.5281/zenodo.3457447}
+    }
+    """
+
+    name = "wisesight"
+    description = "Wisesight Sentiment Thai sentiment analysis task"
+    tags = ["sentiment_analysis"]
+
+    def __init__(self):
+        super().__init__()
+        self.splits = {"train": TRAIN_SPLIT, "test": TEST_SPLIT}
+        self.sentiment2label = {
+            "pos": "แง่บวก",
+            "neg": "แง่ลบ",
+            "neu": "เฉยๆ",
+        }
+
+    def download_dataset(self, output_path: str):
+        URL = "https://github.com/PyThaiNLP/wisesight-sentiment/raw/master/huggingface/data.zip"
+        data_path = os.path.join(output_path, "data")
+        ensure_file_downloaded(source_url=URL, target_path=data_path, unpack=True)
+
+        dataset: Dict[str, pd.DataFrame] = {}
+        for split in self.splits.keys():
+            target_path_file = os.path.join(data_path, "data", f"{split}.jsonl")
+            df = pd.read_json(target_path_file, lines=True)
+            df = df[df["category"] != "q"]  # Drop instances with the "question" label
+            if split == "test":
+                dataset[split] = df.groupby("category", group_keys=False).apply(
+                    lambda x: x.sample(frac=1000 / len(df), random_state=4183)
+                )
+            else:
+                dataset[split] = df
+        return dataset
+
+    def get_instances(self, output_path) -> List[Instance]:
+        dataset = self.download_dataset(output_path)
+        outputs = []
+        for split in self.splits.keys():
+            data = dataset[split]
+            for _, row in data.iterrows():
+                input = Input(row["texts"].strip())
+                output = Output(text=self.sentiment2label[row["category"]])
+                references = [
+                    Reference(output, tags=[CORRECT_TAG]),
+                ]
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# 2.4 Tamil: IndicSentiment
+class IndicSentimentScenario(Scenario):
+    """
+    IndicSentiment is a sentiment analysis scenario for 10 Indic languages. The data consists of
+    product reviews written in English that were then translated by native speakers of the
+    respective languages, resulting in a parallel dataset across the 10 languages.
+
+    Only the Tamil subset of the dataset is used for this scenario. Labels are positive or negative.
+
+    The models are prompted using the following format:
+
+        பின்வரும் வாக்கியத்தில் வெளிப்படுத்தப்படும் உணர்வு எது?
+        ஒரு சொல்லில் மட்டும் பதிலளிக்கவும்:
+        - நேர்மறை
+        - எதிர்மறை
+
+        வாக்கியம்: <text>
+        பதில்:
+
+        ...
+
+        வாக்கியம்: <text>
+        பதில்: <answer>
+
+    Target completion:
+        <sentiment> (<sentiment>:positive or negative)
+
+    @inproceedings{doddapaneni-etal-2023-towards,
+        title = "Towards Leaving No {I}ndic Language Behind: Building Monolingual Corpora, Benchmark and Models for
+            {I}ndic Languages",
+        author = "Doddapaneni, Sumanth  and
+            Aralikatte, Rahul  and
+            Ramesh, Gowtham  and
+            Goyal, Shreya  and
+            Khapra, Mitesh M.  and
+            Kunchukuttan, Anoop  and
+            Kumar, Pratyush",
+        editor = "Rogers, Anna  and
+            Boyd-Graber, Jordan  and
+            Okazaki, Naoaki",
+        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.693",
+        doi = "10.18653/v1/2023.acl-long.693",
+        pages = "12402--12426",
+    }
+    """
+
+    name = "indicsentiment"
+    description = "IndicSentiment Tamil sentiment analysis task"
+    tags = ["sentiment_analysis"]
+
+    def __init__(self):
+        super().__init__()
+        self.splits = {"validation": TRAIN_SPLIT, "test": TEST_SPLIT}
+        self.sentiment2label = {
+            "Positive": "நேர்மறை",
+            "Negative": "எதிர்மறை",
+        }
+
+    def get_instances(self, output_path) -> List[Instance]:
+        dataset = datasets.load_dataset(
+            "ai4bharat/IndicSentiment",
+            "translation-ta",
+            revision="dc8f3f66886531c6897fedffca1e938a68fc5013",
+            trust_remote_code=True,
+        )
+
+        outputs = []
+        for split in self.splits.keys():
+            data = dataset[split].to_pandas()
+            # Current version on HuggingFace datasets has 2 instances without labels across all languages.
+            # Confirmed with first author that the labels for these instances should be Positive.
+            data["LABEL"] = data["LABEL"].fillna("Positive")
+            for _, row in data.iterrows():
+                input = Input(row["INDIC REVIEW"].strip())
+                output = Output(text=self.sentiment2label[row["LABEL"]])
+                references = [
+                    Reference(output, tags=[CORRECT_TAG]),
+                ]
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# 3. Toxicity Detection/Classification
+# 3.1 Indonesian: Multi-Label Hate Speech Detection
+class MLHSDScenario(Scenario):
+    """
+    Multi-Label Hate Speech and Abusive Language Detection (MLHSD) is an Indonesian toxicity
+    classification scenario.  The data is obtained from Twitter and PII have been anonymized to
+    USER and URL.
+
+    The original dataset was used for a multi-label classification task, but it has been repurposed
+    as a multi-class classification task to be more aligned with the task for other languages.
+    The mapping is done as follows:
+    - Clean: No abusive language or hate speech labels
+    - Abusive: Only abusive language label but no hate speech labels
+    - Hate: As long as one hate speech label is present
+
+    The models are prompted using the following format:
+
+        Anda adalah pendeteksi ujaran kebencian. Definisi dari labelnya adalah sebagai berikut:
+        Bersih: Tidak ada ujaran kebencian.
+        Kasar: Ada ujaran kebencian dan kata-kata kasar, namun tidak menyerang pihak tertentu.
+        Benci: Ada ujaran kebencian atau serangan langsung terhadap pihak tertentu.
+        Berdasarkan definisi labelnya, klasifikasikan kalimat berikut ini dengan satu kata saja:
+        - Bersih
+        - Kasar
+        - Benci
+
+        Kalimat: <text>
+        Jawaban: <answer>
+
+        ...
+
+        Kalimat: <text>
+        Jawaban:
+
+    Target completion:
+        <answer>
+
+    @inproceedings{ibrohim-budi-2019-multi,
+        title = "Multi-label Hate Speech and Abusive Language Detection in {I}ndonesian {T}witter",
+        author = "Ibrohim, Muhammad Okky  and
+            Budi, Indra",
+        editor = "Roberts, Sarah T.  and
+            Tetreault, Joel  and
+            Prabhakaran, Vinodkumar  and
+            Waseem, Zeerak",
+        booktitle = "Proceedings of the Third Workshop on Abusive Language Online",
+        month = aug,
+        year = "2019",
+        address = "Florence, Italy",
+        publisher = "Association for Computational Linguistics",
+        url = "https://aclanthology.org/W19-3506",
+        doi = "10.18653/v1/W19-3506",
+        pages = "46--57",
+    }
+    """
+
+    name = "mlhsd"
+    description = (
+        "Multi-Label Hate Speech and Abusive Language Detection (MLHSD) Indonesian toxicity classification task"
+    )
+    tags = ["toxicity_detection"]
+
+    def __init__(self):
+        super().__init__()
+        self.splits = {"train": TRAIN_SPLIT, "test": TEST_SPLIT}
+
+    def download_dataset(self, output_path: str):
+        BASE_URL = "https://raw.githubusercontent.com/okkyibrohim/"
+        URL = f"{BASE_URL}id-multi-label-hate-speech-and-abusive-language-detection/master/re_dataset.csv"
+        target_path_file = os.path.join(output_path, "mlhsd")
+        ensure_file_downloaded(source_url=URL, target_path=target_path_file)
+        df = pd.read_csv(target_path_file, encoding="ISO-8859-1")
+
+        # Map multi-label task to multi-class task
+        df["label"] = df.apply(lambda x: self.get_label(x), axis=1)
+
+        df_test = df.groupby("label", group_keys=False).apply(
+            lambda x: x.sample(frac=1000 / len(df), random_state=7123)
+        )
+
+        # In-context examples to be drawn from remaining examples (since there is no train/dev data)
+        df_train = df[~df.index.isin(df_test.index)]
+        dataset = {
+            "train": df_train,
+            "test": df_test,
+        }
+        return dataset
+
+    def get_label(self, row) -> str:
+        if int(row["HS"]) == 1:
+            return "Benci"
+        elif int(row["Abusive"]) == 1:
+            return "Kasar"
+        else:
+            return "Bersih"
+
+    def get_instances(self, output_path) -> List[Instance]:
+        dataset = self.download_dataset(output_path)
+        outputs = []
+        for split in self.splits.keys():
+            data = dataset[split]
+            for _, row in data.iterrows():
+                input = Input(row["Tweet"].strip())
+                output = Output(text=row["label"])
+                references = [
+                    Reference(output, tags=[CORRECT_TAG]),
+                ]
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# 3.2 Vietnamese: ViHSD
+class ViHSDScenario(Scenario):
+    """
+    ViHSD is a Vietnamese toxicity classification scenario. The data is obtained from social media.
+    The labels are Clean, Offensive and Hate.
+
+    The models are prompted using the following format:
+
+        Bạn là máy phát hiện phát ngôn thù ghét. Các nhãn được định nghĩa như sau:
+        Sạch: Không quấy rối.
+        Công kích: Bao gồm quấy rối và thậm chí chửi thề, nhưng không tấn công bất kì đối tượng cụ thể nào.
+        Thù ghét: Trực tiếp quấy rối hay lăng mạ một đối tượng cụ thể.
+        Với các định nghĩa của nhãn, hãy phân loại câu dưới đây với một từ duy nhất:
+        - Sạch
+        - Công kích
+        - Thù ghét
+
+
+        Câu văn: <text>
+        Câu trả lời: <toxicity>
+
+        ...
+
+        Câu văn: <text>
+        Câu trả lời:
+
+    Target completion:
+        <toxicity>
+
+    @InProceedings{10.1007/978-3-030-79457-6_35,
+        author="Luu, Son T.
+            and Nguyen, Kiet Van
+            and Nguyen, Ngan Luu-Thuy",
+        editor="Fujita, Hamido
+            and Selamat, Ali
+            and Lin, Jerry Chun-Wei
+            and Ali, Moonis",
+        title="A Large-Scale Dataset for Hate Speech Detection on Vietnamese Social Media Texts",
+        booktitle="Advances and Trends in Artificial Intelligence. Artificial Intelligence Practices",
+        year="2021",
+        publisher="Springer International Publishing",
+        address="Cham",
+        pages="415--426",
+        isbn="978-3-030-79457-6",
+        url="https://link.springer.com/chapter/10.1007/978-3-030-79457-6_35",
+    }
+    """
+
+    name = "vihsd"
+    description = "ViHSD Vietnamese toxicity classification task"
+    tags = ["toxicity_detection"]
+
+    def __init__(self):
+        super().__init__()
+        self.splits = {"train": TRAIN_SPLIT, "test": TEST_SPLIT}
+        self.id2label = {
+            0: "Sạch",
+            1: "Công kích",
+            2: "Thù ghét",
+        }
+
+    def download_dataset(self, output_path: str):
+        URL = "https://raw.githubusercontent.com/sonlam1102/vihsd/main/data/vihsd.zip"
+        data_path = os.path.join(output_path, "data")
+        ensure_file_downloaded(source_url=URL, target_path=data_path, unpack=True)
+
+        dataset: Dict[str, pd.DataFrame] = {}
+        for split in self.splits.keys():
+            target_path_file = os.path.join(data_path, "vihsd", f"{split}.csv")
+            df = pd.read_csv(target_path_file)
+            data = df.groupby("label_id", group_keys=False).apply(
+                lambda x: x.sample(frac=1000 / len(df), random_state=4878)
+            )
+            dataset[split] = data
+        return dataset
+
+    def get_instances(self, output_path) -> List[Instance]:
+        dataset = self.download_dataset(output_path)
+        outputs = []
+        for split in self.splits.keys():
+            data = dataset[split]
+            for _, row in data.iterrows():
+                input = Input(str(row["free_text"]).strip())
+                output = Output(text=self.id2label[int(row["label_id"])])
+                references = [
+                    Reference(output, tags=[CORRECT_TAG]),
+                ]
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# 3.3 Thai: Thai Toxicity Tweets
+class ThaiToxicityTweetsScenario(Scenario):
+    """
+    Thai Toxicity Tweets is a Thai toxicity detection scenario. The data is obtained from Twitter.
+    Instances with no labels or had "TWEET_NOT_FOUND" as the text were dropped from the dataset.
+    The labels are either Y (the text is toxic) or N (the text is clean).
+
+    The models are prompted using the following format:
+
+        คุณเป็นเครื่องตรวจจับข้อความที่เป็นพิษ
+        ข้อความเป็นพิษถ้ามันทำร้าย สร้างความเสียหาย หรือสื่อเจตนาเชิงลบถึงบุคคลหรือกลุ่มบุคคล
+        และคุณควรคำนึงถึงการประชดประชันด้วย
+        เมื่อได้รับข้อความ ให้ตอบด้วยตัวอักษร Y หากเป็นพิษ และ N หากไม่เป็นพิษ
+
+        ข้อความ: <text>
+        คำตอบ: <toxicity>
+
+        ...
+
+        ข้อความ: <text>
+        คำตอบ:
+
+    Target completion:
+        <toxicity>
+
+    @inproceedings{sirihattasak2018annotation,
+        title={Annotation and classification of toxicity for Thai Twitter},
+        author={Sirihattasak, Sugan and Komachi, Mamoru and Ishikawa, Hiroshi},
+        booktitle={TA-COS 2018: 2nd Workshop on Text Analytics for Cybersecurity and Online Safety},
+        pages={1},
+        year={2018},
+        url={http://www.lrec-conf.org/workshops/lrec2018/W32/pdf/1_W32.pdf},
+    }
+    """
+
+    name = "thaitoxicitytweets"
+    description = "Thai Toxicity Tweets toxicity detection task"
+    tags = ["toxicity_detection"]
+
+    def __init__(self):
+        super().__init__()
+        self.splits = {"train": TRAIN_SPLIT, "test": TEST_SPLIT}
+        self.id2label = {
+            0: "N",
+            1: "Y",
+        }
+
+    def get_instances(self, output_path) -> List[Instance]:
+        dataset = datasets.load_dataset(
+            "tmu-nlp/thai_toxicity_tweet",
+            split="train",
+            revision="aa021e41d0ee6dbee2975fbed620ec8c586bdaf6",
+            trust_remote_code=True,
+        )
+        df = dataset.to_pandas()
+
+        # Drop instances where there are no labels or text is "TWEET_NOT_FOUND"
+        df = df[df["tweet_text"].str.len() > 0]
+        df = df[df["tweet_text"] != "TWEET_NOT_FOUND"]
+
+        df_test = df.groupby("is_toxic", group_keys=False).apply(
+            lambda x: x.sample(frac=1000 / len(df), random_state=4156)
+        )
+
+        # In-context examples to be drawn from remaining examples (since there is no train/dev data)
+        df_train = df[~df.index.isin(df_test.index)]
+
+        dataset = {
+            "train": df_train,
+            "test": df_test,
+        }
+
+        outputs = []
+        for split in self.splits.keys():
+            data = dataset[split]
+            for _, row in data.iterrows():
+                input = Input(row["tweet_text"].strip())
+                output = Output(text=self.id2label[int(row["is_toxic"])])
+                references = [
+                    Reference(output, tags=[CORRECT_TAG]),
+                ]
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# B. Natural Language Generation
+#   1. Machine Translation
+
+
+# 1. Machine Translation: FLoRes-200
+class FloresScenario(Scenario):
+    """
+    FLoRes-200 is a machine translation scenario for 200+ languages. The data is obtained from English Wikimedia
+    projects (Wikivoyage, Wikijunior and Wikinews), and professionally translated across 200+ languages to obtain a
+    parallel dataset.
+
+    Only the English, Indonesian, Vietnamese, Thai and Tamil subsets are used in this scenario. Both directions
+    (in and out of English) for each Southeast Asian language are included in the scenario.
+
+    The models are prompted using the following general format:
+
+        Translate the following text into <language> language.
+
+        Text: <text>
+        Translation: <translation>
+
+        ...
+
+        Text: <text>
+        Translation:
+
+    Target completion:
+        <translation>
+
+    @article{nllb2022,
+        author = {NLLB Team, Marta R. Costa-jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield,
+            Kevin Heffernan, Elahe Kalbassi,  Janice Lam, Daniel Licht, Jean Maillard, Anna Sun, Skyler Wang,
+            Guillaume Wenzek, Al Youngblood, Bapi Akula, Loic Barrault, Gabriel Mejia Gonzalez, Prangthip Hansanti,
+            John Hoffman, Semarley Jarrett, Kaushik Ram Sadagopan, Dirk Rowe, Shannon Spruit, Chau Tran,
+            Pierre Andrews, Necip Fazil Ayan, Shruti Bhosale, Sergey Edunov, Angela Fan, Cynthia Gao,
+            Vedanuj Goswami, Francisco Guzmán, Philipp Koehn, Alexandre Mourachko, Christophe Ropers,
+            Safiyyah Saleem, Holger Schwenk, Jeff Wang
+        },
+        title = {No Language Left Behind: Scaling Human-Centered Machine Translation},
+        year = {2022},
+        url = {https://research.facebook.com/publications/no-language-left-behind/},
+    }
+
+    """
+
+    name = "flores"
+    description = "FLoRes-200 machine translation task"
+    tags = ["machine_translation"]
+
+    def __init__(self, pair: str):
+        super().__init__()
+        self.pair = pair
+        self.source = pair.split("_")[0]
+        self.target = pair.split("_")[1]
+
+        self.splits = {"dev": TRAIN_SPLIT, "devtest": TEST_SPLIT}
+
+        self.languages = {
+            "en": "eng_Latn",
+            "id": "ind_Latn",
+            "vi": "vie_Latn",
+            "th": "tha_Thai",
+            "ta": "tam_Taml",
+        }
+
+        if self.source not in self.languages.keys() or self.target not in self.languages.keys():
+            raise Exception(f"Unsupported language/s - supported languages are {self.languages.keys()}")
+
+    def get_instances(self, output_path) -> List[Instance]:
+        source_dataset = datasets.load_dataset(
+            "facebook/flores",
+            self.languages[self.source],
+            revision="2db78afdeaccaedc3b33a95442a4e55766887e17",
+            trust_remote_code=True,
+        )
+        target_dataset = datasets.load_dataset(
+            "facebook/flores",
+            self.languages[self.target],
+            revision="2db78afdeaccaedc3b33a95442a4e55766887e17",
+            trust_remote_code=True,
+        )
+
+        outputs = []
+        for split in self.splits.keys():
+            source_df = source_dataset[split].to_pandas()
+            target_df = target_dataset[split].to_pandas()
+            data = source_df.join(target_df, lsuffix="_source", rsuffix="_target")
+            for _, row in data.iterrows():
+                input = Input(row["sentence_source"].strip())
+                output = Output(row["sentence_target"].strip())
+                references = [
+                    Reference(output, tags=[CORRECT_TAG]),
+                ]
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# C. Natural Language Reasoning
+#   1. Natural Language Inference
+#   2. Causal Reasoning
+
+
+# 1. Natural Language Inference
+# 1.1 Indonesian: IndoNLI
+class IndoNLIScenario(Scenario):
+    """
+    IndoNLI is an Indonesian Natural Language Inference (NLI) scenario. The data is sourced from Wikipedia, news,
+    and web articles. Native speakers use premise text from these sources and write hypothesis sentences for each
+    NLI label. The labels are entailment, contradiction, or neutral.
+
+    The models are prompted using the following format:
+
+        Anda akan diberikan dua kalimat, X dan Y.
+        Tentukan mana dari pernyataan berikut ini yang paling sesuai untuk kalimat X dan Y.
+        A: Kalau X benar, maka Y juga harus benar.
+        B: X bertentangan dengan Y.
+        C: Ketika X benar, Y mungkin benar atau mungkin tidak benar.
+        Jawablah dengan satu huruf saja, A, B atau C.
+
+        X: <sentence1>
+        Y: <sentence2>
+        Jawaban: <entailment>
+
+        ...
+
+        X: <sentence1>
+        Y: <sentence2>
+        Jawaban:
+
+    Target completion:
+        <entailment>
+
+    @inproceedings{mahendra-etal-2021-indonli,
+        title = "{I}ndo{NLI}: A Natural Language Inference Dataset for {I}ndonesian",
+        author = "Mahendra, Rahmad and Aji, Alham Fikri and Louvan, Samuel and Rahman, Fahrurrozi and Vania, Clara",
+        booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
+        month = nov,
+        year = "2021",
+        address = "Online and Punta Cana, Dominican Republic",
+        publisher = "Association for Computational Linguistics",
+        url = "https://aclanthology.org/2021.emnlp-main.821",
+        pages = "10511--10527",
+    }
+    """
+
+    name = "indonli"
+    description = "IndoNLI Indonesian Natural Language Inference task"
+    tags = ["natural_language_inference"]
+
+    def __init__(self):
+        super().__init__()
+        self.splits = {
+            "train": TRAIN_SPLIT,
+            "test": TEST_SPLIT,
+        }
+        self.id2label = {"e": "A", "c": "B", "n": "C"}
+
+    def download_dataset(self, output_path: str):
+        URLS = {
+            "train": "https://raw.githubusercontent.com/ir-nlp-csui/indonli/main/data/indonli/train.jsonl",
+            "test": "https://raw.githubusercontent.com/ir-nlp-csui/indonli/main/data/indonli/test_lay.jsonl",
+        }
+
+        dataset: Dict[str, pd.DataFrame] = {}
+        for split in self.splits.keys():
+            target_path_file = os.path.join(output_path, split)
+            ensure_file_downloaded(source_url=URLS[split], target_path=target_path_file)
+            df = pd.read_json(target_path_file, lines=True)
+            if split == "test":
+                dataset[split] = df.groupby("label", group_keys=False).apply(
+                    lambda x: x.sample(frac=1000 / len(df), random_state=4685)
+                )
+            else:
+                dataset[split] = df
+        return dataset
+
+    def get_instances(self, output_path) -> List[Instance]:
+        dataset = self.download_dataset(output_path)
+        outputs = []
+        for split in self.splits.keys():
+            data = dataset[split]
+            for _, row in data.iterrows():
+                passage = "X: " + row["premise"].strip() + "\nY: " + row["hypothesis"].strip()
+                input = Input(passage)
+                output = Output(self.id2label[row["label"]])
+                references = [
+                    Reference(output, tags=[CORRECT_TAG]),
+                ]
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# 1.2 Vietnamese & Thai: XNLI
+class XNLIScenario(Scenario):
+    """
+    XNLI is a Natural Language Inference scenario for 15 languages. The data was constructed following the
+    MultiNLI crowdsourcing procedure to obtain English data, which was then professionally translated across
+    14 other languages. Labels are entailment, neutral, or contradiction.
+
+    The models are prompted using the following general format:
+
+        You will be given two sentences, X and Y.
+        Determine which of the following statements applies to sentences X and Y the best.
+        A: If X is true, Y must be true.
+        B: X contradicts Y.
+        C: When X is true, Y may or may not be true.
+        Answer strictly with a single letter A, B or C.
+
+        X: <sentence1>
+        Y: <sentence2>
+        Answer: <entailment>
+
+        ...
+
+        X: <sentence1>
+        Y: <sentence2>
+        Answer:
+
+    Target completion:
+        <entailment>
+
+    @inproceedings{conneau-etal-2018-xnli,
+        title = "{XNLI}: Evaluating Cross-lingual Sentence Representations",
+        author = "Conneau, Alexis  and
+            Rinott, Ruty  and
+            Lample, Guillaume  and
+            Williams, Adina  and
+            Bowman, Samuel  and
+            Schwenk, Holger  and
+            Stoyanov, Veselin",
+        editor = "Riloff, Ellen  and
+            Chiang, David  and
+            Hockenmaier, Julia  and
+            Tsujii, Jun{'}ichi",
+        booktitle = "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing",
+        month = oct # "-" # nov,
+        year = "2018",
+        address = "Brussels, Belgium",
+        publisher = "Association for Computational Linguistics",
+        url = "https://aclanthology.org/D18-1269",
+        doi = "10.18653/v1/D18-1269",
+        pages = "2475--2485",
+    }
+    """
+
+    name = "xnli"
+    description = "XNLI Natural Language Inference task"
+    tags = ["natural_language_inference"]
+
+    def __init__(self, language: str):
+        super().__init__()
+        self.language = language
+        self.splits = {
+            "validation": TRAIN_SPLIT,
+            "test": TEST_SPLIT,
+        }
+        self.id2label = {0: "A", 2: "B", 1: "C"}
+        self.supported_languages = ["th", "vi"]
+        if self.language not in self.supported_languages:
+            raise Exception(f"{self.language} not supported. Supported languages are {self.supported_languages}.")
+
+    def get_instances(self, output_path) -> List[Instance]:
+        dataset = datasets.load_dataset("xnli", self.language)
+        outputs = []
+        for split in self.splits.keys():
+            df = dataset[split].to_pandas()
+            if split == "validation":
+                data = df
+            else:
+                # This produces 999 instances
+                data = df.groupby("label", group_keys=False).apply(
+                    lambda x: x.sample(frac=1000 / len(df), random_state=4156)
+                )
+
+                # Add 1 neutral instance from remaining instances to the test data to make 1000 in total
+                remainder = df[~df.index.isin(data.index)]
+                neutral_instance = remainder[remainder["label"] == 1].iloc[0].to_frame().transpose()
+                data = pd.concat([data, neutral_instance], axis=0, ignore_index=True)
+            for _, row in data.iterrows():
+                passage = "X: " + row["premise"].strip() + "\nY: " + row["hypothesis"].strip()
+                input = Input(passage)
+                output = Output(self.id2label[int(row["label"])])
+                references = [
+                    Reference(output, tags=[CORRECT_TAG]),
+                ]
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# 1.3 Tamil: IndicXNLI
+class IndicXNLIScenario(Scenario):
+    """
+    IndicXNLI is a Natural Language Inference scenario for 11 Indic languages. The data was
+    automatically translated from the English XNLI dataset into 11 Indic languages using
+    IndicTrans (Ramesh et al., 2021).
+
+    Only the Tamil subset of the data is used in this scenario. The labels are
+    entailment, contradiction and neutral.
+
+    The models are prompted using the following format:
+
+        உங்களுக்கு இரண்டு வாக்கியங்கள், X மற்றும் Y, தரப்படும்.
+        பின்வரும் கூற்றுகளில் எது X மற்றும் Y வாக்கியங்களுடன் மிகப் பொருந்துகிறது எனக் கண்டறியவும்.
+        A: X உண்மை என்றால் Y உம் உண்மையாக இருக்க வேண்டும்.
+        B: X உம் Y உம் முரண்படுகின்றன.
+        C: X உண்மையாக இருக்கும்போது Y உண்மையாக இருக்கலாம் அல்லது இல்லாமல் இருக்கலாம்.
+        A அல்லது B அல்லது C என்ற ஒறே எழுத்தில் மட்டும் பதிலளிக்கவும்.
+
+        X: <premise>
+        Y: <hypothesis>
+        பதில்: <entailment>
+
+        ...
+
+        X: <premise>
+        Y: <hypothesis>
+        பதில்:
+
+    Target completion:
+        <entailment>
+
+    @inproceedings{aggarwal-etal-2022-indicxnli,
+        title = "{I}ndic{XNLI}: Evaluating Multilingual Inference for {I}ndian Languages",
+        author = "Aggarwal, Divyanshu  and
+            Gupta, Vivek  and
+            Kunchukuttan, Anoop",
+        editor = "Goldberg, Yoav  and
+            Kozareva, Zornitsa  and
+            Zhang, Yue",
+        booktitle = "Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing",
+        month = dec,
+        year = "2022",
+        address = "Abu Dhabi, United Arab Emirates",
+        publisher = "Association for Computational Linguistics",
+        url = "https://aclanthology.org/2022.emnlp-main.755",
+        doi = "10.18653/v1/2022.emnlp-main.755",
+        pages = "10994--11006",
+    }
+    """
+
+    name = "indicxnli"
+    description = "IndicXNLI Natural Language Inference task"
+    tags = ["natural_language_inference"]
+
+    def __init__(self):
+        super().__init__()
+        self.splits = {
+            "validation": TRAIN_SPLIT,
+            "test": TEST_SPLIT,
+        }
+        self.id2label = {0: "A", 2: "B", 1: "C"}
+
+    def get_instances(self, output_path) -> List[Instance]:
+        dataset = datasets.load_dataset("Divyanshu/indicxnli", "ta")
+
+        outputs = []
+        for split in self.splits.keys():
+            df = dataset[split].to_pandas()
+            if split == "validation":
+                data = df
+            else:
+                # This produces 999 instances
+                data = df.groupby("label", group_keys=False).apply(
+                    lambda x: x.sample(frac=1000 / len(df), random_state=4156)
+                )
+
+                # Add 1 neutral instance from remaining instances to the test data to make 1000 in total
+                remainder = df[~df.index.isin(data.index)]
+                neutral_instance = remainder[remainder["label"] == 2].iloc[0].to_frame().transpose()
+                data = pd.concat([data, neutral_instance], axis=0, ignore_index=True)
+            for _, row in data.iterrows():
+                passage = "X: " + row["premise"].strip() + "\nY: " + row["hypothesis"].strip()
+                input = Input(passage)
+                output = Output(text=self.id2label[row["label"]])
+                references = [
+                    Reference(output, tags=[CORRECT_TAG]),
+                ]
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# 2. Causal Reasoning: XCOPA
+class XCOPAScenario(Scenario):
+    """
+    XCOPA is a commonsense causal reasoning scenario for 11 languages. The data is sourced from the English
+    COPA dataset and professionally translated across 11 languages to create a parallel dataset.
+
+    Only the Indonesian, Vietnamese, Thai and Tamil subsets were used for this scenario. Each instance consists of
+    a premise and two sentences. The system under test needs to determine which of the two sentences is more likely
+    to be the cause/effect of the premise. Whether the cause or the effect is asked for differs from instance to
+    instance. Although there should be an equal number of instances asking for the cause and for the effect, it was
+    found in the BHASA paper (Leong et al., 2023) that this was not the case for Indonesian and Thai. The
+    cause/effect label is fixed in this scenario by harmonizing the labels across the four languages based on the
+    Tamil subset as the reference.
+
+    The models are prompted using the following general format:
+
+        Based on the following situation, which of the following choices is most likely to be its {cause/effect}?
+        Answer only with a single letter A or B.
+
+        Situation: <premise>
+        A: <choice1>
+        B: <choice2>
+        Answer: <answer>
+
+        ...
+
+        Situation: <premise>
+        A: <choice1>
+        B: <choice2>
+        Answer:
+
+    Target completion:
+        <answer>
+
+    @article{ponti2020xcopa,
+    title={{XCOPA: A} Multilingual Dataset for Causal Commonsense Reasoning},
+    author={Edoardo M. Ponti, Goran Glava
+    {s}, Olga Majewska, Qianchu Liu, Ivan Vuli'{c} and Anna Korhonen},
+    journal={arXiv preprint},
+    year={2020},
+    url={https://ducdauge.github.io/files/xcopa.pdf}
+    }
+
+    @inproceedings{roemmele2011choice,
+    title={Choice of plausible alternatives: An evaluation of commonsense causal reasoning},
+    author={Roemmele, Melissa and Bejan, Cosmin Adrian and Gordon, Andrew S},
+    booktitle={2011 AAAI Spring Symposium Series},
+    year={2011},
+    url={https://people.ict.usc.edu/~gordon/publications/AAAI-SPRING11A.PDF},
+    }
+    """
+
+    name = "xcopa"
+    description = "XCOPA causal reasoning task"
+    tags = ["causal_reasoning"]
+
+    def __init__(self, language: str):
+        super().__init__()
+        self.language = language
+        self.splits = {
+            "validation": TRAIN_SPLIT,
+            "test": TEST_SPLIT,
+        }
+        self.id2label = {
+            0: "A",
+            1: "B",
+        }
+        self.language_to_prompt_components = {
+            "id": {
+                "cause": "sebab",
+                "effect": "akibat",
+                "instruction1": "Berdasarkan situasi di atas, mana dari pilihan-pilihan berikut ini yang lebih "
+                "mungkin menjadi {}?",
+                "instruction2": "Jawablah dengan satu huruf saja, A atau B.",
+            },
+            "ta": {
+                "cause": "காரணமாக",
+                "effect": "விளைவாக",
+                "instruction1": "பின்வரும் வாக்கியங்களில் பெரும்பாலும் எது தரப்பட்ட சூழ்நிலைக்குரிய {} இருக்கும்?",
+                "instruction2": "A அல்லது B என்ற ஒறே எழுத்தில் மட்டும் பதிலளிக்கவும்.",
+            },
+            "th": {
+                "cause": "สาเหตุ",
+                "effect": "ผล",
+                "instruction1": "เมื่อพิจารณาจากสถานการณ์นี้ ตัวเลือกใดต่อไปนี้น่าจะเป็น{}มากกว่ากัน?",
+                "instruction2": "กรุณาตอบด้วยตัวอักษร A หรือ B ตัวเดียวเท่านั้น",
+            },
+            "vi": {
+                "cause": "nguyên nhân",
+                "effect": "kết quả",
+                "instruction1": "Với tình huống trên, lựa chọn nào dưới đây có khả năng cao là {} của nó hơn?",
+                "instruction2": "Trả lời với một chữ cái duy nhất A hoặc B.",
+            },
+        }
+        if self.language not in self.language_to_prompt_components.keys():
+            raise Exception(
+                f"{self.language} not supported. Supported languages are {self.language_to_prompt_components.keys()}."
+            )
+        else:
+            self.prompt_components = self.language_to_prompt_components[self.language]
+
+    def get_instances(self, output_path) -> List[Instance]:
+        language_dataset = datasets.load_dataset("xcopa", self.language)
+        tamil_dataset = datasets.load_dataset("xcopa", "ta")
+
+        outputs = []
+        for split in self.splits.keys():
+            language_df = language_dataset[split].to_pandas()
+            tamil_df = tamil_dataset[split].to_pandas()
+            data = pd.merge(
+                language_df, tamil_df[["question", "idx"]], on="idx"
+            )  # Use the Tamil split's question column
+            for _, row in data.iterrows():
+                instruction1 = self.prompt_components["instruction1"].format(self.prompt_components[row["question_y"]])
+                passage = "{premise}\n{instruction1}\nA: {choice1}\nB: {choice2}\n{instruction2}".format(
+                    premise=row["premise"].strip(),
+                    instruction1=instruction1,
+                    choice1=row["choice1"].strip(),
+                    choice2=row["choice2"].strip(),
+                    instruction2=self.prompt_components["instruction2"],
+                )
+                input = Input(passage)
+                output = Output(self.id2label[int(row["label"])])
+                references = [
+                    Reference(output, tags=[CORRECT_TAG]),
+                ]
+                instance = Instance(input=input, references=references, split=self.splits[split])
+                outputs.append(instance)
+        return outputs
+
+
+# 1. Syntax: LINDSEA Minimal Pairs
+class LINDSEASyntaxMinimalPairsScenario(Scenario):
+    """
+    The LINDSEA Minimal Pairs dataset is a linguistic diagnostic scenario targeting syntactic phenomena.
+    The data is manually handcrafted by linguists and native speakers and verified through multiple rounds
+    of quality control. The high-level categories tested for include morphology, argument structure,
+    filler-gap dependencies, as well as negative polarity items and negation.
+
+    The test is designed as a minimal pair, with a pair of sentences that differ minimally from each other
+    and which exemplify a specific syntactic phenomenon. The system under test needs to determine which
+    sentence of the pair is more acceptable.
+
+    The models are prompted using the following general format:
+
+        Which sentence is more acceptable?
+        Answer only with a single letter A or B.
+        <sentence>
+
+    Target completion:
+        <sentence>
+
+    @misc{leong2023bhasa,
+        title={BHASA: A Holistic Southeast Asian Linguistic and Cultural Evaluation Suite for Large Language Models},
+        author={Wei Qi Leong
+            and Jian Gang Ngui
+            and Yosephine Susanto
+            and Hamsawardhini Rengarajan
+            and Kengatharaiyer Sarveswaran
+            and William Chandra Tjhi
+        },
+        year={2023},
+        eprint={2309.06085},
+        archivePrefix={arXiv},
+        primaryClass={cs.CL},
+        url={https://arxiv.org/abs/2309.06085},
+    }
+    """
+
+    name = "lindsea_minimal_pairs"
+    description = "LINDSEA minimal pairs task"
+    tags = ["linguistic_diagnostic", "syntax", "minimal_pairs"]
+
+    def __init__(self, method: str, language: str):
+        super().__init__()
+        self.method = method
+        self.language = language
+        self.language_to_prompt_components = {
+            "id": {
+                "instructions": "Kalimat mana yang lebih mungkin?",
+                "output_prefix": "Jawablah dengan satu huruf saja, A atau B.",
+            }
+        }
+        if self.language not in self.language_to_prompt_components.keys():
+            raise Exception(
+                f"{self.language} not supported. Supported languages are {self.language_to_prompt_components.keys()}."
+            )
+        else:
+            self.prompt_components = self.language_to_prompt_components[self.language]
+
+    def download_dataset(self, output_path: str):
+        BASE_URL = "https://raw.githubusercontent.com/aisingapore/BHASA/main/lindsea/"
+        URLS = {
+            "npis_and_negation": f"{BASE_URL}{self.language}/syntax/NPIs_and_negation.jsonl",
+            "argument_structure": f"{BASE_URL}{self.language}/syntax/argument_structure.jsonl",
+            "filler_gap_dependencies": f"{BASE_URL}{self.language}/syntax/filler-gap_dependencies.jsonl",
+            "morphology": f"{BASE_URL}{self.language}/syntax/morphology.jsonl",
+        }
+
+        data_files = {}
+        for file in list(URLS.keys()):
+            target_path_file = os.path.join(output_path, file)
+            ensure_file_downloaded(source_url=URLS[file], target_path=target_path_file)
+            data_files[file] = pd.read_json(target_path_file, lines=True)
+        dataset = pd.concat(data_files)
+
+        return dataset
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        data = self.download_dataset(output_path)
+
+        outputs = []
+        if self.method == "mcq":
+            category_list = data["category"].value_counts().keys()
+
+            hlog("MCQ method for LINDSEA Minimal Pairs chosen. Shuffling options...")
+            for category in category_list:
+                # Fix shuffling within each category
+                random.seed(1)
+                for _, row in data[data["category"] == category].iterrows():
+                    options = [(row["correct"], 1), (row["wrong"], 2)]
+                    random.shuffle(options)
+                    options_reversed = True if options[0][1] == 2 else False
+                    instructions = self.prompt_components["instructions"]
+                    output_prefix = self.prompt_components["output_prefix"]
+                    prompt = f"{instructions}\nA: {options[0][0]}\nB: {options[1][0]}\n{output_prefix}"
+                    input = Input(text=prompt)
+                    # Determine correct option based on whether shuffling reversed the options
+                    references = [
+                        Reference(Output(text="A"), tags=[] if options_reversed else [CORRECT_TAG]),
+                        Reference(Output(text="B"), tags=[CORRECT_TAG] if options_reversed else []),
+                    ]
+                    instance = Instance(input=input, references=references, split=TEST_SPLIT)
+                    outputs.append(instance)
+
+        else:
+            for _, row in data.iterrows():
+                # No need to shuffle since we are comparing logprobs of the options separately
+                input = Input(text="")
+                references = [
+                    Reference(Output(text=row["correct"].strip()), tags=[CORRECT_TAG]),
+                    Reference(Output(text=row["wrong"].strip()), tags=[]),
+                ]
+                instance = Instance(
+                    input=input,
+                    references=references,
+                    split=TEST_SPLIT,
+                )
+                outputs.append(instance)
+        return outputs
+
+
+# 2.1 Pragmatics: LINDSEA Presuppositions
+class LINDSEAPragmaticsPresuppositionsScenario(Scenario):
+    """
+    The LINDSEA Presuppositions dataset is a linguistic diagnostic scenario targeting pragmatics.
+    The data is manually handcrafted by linguists and native speakers and verified through multiple rounds
+    of quality control.
+
+    The presuppositions dataset involves two formats: single and pair sentences.
+    For single sentence questions, the system under test needs to determine if the sentence is true/false.
+    For pair sentence questions, the system under test needs to determine whether a conclusion can be drawn
+    from another sentence.
+
+    For the single format, the models are prompted using the following general format:
+
+        Is the following statement true or false?
+        Statement: <sentence>
+        Answer only with True or False.
+
+    For the pair format, the models are prompted using the following general format:
+
+        Situation: <premise>
+        Given this situation, is the following statement true or false?
+        Statement: <hypothesis>
+        Answer only with True or False.
+
+    Target completion:
+        <answer>
+
+    @misc{leong2023bhasa,
+        title={BHASA: A Holistic Southeast Asian Linguistic and Cultural Evaluation Suite for Large Language Models},
+        author={Wei Qi Leong
+            and Jian Gang Ngui
+            and Yosephine Susanto
+            and Hamsawardhini Rengarajan
+            and Kengatharaiyer Sarveswaran
+            and William Chandra Tjhi
+        },
+        year={2023},
+        eprint={2309.06085},
+        archivePrefix={arXiv},
+        primaryClass={cs.CL}
+    }
+    """
+
+    name = "lindsea_pragmatics_presuppositions"
+    description = "LINDSEA presuppositions task"
+    tags = ["linguistic_diagnostic", "pragmatics", "presuppositions"]
+
+    def __init__(self, language: str, subset: str):
+        super().__init__()
+        self.language = language
+        self.subsets = [subset] if subset != "all" else ["single", "pair"]
+        self.language_to_prompt_components = {
+            "id": {
+                "text_noun": "Pernyataan",
+                "premise_noun": "Situasi",
+                "conclusion_noun": "Pernyataan",
+                "single_question": "Apakah pernyataan berikut ini {}?",
+                "single_instruction": "Jawablah dengan {} saja.",
+                "pair_question": "Berdasarkan situasi ini, apakah pernyataan berikut ini benar atau salah?",
+                "pair_instruction": "Jawablah dengan Benar atau Salah saja.",
+                "True": "Benar",
+                "False": "Salah",
+            },
+        }
+        if self.language not in self.language_to_prompt_components.keys():
+            raise Exception(
+                f"{self.language} not supported. Supported languages are {self.language_to_prompt_components.keys()}."
+            )
+        else:
+            self.prompt_components = self.language_to_prompt_components[self.language]
+
+    def download_dataset(self, output_path: str):
+        BASE_URL = "https://raw.githubusercontent.com/aisingapore/BHASA/main/lindsea/"
+        datasets = []
+        for subset in self.subsets:
+            URL = f"{BASE_URL}{self.language}/pragmatics/pragmatic_reasoning_{subset}.jsonl"
+            file = f"pragmatic_reasoning_{subset}.jsonl"
+            target_path_file = os.path.join(output_path, file)
+            ensure_file_downloaded(source_url=URL, target_path=target_path_file)
+            data = pd.read_json(target_path_file, lines=True)
+            data["subset"] = subset
+            data = data[data["linguistic_phenomenon"] == "presuppositions"]
+            datasets.append(data)
+        dataset = pd.concat(datasets)
+        return dataset
+
+    def get_instances(self, output_path) -> List[Instance]:
+        data = self.download_dataset(output_path)
+        outputs = []
+        for _, row in data.iterrows():
+            passage = None
+            references = []
+
+            if row["subset"] == "single":
+                question = self.prompt_components["single_question"]
+                text_noun = self.prompt_components["text_noun"]
+                instruction = self.prompt_components["single_instruction"]
+
+                passage = "{question}\{text_noun}: {text}\n{instruction}".format(
+                    question=question.format(row["question_translated"]),
+                    text_noun=text_noun,
+                    text=row["text"],
+                    instruction=instruction.format(row["choices_translated"]),
+                )
+                # Split "True or False" into ["True", "or", "False"]
+                choices = row["choices"].split()
+                choices_translated = row["choices_translated"].split()
+                label2choice = {
+                    choices[0]: choices_translated[0],
+                    choices[2]: choices_translated[2],
+                }
+                references.append(
+                    Reference(Output(text=label2choice[row["label"].strip()]), tags=[CORRECT_TAG]),
+                )
+
+            elif row["subset"] == "pair":
+                premise_noun = self.prompt_components["premise_noun"]
+                question = self.prompt_components["pair_question"]
+                conclusion_noun = self.prompt_components["conclusion_noun"]
+                instruction = self.prompt_components["pair_instruction"]
+                label = self.prompt_components[str(row["label"])]
+
+                passage = (
+                    "{premise_noun}: {premise}\n{question}\n{conclusion_noun}: {conclusion}\n{instruction}".format(
+                        premise_noun=premise_noun,
+                        premise=row["text"],
+                        question=question,
+                        conclusion_noun=conclusion_noun,
+                        conclusion=row["conclusion"],
+                        instruction=instruction,
+                    )
+                )
+
+                references.append(
+                    Reference(Output(text=label), tags=[CORRECT_TAG]),
+                )
+
+            input = Input(text=str(passage))
+            instance = Instance(
+                input=input,
+                references=references,
+                split=TEST_SPLIT,
+            )
+            outputs.append(instance)
+        return outputs
+
+
+# 2.2 Pragmatics: LINDSEA Scalar Implicatures
+class LINDSEAPragmaticsScalarImplicaturesScenario(Scenario):
+    """
+    The LINDSEA Scalar Implicatures Scenario dataset is a linguistic diagnostic scenario targeting pragmatics.
+    The data is manually handcrafted by linguists and native speakers and verified through multiple rounds
+    of quality control.
+
+    The scalar implicatures dataset involves two formats: single and pair sentences.
+    For single sentence questions, the system under test needs to determine if the sentence is true/false.
+    For pair sentence questions, the system under test needs to determine whether a conclusion can be drawn
+    from another sentence.
+
+    For the single format, the models are prompted using the following general format:
+
+        Is the following statement true or false?
+        Statement: <sentence>
+        Answer only with True or False.
+
+    For the pair format, the models are prompted using the following general format:
+
+        Situation: <premise>
+        Given this situation, is the following statement true or false?
+        Statement: <hypothesis>
+        Answer only with True or False.
+
+    Target completion:
+        <answer>
+
+    @misc{leong2023bhasa,
+        title={BHASA: A Holistic Southeast Asian Linguistic and Cultural Evaluation Suite for Large Language Models},
+        author={Wei Qi Leong
+            and Jian Gang Ngui
+            and Yosephine Susanto
+            and Hamsawardhini Rengarajan
+            and Kengatharaiyer Sarveswaran
+            and William Chandra Tjhi
+        },
+        year={2023},
+        eprint={2309.06085},
+        archivePrefix={arXiv},
+        primaryClass={cs.CL}
+    }
+    """
+
+    name = "lindsea_pragmatics_scalar_implicatures"
+    description = "LINDSEA scalar implicatures task"
+    tags = ["linguistic_diagnostic", "pragmatics", "scalar_implicatures"]
+
+    def __init__(self, language: str, subset: str):
+        super().__init__()
+        self.language = language
+        self.subsets = [subset] if subset != "all" else ["single", "pair"]
+        self.language_to_prompt_components = {
+            "id": {
+                "text_noun": "Pernyataan",
+                "premise_noun": "Situasi",
+                "conclusion_noun": "Pernyataan",
+                "single_question": "Apakah pernyataan berikut ini {}?",
+                "single_instruction": "Jawablah dengan {} saja.",
+                "pair_question": "Berdasarkan situasi ini, apakah pernyataan berikut ini benar atau salah?",
+                "pair_instruction": "Jawablah dengan Benar atau Salah saja.",
+                "True": "Benar",
+                "False": "Salah",
+            },
+        }
+        if self.language not in self.language_to_prompt_components.keys():
+            raise Exception(
+                f"{self.language} not supported. Supported languages are {self.language_to_prompt_components.keys()}."
+            )
+        else:
+            self.prompt_components = self.language_to_prompt_components[self.language]
+
+    def download_dataset(self, output_path: str):
+        BASE_URL = "https://raw.githubusercontent.com/aisingapore/BHASA/main/lindsea/"
+        datasets = []
+        for subset in self.subsets:
+            URL = f"{BASE_URL}{self.language}/pragmatics/pragmatic_reasoning_{subset}.jsonl"
+            file = f"pragmatic_reasoning_{subset}.jsonl"
+            target_path_file = os.path.join(output_path, file)
+            ensure_file_downloaded(source_url=URL, target_path=target_path_file)
+            data = pd.read_json(target_path_file, lines=True)
+            data["subset"] = subset
+            data = data[data["linguistic_phenomenon"] == "scalar_implicatures"]
+            datasets.append(data)
+        dataset = pd.concat(datasets)
+        return dataset
+
+    def get_instances(self, output_path) -> List[Instance]:
+        data = self.download_dataset(output_path)
+        outputs = []
+        for _, row in data.iterrows():
+            passage = None
+            references = []
+
+            if row["subset"] == "single":
+                question = self.prompt_components["single_question"]
+                text_noun = self.prompt_components["text_noun"]
+                instruction = self.prompt_components["single_instruction"]
+
+                passage = "{question}\{text_noun}: {text}\n{instruction}".format(
+                    question=question.format(row["question_translated"]),
+                    text_noun=text_noun,
+                    text=row["text"],
+                    instruction=instruction.format(row["choices_translated"]),
+                )
+                # Split "True or False" into ["True", "or", "False"]
+                choices = row["choices"].split()
+                choices_translated = row["choices_translated"].split()
+                label2choice = {
+                    choices[0]: choices_translated[0],
+                    choices[2]: choices_translated[2],
+                }
+                references.append(
+                    Reference(Output(text=label2choice[row["label"].strip()]), tags=[CORRECT_TAG]),
+                )
+
+            elif row["subset"] == "pair":
+                premise_noun = self.prompt_components["premise_noun"]
+                question = self.prompt_components["pair_question"]
+                conclusion_noun = self.prompt_components["conclusion_noun"]
+                instruction = self.prompt_components["pair_instruction"]
+                label = self.prompt_components[str(row["label"])]
+
+                passage = (
+                    "{premise_noun}: {premise}\n{question}\n{conclusion_noun}: {conclusion}\n{instruction}".format(
+                        premise_noun=premise_noun,
+                        premise=row["text"],
+                        question=question,
+                        conclusion_noun=conclusion_noun,
+                        conclusion=row["conclusion"],
+                        instruction=instruction,
+                    )
+                )
+
+                references.append(
+                    Reference(Output(text=label), tags=[CORRECT_TAG]),
+                )
+
+            input = Input(text=str(passage))
+            instance = Instance(
+                input=input,
+                references=references,
+                split=TEST_SPLIT,
+            )
+            outputs.append(instance)
+        return outputs