"""Reference words the target word is compared with: 9 groups x 20 words.

Rules for this list
- every word is a single token in both BERT (uncased) and GPT-2 (with a leading space). Words split into
  several pieces get blurry "average" vectors that end up close to everything, which spoils the neighbours.
- strongly ambiguous words (apple, mouse, bank, python, jaguar, rock, crane, orange, java, ...) are left out,
  so each group stands for one clear meaning.
"""

REFERENCE_WORDS = {
    "fruit & food": [
        "banana", "pear", "cherry", "grape", "peach", "lemon", "mango", "strawberry", "plum", "bread",
        "cheese", "pie", "juice", "salad", "honey", "butter", "cake", "soup", "potato", "fruit",
    ],
    "tech companies & devices": [
        "Microsoft", "Google", "Samsung", "Intel", "Sony", "Nokia", "Facebook", "IBM", "Dell", "Motorola",
        "Android", "laptop", "computer", "smartphone", "tablet", "software", "app", "website", "internet", "phone",
    ],
    "animals": [
        "dog", "cat", "horse", "lion", "tiger", "wolf", "fox", "rabbit", "deer", "snake",
        "frog", "eagle", "owl", "cow", "sheep", "elephant", "monkey", "rat", "bird", "goat",
    ],
    "nature & landscape": [
        "river", "lake", "ocean", "forest", "mountain", "valley", "hill", "beach", "island", "stone",
        "cliff", "shore", "stream", "desert", "cave", "field", "pond", "sea", "creek", "woods",
    ],
    "money & business": [
        "money", "cash", "loan", "credit", "salary", "investment", "stock", "shares", "profit", "finance",
        "account", "payment", "budget", "debt", "tax", "price", "market", "economy", "dollar", "mortgage",
    ],
    "music": [
        "guitar", "drums", "piano", "song", "melody", "concert", "album", "singer", "violin", "jazz",
        "rhythm", "orchestra", "lyrics", "trumpet", "opera", "chorus", "punk", "blues", "band", "music",
    ],
    "sports": [
        "football", "soccer", "tennis", "baseball", "basketball", "golf", "hockey", "goal", "stadium", "referee",
        "team", "player", "athlete", "coach", "marathon", "swimming", "boxing", "skiing", "cricket", "league",
    ],
    "cars & machines": [
        "car", "truck", "engine", "motor", "wheel", "tire", "brake", "tractor", "machine", "factory",
        "robot", "drill", "hammer", "Porsche", "Ferrari", "sedan", "motorcycle", "vehicle", "pump", "bus",
    ],
    "programming": [
        "code", "compiler", "function", "variable", "algorithm", "database", "programming", "developer",
        "syntax", "script", "server", "Linux", "API", "terminal", "framework", "HTML", "coding", "query",
        "array", "kernel",
    ],
}