tool.py

from smolagents.tools import Tool
import pronouncing
import json
import string
import difflib

class ParodyWordSuggestionTool(Tool):
    name = "parody_word_suggester"
    description = """Suggests rhyming funny words using CMU dictionary pronunciations.
    Returns similar-sounding words that rhyme, especially focusing on common vowel sounds."""
    inputs = {'target': {'type': 'string', 'description': 'The word you want to find rhyming alternatives for'}, 'word_list_str': {'type': 'string', 'description': 'JSON string of word list (e.g. \'["word1", "word2"]\')'}, 'min_similarity': {'type': 'string', 'description': 'Minimum similarity threshold (0.0-1.0)', 'nullable': True, 'default': '0.5'}}
    output_type = "string"
    VOWEL_REF = "AH,AX|UH|AE,EH|IY,IH|AO,AA|UW|AY,EY|OW,AO|AW,AO|OY,OW|ER,AXR"
    CONSONANT_REF = "M,N,NG|P,B|T,D|K,G|F,V|TH,DH|S,Z|SH,ZH|L,R|W,Y"

    def _get_vowel_groups(self):
        groups = []
        group_strs = self.VOWEL_REF.split("|")
        for group_str in group_strs:
            groups.append(group_str.split(","))
        return groups


    def _get_consonant_groups(self):
        groups = []
        group_strs = self.CONSONANT_REF.split("|")
        for group_str in group_strs:
            groups.append(group_str.split(","))
        return groups


    def _get_last_syllable(self, phones: list) -> tuple:
        """Extract the last syllable (vowel + remaining consonants)."""
        last_vowel_idx = -1
        last_vowel = None
        vowel_groups = self._get_vowel_groups()
    
        for i, phone in enumerate(phones):
            base_phone = phone.rstrip('012')
            for group in vowel_groups:
                if base_phone in group:
                    last_vowel_idx = i
                    last_vowel = base_phone
                    break
    
        if last_vowel_idx == -1:
            return None, []
        
        remaining = phones[last_vowel_idx + 1:]
        return last_vowel, remaining


    def _strip_stress(self, phones: list) -> list:
        result = []
        for phone in phones:
            result.append(phone.rstrip('012'))
        return result


    def _vowels_match(self, v1: str, v2: str) -> bool:
        v1 = v1.rstrip('012')
        v2 = v2.rstrip('012')
    
        if v1 == v2:
            return True
        
        vowel_groups = self._get_vowel_groups()
        for group in vowel_groups:
            if v1 in group and v2 in group:
                return True
        return False


    def _calculate_similarity(self, word1, phones1, word2, phones2):
        """Calculate similarity with heavy emphasis on rhyming."""
        from difflib import SequenceMatcher
        import pronouncing
    
        # Initialize all variables
        rhyme_score = 0.0
        string_score = 0.0
        pattern_score = 0.0
        phone_list1 = []
        phone_list2 = []
        vowel1 = None
        vowel2 = None
        end1 = []
        end2 = []
        end1_clean = []
        end2_clean = []
        matching_consonants = 0
    
        phone_list1 = phones1.split()
        phone_list2 = phones2.split()
    
        # Get last syllables
        vowel1, end1 = self._get_last_syllable(phone_list1)
        vowel2, end2 = self._get_last_syllable(phone_list2)
    
        # Calculate rhyme score (60%)
        if vowel1 and vowel2:
            # Perfect vowel match
            if vowel1.rstrip('012') == vowel2.rstrip('012'):
                rhyme_score = 1.0
            # Similar vowel match
            elif self._vowels_match(vowel1, vowel2):
                rhyme_score = 0.8
            
            # Check endings
            if end1 and end2:
                end1_clean = self._strip_stress(end1)
                end2_clean = self._strip_stress(end2)
            
                # Perfect ending match
                if end1_clean == end2_clean:
                    rhyme_score = min(1.0, rhyme_score + 0.2)
                # Partial ending match
                else:
                    consonant_groups = self._get_consonant_groups()
                    matching_consonants = 0
                    for c1, c2 in zip(end1_clean, end2_clean):
                        if c1 == c2:
                            matching_consonants += 1
                        else:
                            # Check if consonants are in same group
                            for group in consonant_groups:
                                if c1 in group and c2 in group:
                                    matching_consonants += 0.5
                                    break
                
                    if matching_consonants > 0:
                        rhyme_score = min(1.0, rhyme_score + (0.1 * matching_consonants))
    
        # String similarity (25%)
        if len(word1) > 1 and len(word2) > 1:
            end_similarity = SequenceMatcher(None, word1[1:], word2[1:]).ratio()
            string_score = end_similarity
        else:
            string_score = SequenceMatcher(None, word1, word2).ratio()
    
        # Pattern/Length score (15%)
        if len(phone_list1) == len(phone_list2):
            pattern_score = 1.0
        else:
            pattern_score = 1.0 - (abs(len(phone_list1) - len(phone_list2)) / max(len(phone_list1), len(phone_list2)))
    
        # Final weighted score
        similarity = (
            (rhyme_score * 0.60) +
            (string_score * 0.25) +
            (pattern_score * 0.15)
        )
    
        # Extra boost for exact matches minus first letter
        if len(word1) == len(word2) and word1[1:] == word2[1:]:
            similarity = min(1.0, similarity * 1.2)
    
        # Extra penalty for very different lengths
        if abs(len(word1) - len(word2)) > 2:
            similarity *= 0.7
    
        return {
            "similarity": round(similarity, 3),
            "rhyme_score": round(rhyme_score, 3),
            "string_score": round(string_score, 3),
            "pattern_score": round(pattern_score, 3),
            "details": {
                "last_vowel_match": vowel1.rstrip('012') == vowel2.rstrip('012') if vowel1 and vowel2 else False,
                "similar_vowels": self._vowels_match(vowel1, vowel2) if vowel1 and vowel2 else False,
                "ending_match": " ".join(end1_clean) == " ".join(end2_clean) if end1 and end2 else False,
                "string_length_diff": abs(len(word1) - len(word2))
            }
        }


    def forward(self, target: str, word_list_str: str, min_similarity: str = "0.5") -> str:
        import pronouncing
        import string
        import json
    
        # Initialize all variables
        target = target.lower().strip(string.punctuation)
        min_similarity = float(min_similarity)
        suggestions = []
        word_vowel = None
        word_end = []
        target_vowel = None
        target_end = []
        valid_words = []
        invalid_words = []
        target_phone_list = []
        words = []
        target_phones = ""
        word_phones = ""
        word = ""
        word_phone_list = []
        similarity_result = {}
    
        # Parse JSON string to list
        try:
            words = json.loads(word_list_str)
        except json.JSONDecodeError:
            return json.dumps({
                "error": "Invalid JSON string for word_list_str",
                "suggestions": []
            }, indent=2)
        
        # Get target pronunciation
        target_phones = pronouncing.phones_for_word(target)
        if not target_phones:
            return json.dumps({
                "error": f"Target word '{target}' not found in CMU dictionary",
                "suggestions": []
            }, indent=2)
    
        # Filter word list to only words in CMU dictionary
        valid_words = []
        invalid_words = []
        for word in words:
            word = word.lower().strip(string.punctuation)
            if pronouncing.phones_for_word(word):
                valid_words.append(word)
            else:
                invalid_words.append(word)
    
        if not valid_words:
            return json.dumps({
                "error": "No valid words found in CMU dictionary",
                "invalid_words": invalid_words,
                "suggestions": []
            }, indent=2)
    
        target_phones = target_phones[0]
        target_phone_list = target_phones.split()
        target_vowel, target_end = self._get_last_syllable(target_phone_list)
    
        # Check each word
        for word in valid_words:
            phones = pronouncing.phones_for_word(word)
            if phones:
                word_phones = phones[0]
                similarity_result = self._calculate_similarity(word, word_phones, target, target_phones)
            
                if similarity_result["similarity"] >= min_similarity:
                    word_phone_list = word_phones.split()
                    word_vowel, word_end = self._get_last_syllable(word_phone_list)
                
                    suggestions.append({
                        "word": word,
                        "similarity": similarity_result["similarity"],
                        "rhyme_score": similarity_result["rhyme_score"],
                        "string_score": similarity_result["string_score"],
                        "pattern_score": similarity_result["pattern_score"],
                        "phones": word_phones,
                        "last_vowel": word_vowel,
                        "ending": " ".join(word_end) if word_end else "",
                        "details": similarity_result["details"]
                    })
    
        # Sort by similarity score descending
        suggestions.sort(key=lambda x: x["similarity"], reverse=True)
    
        result = {
            "target": target,
            "target_phones": target_phones,
            "target_last_vowel": target_vowel,
            "target_ending": " ".join(target_end) if target_end else "",
            "invalid_words": invalid_words,
            "suggestions": suggestions
        }
    
        return json.dumps(result, indent=2)


    def __init__(self, *args, **kwargs):
        self.is_initialized = False