File size: 7,802 Bytes
2ba7d76
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
import json
from typing import Any, Dict, List

from anthropic import Anthropic


class LLMTranslationEditor:
    def __init__(self, validation_results: dict, anthropic_api_key: str):
        """Initialize with validation results and Anthropic API key.

        Args:
            validation_results (dict): Results from TranslationValidator
            anthropic_api_key (str): Anthropic API key for Claude access
        """
        self.results = validation_results
        self.client = Anthropic(api_key=anthropic_api_key)

    def edit_translation(self, source_text: str, current_translation: str,
                        terms_info: List[Dict[str, Any]]) -> Dict[str, Any]:
        """Use Claude to edit the translation considering validation results and context.

        Args:
            source_text (str): Original Tibetan text
            current_translation (str): Current English translation
            terms_info (list): Terms information from validation results

        Returns:
            Dict[str, Any]: Edited translation with analysis
        """
        # Build context for terms that need attention
        terms_context = []
        for term in terms_info:
            analysis = term['analysis']
            assessment = analysis['translation_assessment']

            if assessment['should_be_counted'] and not assessment['translated_correctly']:
                term_context = {
                    'term': term['source_term'],
                    'current': analysis['translated_as'],
                    'suggested': analysis['glossary_translation'],
                    'categories': {}
                }

                # Add category information
                for cat_name, cat_data in term['categories'].items():
                    if cat_name in analysis['matching_categories']:
                        term_context['categories'][cat_name] = {
                            'translations': cat_data.get('translations', []),
                            'definitions': cat_data.get('definitions', [])
                        }

                terms_context.append(term_context)

        if not terms_context:
            return {
                'edited_translation': current_translation,
                'modified': False,
                'reasoning': 'No terms requiring editing'
            }

        prompt = f"""You are an expert Tibetan translator. Review and improve this translation, focusing on accuracy and natural English:

Tibetan text: {source_text}
Current translation: {current_translation}

The following terms need attention:"""

        for term in terms_context:
            prompt += f"\n\nTibetan term: {term['term']}"
            prompt += f"\nCurrently translated as: {term['current']}"
            prompt += f"\nGlossary suggestion: {term['suggested']}"

            for cat_name, cat_data in term['categories'].items():
                prompt += f"\n{cat_name}:"
                if cat_data['definitions']:
                    prompt += f"\n- Definitions: {', '.join(cat_data['definitions'])}"
                if cat_data['translations']:
                    prompt += f"\n- Translations: {', '.join(cat_data['translations'])}"

        prompt += """

Please provide:
1. An improved translation that:
   - Maintains the meaning of the Tibetan text
   - Maintains the style and tone of the current translation
   - Uses appropriate technical terms from the glossary
   - Preserves any correct parts of the current translation
2. Your reasoning for the changes

Respond in JSON format:
{
  "edited_translation": "your improved translation",
  "reasoning": "explanation of changes and decisions",
  "modified": true/false
}"""

        try:
            message = self.client.messages.create(
                model="claude-3-sonnet-20240229",
                max_tokens=1000,
                temperature=0,
                messages=[{"role": "user", "content": prompt}]
            )

            # Extract JSON from response
            import re
            json_match = re.search(r'\{.*\}', message.content[0].text, re.DOTALL)
            if json_match:
                return json.loads(json_match.group())
            else:
                return {
                    'edited_translation': current_translation,
                    'modified': False,
                    'reasoning': 'Failed to parse LLM response'
                }

        except Exception as e:
            print(f"Error during LLM editing: {e}")
            return {
                'edited_translation': current_translation,
                'modified': False,
                'reasoning': f'LLM editing failed: {str(e)}'
            }

    def post_edit_translations(self) -> List[Dict[str, Any]]:
        """Process all lines and post-edit translations using LLM.

        Returns:
            List[Dict[str, Any]]: List of edited translations with analysis
        """
        edited_translations = []

        for line in self.results['lines']:
            source = line['source']
            target = line['target']
            terms = line['terms']

            if not terms:
                edited_translations.append({
                    'line_number': line['line_number'],
                    'source': source,
                    'original': target,
                    'edited': target,
                    'modified': False,
                    'reasoning': 'No terms to edit'
                })
                continue

            # Get LLM to edit the translation
            edit_result = self.edit_translation(source, target, terms)

            edited_translations.append({
                'line_number': line['line_number'],
                'source': source,
                'original': target,
                'edited': edit_result['edited_translation'],
                'modified': edit_result['modified'],
                'reasoning': edit_result['reasoning']
            })

        return edited_translations

    def save_edits(self, edited_translations: List[Dict[str, Any]],
                  output_path: str) -> None:
        """Save the post-edited translations with analysis to a file.

        Args:
            edited_translations (List[Dict[str, Any]]): Edited translations with analysis
            output_path (str): Path to save results
        """
        with open(output_path, 'w', encoding='utf-8') as f:
            json.dump({
                'summary': {
                    'total_lines': len(edited_translations),
                    'modified_lines': sum(1 for t in edited_translations if t['modified'])
                },
                'translations': edited_translations
            }, f, ensure_ascii=False, indent=2)


# Example usage:
if __name__ == "__main__":
    import os

    # Load validation results
    with open('data/validation_results.json', 'r', encoding='utf-8') as f:
        validation_results = json.load(f)

    # Create editor and process translations
    editor = LLMTranslationEditor(
        validation_results,
        os.getenv('ANTHROPIC_API_KEY')
    )
    edited_translations = editor.post_edit_translations()

    # Save results
    editor.save_edits(edited_translations, 'llm_post_edited_translations.json')

    # Print summary and examples
    print(f"Post-editing completed:")
    print(f"Total lines: {len(edited_translations)}")
    print(f"Modified lines: {sum(1 for t in edited_translations if t['modified'])}")

    print("\nExample modifications:")
    for trans in edited_translations:
        if trans['modified']:
            print(f"\nLine {trans['line_number']}:")
            print(f"Source  : {trans['source']}")
            print(f"Original: {trans['original']}")
            print(f"Edited  : {trans['edited']}")
            print(f"Reasoning: {trans['reasoning']}")