ensure correct output language

climateqa/engine/chains/intent_categorization.py

@@ -1,4 +1,3 @@
-
 from langchain_core.pydantic_v1 import BaseModel, Field
 from typing import List
 from typing import Literal
@@ -44,7 +43,7 @@ def make_intent_categorization_chain(llm):
     llm_with_functions = llm.bind(functions = openai_functions,function_call={"name":"IntentCategorizer"})
 
     prompt = ChatPromptTemplate.from_messages([
-        ("system", "You are a helpful assistant, you will analyze, translate and categorize the user input message using the function provided. Categorize the user input as ai ONLY if it is related to Artificial Intelligence, search if it is related to the environment, climate change, energy, biodiversity, nature, etc. and chitchat if it is just general conversation."),
+        ("system", "You are a helpful assistant, you will analyze, detect the language, and categorize the user input message using the function provided. You MUST detect and return the language of the input message. Categorize the user input as ai ONLY if it is related to Artificial Intelligence, search if it is related to the environment, climate change, energy, biodiversity, nature, etc. and chitchat if it is just general conversation."),
         ("user", "input: {input}")
     ])
 
@@ -58,11 +57,19 @@ def make_intent_categorization_node(llm):
 
     def categorize_message(state):
         print("---- Categorize_message ----")
+        print(f"Input state: {state}")
 
         output = categorization_chain.invoke({"input": state["user_input"]})
-        print(f"\n\nOutput intent categorization: {output}\n")
-        if "language" not in output: output["language"] = "English"
+        print(f"\n\nRaw output from categorization: {output}\n")
+        
+        if "language" not in output:
+            print("WARNING: Language field missing from output, setting default to English")
+            output["language"] = "English"
+        else:
+            print(f"Language detected: {output['language']}")
+            
         output["query"] = state["user_input"]
+        print(f"Final output: {output}")
         return output
     
     return categorize_message

climateqa/engine/chains/standalone_question.py

@@ -18,7 +18,8 @@ def make_standalone_question_chain(llm):
         ("user", """Chat History: {chat_history}
         User Question: {question}
         
-        Transform this into a standalone question:""")
+        Transform this into a standalone question:
+        Make sure to keep the original language of the question.""")
     ])
     
     chain = prompt | llm
@@ -29,6 +30,8 @@ def make_standalone_question_node(llm):
     
     def transform_to_standalone(state):
         chat_history = state.get("chat_history", "")
+        if chat_history == "":
+            return {}
         output = standalone_chain.invoke({
             "chat_history": chat_history,
             "question": state["user_input"]

front/tabs/chat_interface.py

@@ -56,7 +56,7 @@ def create_chat_interface(tab):
     )
     with gr.Accordion("Click here for follow up questions examples", elem_id="follow-up-examples",open = False):
         follow_up_examples_hidden = gr.Textbox(visible=False, elem_id="follow-up-hidden")
-        follow_up_examples = gr.Examples(examples=[], label="", inputs= [follow_up_examples_hidden], elem_id="follow-up-button", run_on_click=False)
+        follow_up_examples = gr.Examples(examples=["What evidence do we have of climate change ?"], label="", inputs= [follow_up_examples_hidden], elem_id="follow-up-button", run_on_click=False)
 
     with gr.Row(elem_id="input-message"):