09 - Dialogue Systems Flashcards
What are some every day examples of dialogue systems?
Siri, Alexa, Cortana etc.
In recent years researchers have gone away from the ‘personal assistants’ approach and focused more on?
Large language models and chat-based apps such as ChatGPT.
What describes a chit-chat based system?
No goal. It is not task-oriented. It is just to chat, and it is very open-ended and has a broad domain. The object is to be indistinguishable from humans (turing test).
What describes a task oriented dialogue system?
Take action in the real world. They are goal oriented and wants to complete tasks. Natural language (conversational) is a nice-to-have.
Chit-chat uses more which kinds of models?
Seq-2-seq models
Goal-oriented models uses which models?
End-2-end models. Sometimes single domain models.
What are some challenges in dialogue systems? Mention atleast 3.
Variability in natural language.
Robustness.
Recall/precision trade-off
Meaningful representations
Common sense and knowledge of the world
Ability to learn
Transparency
Modular task-oriented systems can be divided into for example?
Predefined domains. Restaurant, taxi, movie etc.
A task-oriented system needs to be able to quickly identify what, when it is asked a question?
The domain and the intent of the question.
What is ‘slot filling’?
It is the process where a model, given a question, extracts the correct features for the ‘search’ or ‘goal’ such as: “Find me a good thai restaurant” –> Restaurant –> Good ratings –> Thai.
We sometimes have text classification models that are trained on transcribed text from spoken language. Why is this a challenge?
Because the transcribed text can be mismatched compared to when we write. Difference in speech and writing.
What is Dialogue State Tracking (DST)?
It refers to the process of monitoring and updating the current state of the dialogue during a conversation. It involves keeping track of relevant information, such as user goals, user preferences, and system actions, to maintain a coherent and accurate representation of the ongoing dialogue.
What is the benefit of using reinforcement learning in dialogue systems?
We introduce rewards. These can be both short term and long term and thus a model might learn to postpone immediate reward for a larger reward later on.
What is the MultiWOZ dataset?
It is a dataset that was created to facilitate research and development of dialogue systems that can handle multi-domain conversations.
Metrics in natural language generation that are based on word overlaps are?
BLEU, METEOR, ROGUE
Metrics in natural language generation that are based on word embeddings are?
BERTScore
BLEU uses which scores?
N-gram precision, which is a score between 0 and 1.
How does BERTScore work?
BERTScore is a metric that computes the similarity between two sentences by using contextualized word embeddings from BERT and aligning them based on cosine similarity to capture semantic similarity.
End-to-end dialogue systems are becoming more popular. What are they?
Yet again, it is a system that takes in an input and models an output without relying on intermediate processes or stages.
What is the ‘Proximal Policy Optimization’ (PPO)?
In Proximal Policy Optimization (PPO), a language model acts as a policy that generates text based on a given prompt, and the model’s parameters are updated to maximize reward metrics by combining a preference model, a constraint on policy shift, and computing advantage for each state-action pair.
What makes the ‘Alpaca’ models so unique?
They are trained on input that is generated as output from chatGPT model.
There are in this course 5 components of a prompt. These are?
- Input data
- Exemplars
- Instruction
- Indicators
- Context
For smaller models, if results are poor, what might we do?
Fine-tune the model itself, since fine-tuning the prompt might be ineffective.
For larger models (in dialogue systems), if results are poor, what might we do?
Fine-tune the prompt with advanced techniques such as chain-of-thought or least-to-most.
What is the key idea in ‘Knowledge augmentation’?
The key idea is to find knowledge in the form of relevant textual information that is provided as context in the prompt.
What is unique about conversational question answering?
Tasks like searching the web or recommendations are done through a conversational approach. It feels like talking to a very knowledgable person.
What does Sparrow propose as a conversational question anwering model?
An end-2-end approach using Large Language Models (LLM’s)
Sparrow is trained fine-tuning a LLM with reinforcement learning from human feedback. True or false?
True
How is the Sparrow model finetuned using human feedback?
Using per-turn response preference (finish non-complete sentence) and adversarial probing (make rule, identify violations).
Sparrow ‘introduces’ extra participants into the dialog. These are called search query and search result. What are these?
The search query is the string that queries the search and the result is the feedback from the query - the answer from google for example.
When we compare chit-chat models to task-oriented models which of the two is it more normal to incorporate LLM’s?
In the chit-chat models.
Why is it difficult to incorporate LLM into task-oriented models?
Because task-oriented models have explicit goals and therefore have a more moduralized pipeline for interpretability and controllability.
What is the ‘Few-Shot Bot (FSB)’?
A model that combines few-shot prompting with a skill selector.
it retrieves information depending on the skill and uses a LLM to generate an answer conditioned on dialogue history.
Recent trends in this topic of Dialogue Systems include?
Empathy in dialogue systems and multimodal (think writing AND pictures) dialogue systems.
