Generating More Interesting Responses in Neural Conversation Models with Distributional Constraints

By Ashutosh Baheti, Alan Ritter, Jiwei Li, William B. Dolan. 2018

94 citations recorded by the source corpus.

Neural conversation models tend to generate safe, generic responses for most inputs. This is due to the limitations of likelihood-based decoding objectives in generation tasks with diverse outputs, such as conversation. To address this challenge, we propose a simple yet effective approach for incorporating side information in the form of distributional constraints over the generated responses. We propose two constraints that help generate more content rich responses that are based on a model of syntax and topics (Griffiths et al., 2005) and semantic similarity (Arora et al., 2016). We evaluate our approach against a variety of competitive baselines, using both automatic metrics and human judgments, showing that our proposed approach generates responses that are much less generic without sacrificing plausibility. A working demo of our code can be found at https://github.com/abaheti95/DC-NeuralConversation.

Subfields: Artificial Intelligence, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

Citations in the corpus

Open-Domain Conversational Agents: Current Progress, Open Problems, and Future Directions (2020, methodology): “uses. 8 MILD My dad taught me everything I know - Im a construction worker. 4 BLAND Im a construction worker. 2 Figure 4: Speciﬁcity level using generative control (See et al., 2019). et al., 2017; Baheti et al., 2018). These methods provide a mechanism to control, and hence increase, the rare words”
Plug and Play Language Models: A Simple Approach to Controlled Text Generation (2019, methodology): “Similarly, Baheti et al. (2018) proposed a decoding strategy for generating interesting responses in dialogue systems, using bags of words and word embed-dings.”
What makes a good conversation? How controllable attributes affect human judgments (2019, methodology): “This method of controlling response-relatedness is similar to that described in (Baheti et al., 2018).”
Justifying Recommendations using Distantly-Labeled Reviews and Fine-Grained Aspects (2019, background): “BLEU) for open-domain generation tasks (Baheti et al., 2018; Gao et al., 2018).”
Plug-and-Play Conversational Models (2020, methodology): “…controlled generation include 1) re-weighting the output distribution using discriminators (Holtzman et al., 2018) or bag of words (Ghazvininejad et al., 2017; See et al., 2019; Baheti et al., 2018), and 2) perturbing the models activation with an attribute model (PPLM) (Dathathri et al., 2019).”
Offline RL for Natural Language Generation with Implicit Language Q Learning (2022, methodology): “We compare ILQL to: (1) ILQL (utterance): a per-utterance adaptation of ILQL that is nearly identical to ILQL, except that it removes the conservatism loss term and performs Bellman backups at the utterance level instead of the token level; (2) SARSA (utterance): a per-utterance version of SARSA, wh”
Comparison of Diverse Decoding Methods from Conditional Language Models (2019, methodology): “We match the model architecture and training data of Baheti et al. (2018).”
Policy-Driven Neural Response Generation for Knowledge-Grounded Dialog Systems (2020, methodology): “Weighted decoding (See et al., 2019) instead uses token-level features that are controllable (Ghazvininejad et al., 2017; Baheti et al., 2018) and supplements the scores from the de-coder model output with these features.”
Cluster-based beam search for pointer-generator chatbot grounded by knowledge (2020, background): “(Baheti et al. 2018) added distributional constraints such as latent topics into the model to generate more interesting responses.”
Do Massively Pretrained Language Models Make Better Storytellers? (2019, background): “The issue has inspired multiple explanations (Jiang and de Rijke, 2018) and multiple solutions – such as alternative training objectives (Li et al., 2016), decoding objectives (Baheti et al., 2018; See et al., 2019), and architectural changes (Fan et al., 2018).”
Improving Factual Consistency Between a Response and Persona Facts (2021, methodology): “Inspired by Baheti et al. (2018) and See et al. (2019), we use cosine similarity between ~ r and ~ u T 1 as a proxy for topical coherence: R 2 = cos ( ~ r, ~ u T 1 ) .”