---
title: Direct Response Generation
url: https://www.emergentmind.com/topics/direct-response-generation
type: topic
---

# Direct Response Generation

Searching arXiv for the cited papers and closely related work on direct response generation.
Direct response generation is a family of conditional generation paradigms in which a model maps an input representation directly to a response utterance. In the dialogue-systems literature, this can mean mapping a raw user utterance, optionally with dialogue history, directly to an agent reply without any explicit intermediate intent-classification component; in controllable NLG, it can mean mapping a meaning representation consisting of a dialogue act and semantic attributes directly to a surface utterance without hand-crafted templates or multiple pipeline steps [2509.05006] [2307.14440]. Across app-review automation, open-domain dialogue, target-guided transition generation, and service assistants, the shared objective is to learn \(p_\theta(y \mid x)\) or a closely related conditional distribution while coping with the one-to-many structure of valid replies, the need for control over response properties, and the difficulty of evaluating response quality beyond word overlap [2002.03552] [1906.06050].

## 1. Definition and formal problem setting

In the “single-step” formulation for service assistants, direct response generation trains a single model \(p_\theta(y \mid x)\) end-to-end to maximize the conditional likelihood of an agent response \(y\) given user input \(x\), rather than factorizing response generation through an explicit intent variable \(c\) [2509.05006]. In dialogue-act-conditioned NLG, the same idea is expressed as a direct mapping from a meaning representation \(x=(d,a)\), where \(d\) is a specified dialogue act and \(a\) is a bag of semantic attributes, to an utterance \(y=f(d,a)\) [2307.14440].

Several works make explicit that the response space is intrinsically one-to-many. Gao et al. formulate short-text conversation as generating a set of responses \(\{r_1,\dots,r_K\}\) for a single post and note that a given post is often associated with multiple replies simultaneously in real applications; their machinery can then be collapsed to a single direct reply by taking \(K=1\) or selecting the top latent word and decoding once [1811.05696]. “Neural Response Generation with Meta-Words” formalizes this same one-to-many relation through structured response attributes, or meta-words, that specify what kind of response to generate [1906.06050].

The conditional signal provided to a direct response generator varies by task. RRGen conditions on a review, keyword-topic tags, and review-side attributes such as app category, review length, user rating, and inferred sentiment score [2002.03552]. CODA conditions on dialogue context, a target sentence, and a generated commonsense bridging path [2205.09314]. EM pre-training for multi-party response generation conditions on dialogue history together with an addressee label, observed during downstream training and latent during pre-training [2305.12412]. These variations preserve the same underlying principle: response generation is performed directly from the model’s conditioned input rather than from manually authored templates.

## 2. Architectural families

The literature does not use a single canonical architecture for direct response generation. Instead, it presents a set of conditioning and control mechanisms layered onto seq2seq, Transformer, or LLM backbones.

| Family | Core conditioning signal | Representative work |
|---|---|---|
| Attentional seq2seq | Review attributes and keyword-topic tags | RRGen [2002.03552] |
| Hierarchical recurrent encoder-decoder | Word-level and utterance-level attention over context | HRAN [1701.07149] |
| GPT-2 target-guided generation | Commonsense bridging path plus target sentence | CODA [2205.09314] |
| Encoder-decoder with latent structure | Latent word, addressee label, or meta-word memory | [1811.05696], [2305.12412], [1906.06050] |
| LLM prompting and ranking | Few-shot prompts, evaluator scores, or RAG context | [2307.14440], [2206.04937], [2405.03845] |

RRGen extends a standard attentional Seq2Seq architecture by explicitly injecting an A-component of high-level attributes and a K-component of keyword/topic tags. Its review encoder uses token embeddings enriched with keyword-topic signals and a bidirectional GRU; its relation-learning module embeds each attribute with small MLPs and concatenates them with the encoder’s final state; and its decoder is a unidirectional GRU initialized with the fused context and equipped with attention over encoder states [2002.03552].

HRAN addresses multi-turn response generation with a three-layer encoder-decoder in which a word-level bidirectional GRU produces hidden states per utterance, an utterance-level GRU processes utterance summaries in reverse order, and hierarchical attention attends both within utterances and among utterances at each decoding step [1701.07149]. This design explicitly models the fact that words and utterances in a context are differentially important.

CODA uses a two-stage GPT-2-based framework. A Knowledge Path Generator first produces a multi-hop commonsense path between a source concept in the dialogue context and a target concept in the target sentence; a Commonsense Response Generator then conditions on the generated path, the target sentence, and the dialogue context to decode the transition response [2205.09314]. The explicit path serves as a controllable intermediate structure, but the output remains a directly generated response.

Other models introduce different forms of latent or structured control. Gao et al. use a latent variable \(z\) drawn from the vocabulary as a semantic anchor for response generation [1811.05696]. The EM-based multi-party model builds \(p_\theta(r_t \mid H_t,a_t)\) on BART and injects addressee information through learned addressee embeddings in the encoder input [2305.12412]. Meta-word generation augments seq2seq with a Goal-Tracking Memory Network that embeds response attributes, tracks which attributes have been satisfied, and reads out residual goals to guide the decoder [1906.06050].

LLM-era systems expand the design space further. Ramirez et al. use few-shot prompting plus overgenerate-and-rank for dialogue-act realization [2307.14440]. The generator-evaluator architecture of “Generate, Evaluate, and Select” uses a T5 generator and a BERT evaluator that scores candidate responses and selects the best one [2206.04937]. SCRABLE couples retrieval-augmented generation, an LLM-as-a-Judge, and self-optimizing prompts for customer review responses [2405.03845]. This suggests that direct response generation is best understood as a problem formulation rather than a single model class.

## 3. Objectives, decoding, and control mechanisms

A large portion of the literature trains direct response generators with standard autoregressive maximum likelihood or cross-entropy objectives. RRGen maximizes \(\log p_\theta(y \mid x,\tau,l,r,s,\kappa)\) over review-response pairs and decodes with beam search of size \(B=5\) [2002.03552]. HRAN minimizes the negative log-likelihood of the target response conditioned on the multi-turn context [1701.07149]. The fine-tuned T5 direct-response model for service assistants also uses standard token-level cross-entropy, optimized by AdamW via teacher forcing [2509.05006]. CODA separately trains the path generator and the response generator with negative log-likelihood objectives and combines them as \(L_{\text{total}} = L_{\text{path}} + L_{\text{resp}}\) [2205.09314].

Not all systems rely exclusively on MLE. Gao et al. formulate response generation with a latent-word policy \(p_W(z \mid x)\) and a response generator \(G_G(r \mid x,z)\), then fine-tune with REINFORCE using token-level F1-based rewards defined against a bag of human replies [1811.05696]. The EM pre-training approach for multi-party dialogue treats the addressee as a discrete latent variable, performs hard-EM by selecting the highest-posterior addressee label, and updates the generator on a high-confidence subset of pseudo-labeled examples [2305.12412]. In spoken-dialog response generation, automatic conversation evaluators are incorporated both by reranking beam candidates and by augmenting the training loss with an evaluator-based discriminator term weighted by \(\lambda=10\) [1904.13015].

Control enters direct response generation through multiple channels. RRGen conditions on user rating, review length, app category, inferred sentiment, and topic tags [2002.03552]. CODA controls response direction through a specified target sentence and a generated commonsense path [2205.09314]. Dialogue-act NLG controls the realized act through few-shot prompt styles and ranking functions that combine DA accuracy and semantic accuracy [2307.14440]. Meta-word generation controls length, dialogue act, copy ratio, specificity, and other response attributes through interpretable attribute records tracked by memory [1906.06050]. In multi-party dialogue, the addressee must be specified before the response is generated, making addressee conditioning central to the task formulation [2305.12412].

Candidate generation and selection also vary substantially. RRGen and the evaluator-guided spoken-dialog system use beam search [2002.03552] [1904.13015]. Gao et al. sample latent words from clustered candidate pools to encourage diversity across multiple responses [1811.05696]. The diversity-aware T5+BERT system generates seven candidates by combining greedy search, beam search, and repeated top-50 multinomial sampling, or forty-nine candidates in the combined DADE setting, and then chooses \(y^*=\arg\max_i s_i\) according to the evaluator [2206.04937]. Ramirez et al. overgenerate \(k=10\) candidates from few-shot prompts and rank them with functions such as \(RF_1(y)=DAprob(d\mid y,a)\times SACC(y)\times P(y)\) or the hard-DA variant \(RF_2^{DA}\) [2307.14440]. SCRABLE iteratively regenerates responses under revised prompts until the average judge score reaches a threshold of \(0.95\) or the maximum number of iterations is reached [2405.03845].

## 4. Evaluation methodology and empirical findings

Evaluation in direct response generation is notably heterogeneous. Corpus-level BLEU-4 is used in RRGen, where it measures 1–4 gram overlap between a single human reference response and the model output [2002.03552]. HRAN reports perplexity on development and test sets together with side-by-side human judgments [1701.07149]. Service-assistant work evaluates BLEU, ROUGE-L, BERTScore, Task Success Rate, and human ratings of fluency, relevance, and helpfulness [2509.05006]. Diversity-aware systems report BLEU-n, distinct-1/2, and human judgments of engagingness or semantic distinctness [1811.05696] [2206.04937]. CODA introduces TARGET-COHERENCE, a reference-less trained metric \(f(c,r,t)\in[0,1]\), after observing that BLEU, ROUGE, METEOR, and BERTScore correlate poorly with human judgment for target-guided transitions [2205.09314].

The empirical record shows that direct response generation can perform strongly when task-specific conditioning is designed carefully. On 58 apps and 309,246 review-response pairs, RRGen achieves BLEU-4 \(=36.17\), compared with \(21.61\) for attentional NMT, \(14.08\) for NNGen nearest-neighbor retrieval, and \(6.55\) for random response selection; the reported improvement over NMT is approximately \(+67.4\%\) with \(p<0.01\) and large effect \(d=0.74\) [2002.03552]. In multi-turn dialogue, HRAN reduces dev/test perplexity to \(40.26/41.14\), outperforming S2SA, HRED, and VHRED, and also wins significantly more human pairwise comparisons than it loses against each baseline [1701.07149].

Target-guided direct generation also benefits from explicit control. CODA reaches TARGET-COHERENCE \(=36.7\) on Otters-ID, versus \(26.7\) for GPT-2, \(27.8\) for Multigen, and \(28.3\) for GPT2-Fudge; on Otters-OOD, CODA reaches \(37.9\) versus a best baseline of approximately \(30.8\) [2205.09314]. Ablations show that removing paths drops TARGET-COHERENCE by more than 4 points, removing DailyDialog data drops it by approximately 1 point, and replacing relation-annotated paths with entities only loses about 3 points [2205.09314].

Few-shot controllable generation achieves very high act fidelity. Ramirez et al. report that several prompt settings achieve perfect DA accuracy and near perfect semantic accuracy of \(99.81\%\), outperforming few-shot fine-tuning [2307.14440]. With instruction-tuned models plus a definitional prompt, reported results include ViGGO \(DAC=100\%\), \(SACC=100\%\), \(PERF=99.4\%\); TV \(DAC=100\%\), \(SACC\approx99.9\%\), \(PERF\approx99.6\%\); and Laptop \(DAC\approx99.6\%\), \(SACC\approx99.5\%\), \(PERF\approx97.1\%\) [2307.14440].

Evaluator-guided generation also shows sizable gains. In spoken-dialog response generation, reranking and evaluator-based fine-tuning raise BLEU-4 on the Alexa Prize test set from \(5.9\) for the baseline seq2seq model to \(12.2\) for the combined \(S2S\_RR\_FT\) system, while human coherence and engagement scores increase from \(2.34/1.80\) to \(2.55/2.31\) [1904.13015]. In the T5+BERT generate-evaluate-select system, pairwise human evaluation on 2,000 held-out single-turn contexts shows that \(DE\_Best\) beats greedy decoding in \(44\%\) of cases and loses in \(21\%\), indicating the practical value of selecting among diverse candidates [2206.04937].

For customer review response generation with LLMs, SCRABLE reports an LLM-judge average score of \(0.90\) on the test set for the LLM-optimized prompt, versus \(0.87\) for the base prompt and \(0.89\) for a human-optimized prompt; human evaluation increases from \(0.78\) to \(0.85\), a reported \(+8.97\%\) improvement [2405.03845]. In service assistants, direct response generation with fine-tuned T5 is close to a fully fine-tuned intent-first pipeline: on BiToD, direct generation reaches BERTScore/ROUGE-L/BLEU of \(0.86/0.22/0.58\) against \(0.87/0.25/0.58\) for the two-step system, with Task Success Rate \(92\%\) versus \(94\%\); on Bitext, the corresponding Task Success Rates are \(90\%\) and \(91\%\) [2509.05006]. This directly challenges the assumption that explicit intent recognition is always necessary for high-quality service responses.

## 5. Major application domains

App-review and customer-review response generation is one of the clearest industrial settings for direct response generation. Hassan et al. are cited as finding that responding to a review increases the chances of a user updating their given rating by up to six times compared to not responding, motivating automation [2002.03552]. RRGen is explicitly designed for app-store review response generation, while SCRABLE uses retrieval-augmented generation over app manuals, FAQs, and past responses to produce customer-review replies and iteratively improve prompts [2002.03552] [2405.03845].

Open-domain dialogue remains the most diverse research domain. HRAN addresses multi-turn chat by focusing on hierarchical attention over context [1701.07149]. Gao et al. study short-text conversation and the generation of multiple diverse replies from a single post, then describe how the same machinery can be adapted to produce a single direct reply [1811.05696]. The generate-evaluate-select system focuses on single-turn Twitter dialogue and explicitly targets engagingness through evaluator-guided selection [2206.04937]. The spoken-dialog evaluator work similarly concentrates on coherence and engagement at each turn [1904.13015].

Controlled transition generation forms a separate subarea. CODA aims to move a conversation smoothly from a dialogue context toward a target sentence by generating a commonsense bridging path and conditioning generation on that path [2205.09314]. Dialogue-act-conditioned NLG treats direct response generation as a mapping from semantic intent specifications to utterances in domains such as ViGGO, Laptop, and TV [2307.14440]. A plausible implication is that direct response generation encompasses both response selection from conversational context and structured realization from semantic control variables.

Multi-party and service-assistant settings introduce additional task structure. In multi-party dialogue, the addressee of a response utterance should be specified before it is generated, and the EM pre-training framework addresses the lack of annotated addressee labels during pre-training by inferring them iteratively [2305.12412]. In service assistants, direct generation is compared directly with intent-first pipelines on BiToD and Bitext Customer Service, covering domains such as hotel booking, restaurant search, ride-hailing, and synthetic QA pairs spanning 27 intents [2509.05006].

Several papers also state that the framework can transfer beyond the original task. RRGen notes that the same supervised framework applies to other single-turn feedback-response tasks, including help-desk FAQs and product Q&A, by retraining on task-specific pairs and attributes [2002.03552]. The service-assistant study recommends starting with direct response generation in open-ended or rapidly evolving domains and considering retrieval augmentation or hybrid fallbacks in higher-stakes settings [2509.05006].

## 6. Limitations, controversies, and open directions

A recurring limitation is that many direct response generators are optimized for narrowly defined interaction structures. RRGen only handles the single-reply scenario, described as \(97.5\%\) of app-store dialogues, and extending it to multi-turn interaction would require a dialogue-state encoder [2002.03552]. The target-guided work depends on a small benchmark and compensates by re-purposing DailyDialog through SRL-based target extraction and filtering [2205.09314]. The multi-party setting shows that direct generation may require latent structural inference, here in the form of addressee estimation, before a useful response can be produced [2305.12412].

Another controversy concerns evaluation. CODA explicitly reports that BLEU, ROUGE, METEOR, and BERTScore hover near zero or even negative correlation with human judgment on target-guided response evaluation, motivating TARGET-COHERENCE with correlation \(\rho=0.47\) [2205.09314]. SCRABLE similarly replaces embedding-based similarity with an LLM-as-a-Judge rubric and reports Pearson correlation \(0.46\) with human scores on the test set, versus approximately \(0.13\) for the cited comparator [2405.03845]. These findings make clear that overlap-based metrics are often inadequate for response-generation tasks where multiple semantically acceptable outputs exist.

Direct response generation also raises controllability and reliability issues. The service-assistant study identifies slot omission in structured tasks, hallucination risk under domain shift, and reduced controllability because auditing which intent was recognized is impossible in pure direct generation [2509.05006]. The diversity-aware generator-evaluator model notes that collecting separate evaluation data for each generator type is costly and that scoring dozens of candidates adds inference cost [2206.04937]. The spoken-dialog evaluator work likewise depends on external evaluators and reranking infrastructure [1904.13015].

Task-specific noise sources remain important. RRGen notes that SentiStrength may mislabel user sentiment and proposes joint sentiment learning as a future extension; it also relies on generic placeholders such as “\<email\>” and “\<url\>” that must be post-processed with an app-specific lookup table [2002.03552]. Dialogue-act-conditioned prompting reports that DA-specified response quality can lag due to skewed DA distributions and imperfect DA classifiers in related setups [2206.04937], while the few-shot DA study itself depends on a high-accuracy DA classifier and domain-specific slot-error scripts for ranking [2307.14440].

Open directions are stated explicitly across the literature. RRGen suggests enriching context with review-post time, device information, or user-history embeddings and extending the model to multi-turn interaction [2002.03552]. CODA indicates that commonsense paths and augmentation are both essential, implying that richer knowledge planning may remain valuable for controlled dialogue [2205.09314]. The service-assistant comparison recommends hybrid architectures in high-stakes domains, retaining explicit intent labels for critical flows while using direct response generation for non-critical FAQs [2509.05006]. Taken together, these works suggest that the main frontier is not whether direct response generation is possible, but how much explicit structure, retrieval, planning, or post-hoc evaluation should be coupled to end-to-end generation in a given domain.

Source: https://www.emergentmind.com/topics/direct-response-generation