Surface realisation decisions in language generation can be sensitive to a language model, but also to decisions of content selection. We therefore propose the joint optimisation of content selection and surface realisation using Hierarchical Reinforcement Learning (HRL). To this end, we suggest a novel reward function that is induced from human data and is especially suited for surface realisation. It is based on a generation space in the form of a Hidden Markov Model (HMM). Results in terms of task success and human-likeness suggest that our unified approach performs better than greedy or random baselines. 1

Texts with potential educational value are becoming available through the Internet (e.g., Wikipedia, news services). However, using these new texts in classrooms introduces many challenges, one of which is that they usually lack practice exercises and assessments. Here, we address part of this challenge by automating the creation of a specific type of assessment item. Specifically, we focus on automatically generating factual WH questions. Our goal is to create an automated system that can take as input a text and produce as output questions for assessing a reader’s knowledge of the information in the text. The questions could then be presented to a teacher, who could select and revise the ones that he or she judges to be useful. After introducing the problem, we describe some of the computational and linguistic challenges presented by factual question generation. We then present an implemented system that leverages existing natural language processing techniques to address some of these challenges. The system uses a combination of manually encoded transformation rules and a statistical question ranker trained on a tailored dataset of labeled system output. We present experiments that evaluate individual components of the system as well as the system as a whole. We found, among other things, that the question ranker roughly doubled the acceptability

Studies assessing rating scales are very common in psychology and related fields, but are rare in NLP. In this paper we assess discrete and continuous scales used for measuring quality assessments of computergenerated language. We conducted six separate experiments designed to investigate the validity, reliability, stability, interchangeability and sensitivity of discrete vs. continuous scales. We show that continuous scales are viable for use in language evaluation, and offer distinct advantages over discrete scales. 1

Language generators in situated domains face a number of content selection, utterance planning and surface realisation decisions, which can be strictly interdependent. We therefore propose to optimise these processes in a joint fashion using Hierarchical Reinforcement Learning. To this end, we induce a reward function for content selection and utterance planning from data using the PARADISE framework, and suggest a novel method for inducing a reward function for surface realisation from corpora. It is based on generation spaces represented as Bayesian Networks. Results in terms of task success and humanlikeness suggest that our unified approach performs better than a baseline optimised in isolation or a greedy or random baseline. It receives human ratings close to human authors. 1

This paper describes a novel probabilistic approach for generating natural language sentences from their underlying semantics in the form of typed lambda calculus. The approach is built on top of a novel reduction-based weighted synchronous context free grammar formalism, which facilitates the transformation process from typed lambda calculus into natural language sentences. Sentences can then be generated based on such grammar rules with a log-linear model. To acquire such grammar rules automatically in an unsupervised manner, we also propose a novel approach with a generative model, which maps from sub-expressions of logical forms to word sequences in natural language sentences. Experiments on benchmark datasets for both English and Chinese generation tasks yield significant improvements over results obtained by two state-of-the-art machine translation models, in terms of both automatic metrics and human evaluation. 1