Towards human-like spoken dialogue systems

TL;DR: The two-way mimicry target is presented, a model for measuring how well a human-computer dialogue mimics or replicates some aspect of human-human dialogue, including human flaws and inconsistencies.

About: This article is published in Speech Communication.The article was published on 2008-08-01 and is currently open access. It has received 145 citations till now.

Summary

Introduction

• This is the submitted version of a paper published in The Journal of product innovation management.
• N.B. When citing this work, cite the original published paper.
• This study explores some of this complexity by theoretically detailing and empirically examining the critical role that synchronization plays in the process of leveraging resources to create innovation.
• To do so requires that firms “efficiently orchestrate their resources to innovate and outcompete their competitors in the global market” (De Massis, et al., 2018: 136); more specifically, resources and capabilities need to be managed and leveraged in a way that supports innovation (Lichtenstein and Brush, 2001).

2.1. Innovation

• Prior research has established three concepts key to innovation: risk, slack, and resource recombination.
• One aspect of BTOF is that organizational goals, or aspirations, are set in the context of the firm’s perceived performance relative to its rivals in the market.
• According to the resource-based view, a central task of managers is the optimal allocation of scarce and valuable resources (Barney, 1991, Wernerfelt, 1984).

2.2.1. The role of managers

• RO research concerns managerial strategy, decisions, and actions.
• RO identifies managerial roles in three broad categories: (1) structuring, (2) bundling, and (3) leveraging.
• Structuring captures the subprocesses associated with acquiring, accumulating, and divesting the resources; bundling encompasses the subprocesses associated with combining, integrating, and transforming resources into capabilities; and leveraging involves the deployment of the capabilities to create value (Helfat and Peteraf, 2003, Sirmon, et al., 2007).

2.2.2. Leveraging strategies

• Of the three managerial roles, leveraging is arguably the most crucial.
• Innovations using this leveraging strategy are most likely to produce improvements in existing products and expand existing markets.
• Steve Jobs’ well-known pivot from using touch-screen technology for the iPhone to the iPad reconfigured an existing capability to target a new market opportunity he identified.
• Examples include the creation and proliferation of telemedicine by firms such as CareClix, virtual reality headsets by such firms as Oculus Rift, and even commercial space flight by firms such as SpaceX.
• In summary, the resource advantage strategy leverages current capabilities in a current market, the market opportunity strategy leverages current capabilities in a new market, and the entrepreneurial strategy leverages new capabilities in a new market.

2.2.3. Synchronization

• Synchronization refers to the managerial process of integrating and coordinating RO actions within the firm to support and implement a specific leveraging strategy (Sirmon, et al., 2007, Sirmon, et al., 2011).
• If a firm decides to adopt a new ‘direct to consumer’ distribution model as part of a market opportunity strategy, its managers will have to coordinate changes to internal processes across the organization, which potentially involves changes in human capital, information technology, and/or financial resources, among others to ensure they all support the firm’s overall strategy.
• This bias potentially leaves certain intrafirm groups with unidentified slack, while other groups may lack the resources needed to fully support the implementation of a firm’s leveraging strategy.
• With greater synchronization, managers can coordinate resources across processes and between units allocating resources to meet each unit’s specific needs, thereby enabling greater utilization of resources across the firm (Bonaccorsi and Lipparini, 1994, Emden, et al., 2006).
• The authors predict that synchronization positively moderates the relationship between each leveraging strategy and innovation.

3.1. Differential Impact of Performance

• Next, the authors suggest that the relationship among synchronization, leveraging strategy, and innovation is also influenced by the firm’s relative performance context, which plays an important role in the choice of leveraging strategy.
• By increasing efficiencies and reducing friction, synchronization enables a poorly performing firm to build the slack resources it needs, which can then be deployed to build new capabilities as part of an entrepreneurial leveraging strategy.
• At the same time, the authors ensured that the variance of the levels of each item was balanced so that each item had approximately the same probability of influencing a respondent’s evaluations of the scenarios.

4.2. Sample

• To invite study participants, the authors used a random subsample of 600 entrepreneurial firms that had been identified by the entrepreneurship center of a large research university located in the southwestern United States.
• Alternatively, the authors employ subgroup analysis to test their high and low performance predictions, as these hypotheses suggest changes in the strength of the relationship.
• The authors findings are the first to provide empirical support for the contention that synchronization of the firm’s resources, capabilities, and strategies plays a critical role in the development of innovation (Sirmon, et al., 2011).

Figures

Figure 3: human-computer interaction on the computer's terms (DH’C) and humancomputer interaction on human terms (DHC’)

Frequently asked questions

Q1. What contributions have the authors mentioned in the paper "Towards human-like spoken dialogue systems" ?

In this paper, the authors present an analysis of how users perceive spoken dialogue systems in terms of other, more familiar things. 

Q2. What is the way to test candidates for human-likeness?

In other words, tests for response implicitly test perception and understandingas well, but as response tests are considerably more expensive, it is prudent to test candidates using perception and understanding tests first. 

Q3. What are some examples of features that can be accessed from SAD decisions?

Examples of such features include different durations, such as the lengths of pauses, overlapping speech and utterances, which can be accessed using speech activity detection; prosodic features, such as intensity and pitch; and turn-taking patterns, which can be accessed from SAD decisions and an interaction model (e.g. Brady, 1968). 

Q4. What is the example of a machine that can make a human believe she is talking?

Nigel Ward’s humming machine (Ward & Tsukahara, 2000) is an example of a machine that can potentially make a human believe she is talking to another human, if presented in an appropriate context – like a telephone conversation where one person does most of the talking. 

Q5. What was the lack of realism in traditional wizard-of-oz collections?

the lack of realism in traditional Wizard-of-Oz collections, a method which was coined in-service Wizard-of-Oz data collection was introduced, in which the wizards were real customer care operators and the callers were real customers with real problems. 

Q6. What is the way to use wizards to aid a spoken dialogue system?

If the wizards are allowed to use whatever means they are given to the best of their ability and any restrictions imposed on them are encoded in the software, then the wizards’ actions represent the target the component designer should aim at – an idea akin to Paek (2001), who suggests using human wizards’ behaviour as a gold standard for dialogue components. 

Q7. What are some of the more novel twists in human-like spoken dialogue?

Other more novel twists include manipulating human-human dialogues on-line, effectively treating both participants as subjects and recording continuous judgments of some parameter by a panel of reviewers equipped with different kinds of audience response systems. 

Q8. What is the main argument for human-likeness in spoken dialogue systems?

There is also a growing interest for human-likeness in spoken dialogue systems amongst researchers (e.g. Philips, 2006, keynote speech, Interspeech, Pittsburgh, PA, US; Zue, 2007), and many researchers have made a case for anthropomorphism in spoken dialogue systems. 

Q9. Why is the uncanny valley not known to us?

For one thing, no reports of uncanny valley effects from users actually interacting with human-like spoken dialogue systems are known to us, possibly because spoken dialogue systems aiming at human-likeness are yet too immature.