Natural Language Processing in Accounting, Auditing and Finance: A Synthesis of the Literature with a Roadmap for Future Research

• Author: Margaret R. Garnsey,Mark E. Hughes,Ingrid E. Fisher
• Published date: 01 July 2016
• Date: 01 July 2016
• DOI: http://doi.org/10.1002/isaf.1386

NATURAL LANGUAGE PROCESSING IN ACCOUNTING,

AUDITING AND FINANCE: A SYNTHESIS OF THE LITERATURE

WITH A ROADMAP FOR FUTURE RESEARCH

INGRID E. FISHER

a

*, MARGARET R. GARNSEY

b

AND MARK E. HUGHES

a

a

The University at Albany, State University of New York, Albany, NY, USA

b

Siena College, Loudonville, NY, USA

SUMMARY

Natural language processing (NLP) is a part of the artificial intelligence domain focused on communication be-

tween humans and computers. NLP attempts to address the inherent problem that while human communications

are often ambiguous and imprecise, computers require unambiguous and precise messages to enable understand-

ing. The accounting, auditing and finance domains frequently put forth textual documents intended to communi-

cate a wide variety of messages, including, but not limited to, corporate financial performance, management’s

assessment of current and future firm performance, analysts’assessments of firm performance, domain standards

and regulations as well as evidence of compliance with relevant standards and regulations. NLP applications have

been used to mine these documents to obtain insights, make inferences and to create additional methodologies and

artefacts to advance knowledge in accounting, auditing and finance. This paper synthesizes the extant literature in

NLP in accounting, auditing and finance to establish the state of current knowledge and to identify paths for future

research. Copyright © 2016 John Wiley & Sons, Ltd.

Keywords: computational linguistics; natural language processing; text mining

1. INTRODUCTION

1.1. Background and Motivation

Natural language processing (NLP) focuses on the study and facilitation of communication between

people and computers. One of the central problems in artificial intelligence (AI) is that of communica-

tion. AI is typically defined as intelligence that is demonstrated by software or machinery that imitates

the workings of the human mind (Tung,Quek, & Cheng 2004). It is an inherently interdisciplinary field

of study that draws from domains as diverse as psychology, computer science, linguistics and neurosci-

ence. NLP, more specifically, encompasses a range of computational techniques for analysing and

representing texts at one or more levels of linguistic analysis to enable human-like language processing

for a range of particular tasks or applications.

Although NLP has been described in many ways, the working definition underlying this discussion

follows the notion that NLP ‘is a theoretically motivated range of computational techniques for analyz-

ing and representing naturally occurring texts at one or more levels of linguistic analysis for the purpose

* Correspondence to: Ingrid E. Fisher, The University at Albany, State University of New York, Albany, NY, USA. E-mail:

ifisher@albany.edu

Copyright © 2016 John Wiley & Sons, Ltd.

INTELLIGENT SYSTEMS IN ACCOUNTING, FINANCE AND MANAGEMENT

Intell. Sys. Acc. Fin. Mgmt. 23, 157–214 (2016)

Published online 1 March 2016 in Wiley Online Library (wileyonlinelibrary.com)DOI: 10.1002/isaf.1386

of achieving human-like language processing for a range of tasks or applications’(Liddy, 2001). This

implies, as demonstrated throughout this paper, that ‘there are multiple methods or techniques from

which to choose to accomplish a particular type of language analysis’(Liddy, 2001). We reference a

variety of analytical methods that incorporate NLP data, at various ‘levels of linguistic analysis’(Liddy,

2001). However, since ‘NLP is considered a discipline within Artificial Intelligence (AI)’we focus

most heavily on the AI-related applications of NLP (Liddy, 2001). Despite the plethora of analytical

methods that are associated with and use data produced by NLP, it comprises a unique set of ‘compu-

tational techniques’that should not be confused or conflated with the many analytical tools referenced

in the text that follows.

Fifty years ago Goldberg (1965) stated: ‘it is scarcely an exaggeration to say that the problem of

communication is the axial problem in accounting’. Accounting practice is replete with written docu-

ments intended to communicate such messages as, but not limited to: current accounting standards, past

and expected future corporate performance (along multiple dimensions), policies and practices embod-

ied by financial statements and the results of financial statement audits. The growth in digital and social

media usage by businesses has further increased the volume of unstr uctured text documents. Since the

dissemination of these documents and messages is now largely automated by and through computers,

we believe that NLP research and applications have considerable potential to enhance communication

in the areas of accounting, auditing and finance. Our paper is motivated by the desire to determine the

state of the extant literature in NLP in accounting, auditing and finance so as to inform and guide future

research efforts.

1.2. Objectives

This paper examines the NLP literature in accounting, auditing and finance. Our objectives are: (1) to

synthesize the literature and present the lessons to be learned from prior work, (2) to identify

unanswered questions that present ripe opportunity for future research and (3) to identify the constraints

that are likely to present challenges to future research. We focus our literature reviewon combinations

of NLP with AI. Machine learning (ML) is often considered to be a subfield of AI. However, the large

number of studies employing ML, as opposed to other AI methodologies, drove us to acknowledge the

ML literature separately, where appropriate.

While we identify the various methods that accounting, audit and finance-related researchers have

used in association with NLP, we do not seek to create an artificial classification of NLP-related

studies. Rather, we acknowledge the existence of groups of studies that have applied similar NLP

analyses, or that sought to accomplish similar tasks by employing NLP, within the fields of

accounting, audit and finance. We also note the diverse analyti cal methods that have been

employed in addition to, or in combination with, NLP. Most importantly, we seek to highlight

the variety of applications of NLP that have developed and the ever-increasing uses of NLP-

generated data.

The remainder of this paper is organized as follows. Section 2 describes the methodology used to

identify the relevant literature and provides an overview of the evolution of NLP literature. A synthesis

of the NLP literature follows. We discuss NLP in accounting, audit and finance, focusing on two major

uses of NLP: classification (Section 3) and prediction (Section 4). Frequently-observed applications

and data sources are highlighted, followed by a review of readability studies in accounting, audit and

finance. In Section 5 we identify future research opportunities as well as likely obstacles and provide

concluding observations in Section 6.

158 I. E. FISHER ET AL.

Copyright © 2016 John Wiley & Sons, Ltd. Intell. Sys. Acc. Fin. Mgmt., 23, 157–214 (2016)

DOI: 10.1002/isaf

2. METHODOLOGY

2.1. Literature Selection

We began our search for relevant literature byscanning the bibliographies of four literature reviews that

surveyed NLP developments in the accounting, audit, and finance domains (Fisher et al., 2010; Li,

2010c; Kearney & Liu, 2014; Nassirtoussi et al., 2014). We expanded our exploration by performing

keyword searches of databases, including the Association for Computing Machinery (ACM) Digital

Library, ProQuest, and the linked databases accessible via EBSCO Host. The primary search terms

used included: text analysis, text mining, natural language processing (NLP), ML, artificial neural

networks (ANNs), support vector machines (SVMs), expert systems (ESs), artificial intelligence

(AI), latent semantic analysis (LSA), content analysis and computerized content analysis. Finally, the

bibliographies of retrieved papers were manually scanned, identifying additional sources.

2.2. Literature Assessment

An analysis of the diverse bibliographic sources included in this study demonstrates the ongoing inter-

est in NLP. We assessed a total of 266 monographs, including 192 journal articles, 38 conference

papers, 25 working papers, 1 book chapter and 10 doctoral dissertations, as shown in Table I. Since

our primary focus is NLP combined with ML and/or AI, we excluded 86 studies that addressed manual

text analysis, a precursor to the computer-facilitated NLP often employed today, and 81 involving basic

text mining. Excluding the four literature reviews noted above, 95 studies featured NLP combined with

and/or augmented by ML and/or AI. The studies featuring co-occurrences of NLP, ML and/or AI come

from three primary domains: accounting, audit and finance, represented by 20, 15 and 60 items

respectively, as shown in Table II. Since NLP literature is replete with protracted terms, in the interest

of brevity we use complete designations, followed by acronyms, only for initial references. Thereafter,

acronyms are used. To aid the reader, a glossary of all acronyms and corresponding terms is provided in

Appendix A.

2.3. Brief Overview of the Evolution of Natural Language Processing Research

Assessing the literature from a chronological perspective, it is clear that NLP research has evolved, over

time, as illustrated in Figure 1. Researchers’original focus was on manual text analysis, a non-

computerized approach used to tap the informational value of linguistic patterns in text-based data.

The earliest research we found employed manual content analysis in order to assess the readability of

corporate reports. It found that the ‘general level of reading was difficult, the human interest value dull,

Table I. Literature categorized by bibliographic source

Retrieved literature Accounting Audit Finance Total

Journal article 72 26 94 192

Conference papers 11 3 24 38

Working papers 2 2 21 25

Book chapters 1 —— 1

Doctoral dissertations 3 1 6 10

Total articles retrieved 89 32 145 266

NLP IN ACCOUNTING, AUDITING AND FINANCE 159

Copyright © 2016 John Wiley & Sons, Ltd. Intell. Sys. Acc. Fin. Mgmt., 23, 157–214 (2016)

DOI: 10.1002/isaf