108 Turing’s Test
agenda or mood. Recognising, therefore, the important limitations of
machine learning, it is helpful to consider some of the factors that have
contributed to the current state of automated journalism, employing algorithms as a “finite series of precisely described rules or processes to
solve a problem”.
36
The idea of teaching computers to understand human language
emerged as a field in the 1950s as part of machine translation, as when
IBM and Georgetown University demonstrated a computer that could
translate Russian sentences into English in 1954. Although the researchers believed that the problem of machine translation would be solved in a
few years, their own device was extremely limited, with a lexicon of only
250 words and a set of six grammatical rules.
37
The demand for translation, however, drove research throughout the 1970s and into the 1990s,
with statistical data analysis providing an early methodology for formalising the rules of language so that they were machine-readable and
thus, potentially, capable of generating text. An early example of such
machine-generated information was weather forecasting in the 1990s,
with a Forecast Generator (FOG) producing routine text forecasts from
weather maps.
38
Such developments then moved into sports,
39
medical
data,
40
and simple forms of storytelling.
41
Dörr offers a useful summary of recent developments in Natural Language Generation (NLG),
a subset of Natural Language Processing (NLP), defined by Reiter and
Dale as the ability of computer systems to automatically produce human
(natural) language from computational information.
42
The contribution
of NLG to algorithmic journalism has the potential, as has already been
noted in this chapter, to change greatly the role of journalists – but only
in restricted areas. The role of algorithmic journalism, as considered by
Dörr, follows what he refers to as an input-throughput-output (ITO)
model, taken from Latzer et al, in which electronic data are taken from
private or public databases (input), organised into relevant semantic
structures (throughput), and then published to a platform (output): the
technology behind NLG is what enables algorithmic journalism to take
place, and obviously depended on considerable advances in the ability of
machines to read such data in the first instance.
43
Significant advances appeared to be in place when, in 2015, stories
surfaced that Google had taught an AI, DeepMind, to read. The parent
company of DeepMind had been founded in London in 2010 and was
acquired by Google in 2014, having built a neural network that could
be taught to perform cognitive actions such as learning to play games
or to read text. In contrast to IBM’s Deep Blue and Watson, which were
developed to advance one clearly defined function (applying advanced
natural language functioning processes in the case of Watson), DeepMind claimed that its system was not pre-programmed but could learn
from experience, and it was put to the test against computer games –
for example, quickly mastering the video game Breakout and playing
agenda or mood. Recognising, therefore, the important limitations of
machine learning, it is helpful to consider some of the factors that have
contributed to the current state of automated journalism, employing algorithms as a “finite series of precisely described rules or processes to
solve a problem”.
36
The idea of teaching computers to understand human language
emerged as a field in the 1950s as part of machine translation, as when
IBM and Georgetown University demonstrated a computer that could
translate Russian sentences into English in 1954. Although the researchers believed that the problem of machine translation would be solved in a
few years, their own device was extremely limited, with a lexicon of only
250 words and a set of six grammatical rules.
37
The demand for translation, however, drove research throughout the 1970s and into the 1990s,
with statistical data analysis providing an early methodology for formalising the rules of language so that they were machine-readable and
thus, potentially, capable of generating text. An early example of such
machine-generated information was weather forecasting in the 1990s,
with a Forecast Generator (FOG) producing routine text forecasts from
weather maps.
38
Such developments then moved into sports,
39
medical
data,
40
and simple forms of storytelling.
41
Dörr offers a useful summary of recent developments in Natural Language Generation (NLG),
a subset of Natural Language Processing (NLP), defined by Reiter and
Dale as the ability of computer systems to automatically produce human
(natural) language from computational information.
42
The contribution
of NLG to algorithmic journalism has the potential, as has already been
noted in this chapter, to change greatly the role of journalists – but only
in restricted areas. The role of algorithmic journalism, as considered by
Dörr, follows what he refers to as an input-throughput-output (ITO)
model, taken from Latzer et al, in which electronic data are taken from
private or public databases (input), organised into relevant semantic
structures (throughput), and then published to a platform (output): the
technology behind NLG is what enables algorithmic journalism to take
place, and obviously depended on considerable advances in the ability of
machines to read such data in the first instance.
43
Significant advances appeared to be in place when, in 2015, stories
surfaced that Google had taught an AI, DeepMind, to read. The parent
company of DeepMind had been founded in London in 2010 and was
acquired by Google in 2014, having built a neural network that could
be taught to perform cognitive actions such as learning to play games
or to read text. In contrast to IBM’s Deep Blue and Watson, which were
developed to advance one clearly defined function (applying advanced
natural language functioning processes in the case of Watson), DeepMind claimed that its system was not pre-programmed but could learn
from experience, and it was put to the test against computer games –
for example, quickly mastering the video game Breakout and playing