Packaging Information as Fact Versus Opinion: Consequences of the (Information-)Structural Position of Subjective Adjectives

ABSTRACT

How do we distinguish fact from opinion? We tested whether people’s ability to detect opinion-based content—as indicated by the use of subjective adjectives (e.g., amazing, frustrating)—depends on the linguistic position of the adjective. Our results show that simply changing the linguistic structure of a sentence influences our perception of the sentence’s subjectivity: The same basic information, packaged differently in linguistic terms, yields significantly different subjectivity ratings. Specifically, our results show that texts with subjective adjectives in syntactic positions associated with new information and “main news” are rated as more opinion-based than texts conveying the same core information with the same adjective presented in a position that presents the information as already-known information or as secondary information. We also show that this information-packaging effect is independent of whether the sentence provides grounds/evidence for the opinion. More generally, our results suggest that linguistic-packaging choices can be used to blur the distinction between fact and opinion or, at least, our ability to perceive opinion-based information as such.

Acknowledgments

Thanks to Ian D’Elia and Sarah Hye-yeon Lee for help with implementing Experiments 3 and 4. An earlier version of some of this research was presented at the 94th Annual Meeting of the Linguistic Society of America and at the 42nd Annual Virtual Meeting of the Cognitive Science Society. An earlier version of some of this research is included in the Proceedings of the 42nd Meeting of the Cognitive Science Society. We also gratefully acknowledge funding from the USC Provost’s Undergrad Research Fellowship and from the National Science Foundation.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed on the publisher’s website.

Notes

1. https://www.wsj.com/articles/a-tax-break-for-coronavirus-hit-states-11586112131 retrieved 9/20/2020.

2. https://www.courier-journal.com/story/opinion/2018/10/15/kentuckys-proposed-gas-tax-hike-unfair-and-unnecessary/1647435002/ retrieved 9/20/2020.

3. Our term complex subjective adjectives is largely synonymous with adjectives described as predicates of personal taste and multidimensional adjectives. However, this distinction is not relevant for the issues investigated in this paper.

4. The six practice items expressed opinions or were fully factual but did not use the structures or adjectives tested in targets. For example, “The well-known establishment has the tastiest sandwiches in the whole town” expresses an opinion about taste using a superlative; we expected it to be rated 5 or 6 on the 6-point scale (1 = a sentence is all fact; 6 = it contains opinion-based content). “Afterwards, the president met with the prime minister as well as the foreign minister to discuss the upcoming constitutional debate” is factual; we expected it to be rated 1 or 2. Instructions (below) explained that presence of any opinion-based information means a sentence should be treated as “opinion.”

5. In some cases, singularities arose with z score analyses (even with the simplest random effect structure), presumably due to lack of variance (e.g., in Experiment 4, all conditions receive uniformly low ratings).

6. We asked about (noncorrectable) vision and hearing in all experiments, because auditory and visual impairments can limit individuals’ access to English-language input.

7. A potential concern with MTurk is the decrease in data quality that researchers began to observe in summer 2018 (e.g., Chmielewski & Kucker, 2020; Kennedy et al., 2020). Given these concerns, we used a variety of data-validity indicators (both practice items and catch trials) to exclude participants, in line with the advice of Chmielewski and Kucker. Our exclusion numbers may seem high but are lower than Chmielewski and Kucker’s: They found that 38% to 62% of MTurk participants failing at least one data-quality validity indicator in summer/fall 2018 and in spring 2019.

8. There is a precedent for predicative position being linked to subjectivity: Corpus work shows adjectives that often occur in the predicative position are more likely to be subjective than adjectives in other positions (Wiegand et al., 2013).

Additional information

Funding

This material is partly based on work supported by the National Science Foundation under Grant No. 1749612.