Excite your brain with negativity
« previous post | next post »
Darryl Shpak writes to me:
I came across the following surprising claim this morning; you might find it interesting:
"50 per cent of all words are negative, and only 30 per cent are positive, in both English and Spanish. So we tend to have a much more colorful, rich, negative vocabulary, and it's all because our brains are built to be particularly excited by negative things."
Darryl goes on:
This was in an interview in Maclean's magazine with Prof. Clifford Nass from Stanford, on the topic of human-computer interaction. The interview is available online at http://www2.macleans.ca/2010/09/30/clifford-nass-in-conversation/.
The claim seems obviously false on the surface: it only allows for 20% of words to be neutral, yet in any casual sampling of text, whether it's a couple of paragraphs from that very interview, or a page from a dictionary, or anything else I can find, I'm unable to categorize any more than a small minority of words as positive or negative, even under the most generous criteria.
I've done some web searching to try to find the source of this claim, but variations of the phrase "percent of words are negative/positive" have turned up no results except for that interview. Do you have any idea whether or not this is a misinterpretation or misstatement of a genuine statistic? Or is it complete fiction?
Complete fiction would be my guess. For it to be serious, we'd need a way to measure negativeness (how negative is the word negative?), a decision on how to delineate people's vocabularies (you can't use a dictionary if it's full of words that most people don't actually know), and some serious quantitative work to count them all up. Where has this been done?
And what would it show? Why, if I keep muttering about administrative bullshit in memos, and bullshit checks out as a negative word, would that show that "our brains are built to be particularly excited by negative things"?
Classic nobody-will-try-to-check-it-if-it's-about-language bullshit, I'd say. And it does not excite my brain.
[But: see below for a measured and informative comment by Professor Nass, who explains that (a) he was sorely misrepresented, since his claim was supposed to be about a specific subpart of the vocabulary, and (b) there is at least some quantitative evidence to back up the claim.
Andrew F said,
October 21, 2010 @ 4:57 am
Can a word even have a fixed intrinsic negativity? Is it still negative to say something is the shit, or the dog's bollocks, or even wicked? Is it positive to say the weather is cool?
Alon Lischinsky said,
October 21, 2010 @ 5:00 am
And measuring the negativity (121 times more frequent than 'negativeness' in COCA) or positivity of lexical items in isolation seems a pointless task, given that any specific reading is going to be highly contextual. Bill Louw has shown that, say, 'bent on' has a dominant negative semantic prosody, but 'build up' changes sharply depending on whether it's used as a transitive or intransitive verb. In any case, actual semantic prosodies only rarely match speakers' intuitions.
Adam said,
October 21, 2010 @ 5:29 am
There is a language resource for positivity and negativity:
SentiWordNet is a lexical resource for opinion mining. SentiWordNet assigns to each synset of WordNet three sentiment scores: positivity, negativity, objectivity.
Alon Lischinsky said,
October 21, 2010 @ 5:49 am
@Adam: I still have to read more closely the author's explanation of the SentiWordNet algorithm, but a quick look at the sample results in page 2203 leaves me highly doubtful of its accuracy. I can hardly think of any justification for ranking 'unsound' as more negative than 'miserable', 'wretched' and 'pathetic'.
Of course, this is likely to be related to the intrinsic polysemy of 'bad', which conflates (technical) judgement of fitness for a purpose and (ethical) praiseworthiness.
Mark Liberman said,
October 21, 2010 @ 6:12 am
I suggest asking Clifford Nass himself. He commented in a helpful way on an earlier post.
Thomas McCarthy-Ward said,
October 21, 2010 @ 7:09 am
There do seem to be several isolated negatives in common-ish use though, negatives for which we lack or maybe no longer use the positive, uncouth, disheveled, gormless, etc.
Mark P said,
October 21, 2010 @ 8:04 am
" … our brains are built to be particularly excited by negative things."
I think it's probably fair to say that our brains tend to remember negative things, or probably more accurately, unpleasant things. There are good evolutionary reasons for that. For example, it would be good to remember that eating a particular food caused sickness. but I don't think that has anything to do with language.
Lars said,
October 21, 2010 @ 8:19 am
@Adam and Alon: Since different models of understanding of interjections are of great interest to me, I performed a search on hurrah in the SentiWordNet. This yielded the following result: P: 0 O: 0.875 N: 0.125, i.e. it is considered 12,5% negative and 87,5% neutral. The search on wow P: 0.375 O: 0.625 N: 0. According to this 'lexical resource', hurrah does not score any positive sentiment, and wow no negative sentiment. Based on these results, I concur with Alon's assessment.
Lance said,
October 21, 2010 @ 8:27 am
TMcC-W said:
Yes, but what's proven by three words? Or even if the "etc." is filled out, by more words? I could turn this around and say that we don't have the opposite–positives for which we no longer use the root negative word–because our root words are positive and we derive negative words from the positive ones. Thus the negative-with-positive-root unhappy but not *unsad, unimportant and insignificant but not *intrivial or *unpetty…which tells us that positivity is at the heart of the way we think, whereas negativity is less natural and comes secondarily.
Or else it means that we'll readily form more and more negative words but not more positive ones.
Or else it means that studies show that 53% of all statistics are utter bullshit. In the technical sense.
Mark Liberman said,
October 21, 2010 @ 8:45 am
According to the description in this paper, the SentiWordNet sentiment scores were calculated by starting with two small "seed sets" of "7 'paradigmatically positive' terms" and "7 'paradigmatically negative' terms", and propagating these evaluations through the network via two algorithmic processes.
Exploring the results certainly gives some reason to doubt the effectiveness of this approach, as some commenters have already observed. To add a few examples:
corruption in the sense of "lack of integrity or honesty (especially susceptibility to bribery); use of a position of trust for dishonest gain" is 0.5 positive, 0.25 "objective" (i.e. neutral), and 0.25 negative.
Corruption in the sense of "moral perversion; impairment of virtue and moral principles" (corruption#4, in a synset with putrefaction#3 depravity#1 depravation#1 degeneracy#2) is P: 0.75 O: 0.125 N: 0.125.
Inspiration in the sense of "a sudden intuition as part of solving a problem" is P: 0 O: 0.875 N: 0.125.
Implausible in the sense of "highly imaginative but unlikely; 'a farfetched excuse'; 'an implausible explanation'" is P: 0.625 O: 0.25 N: 0.125.
Hope in the sense of "someone (or something) on which expectations are centered; 'he was their best hope for a victory'" is P: 0 O: 0.625 N: 0.375
Such examples make it clear that the algorithms, though interesting, are not producing uniformly plausible results in this case. It's not clear how well the results would stand up to a systematic test against human judgments, though there is obviously some reason for skepticism. But anyhow, I haven't seen any reason to believe that this has anything to do with Clifford Nass's remark.
Pflaumbaum said,
October 21, 2010 @ 9:59 am
@ Lance –
Your point is supported, I think, by the fact that it is clauses with positive polarity that are canonical.
Rob Malouf said,
October 21, 2010 @ 9:59 am
The MPQA Opinion Corpus might be a better place to check (it's mostly based on hand-annotated text). Their subjectivity lexicon lists 4,912 words with a negative prior polarity and 2,718 words with a positive prior polarity (the polarity is only "prior": positive words can be used negatively in the right context and vice versa).
The corpus itself is about 350,000 words of newspaper text marked up for subjectivity. The annotators marked 13,306 items as having subjective connotations. Of those, 2,037 were marked positive, 6,090 were marked neutral, and 5,179 were marked negative.
So, it's not 50% to 30%, but there are abut twice as many negative words as positive words in this corpus, counting either types or tokens. The fact that it's newspaper text no doubt has something to do with it — I wouldn't want to draw any conclusions about everyday language use from this!
Ryan Denzer-King said,
October 21, 2010 @ 10:29 am
Regardless of the truth of the language claim (which does seem especially dubious), isn't it true that "our brains are built to be particularly excited by negative things"? At least I've always understood that to be true, but maybe I've been brainwashed by bad science writing. However, a priori it definitely seems like while there's a distinct evolutionary advantage in responding strongly to negative eventualities, there is not corresponding pressure to have any response to positive eventualities. E.g., if I retreat in terror from a predator that could kill me, that helps me pass on my genes, whereas if I get excited about Blackfoot phonology, that doesn't really make me a more fit candidate for survival.
Pflaumbaum said,
October 21, 2010 @ 10:38 am
Well, getting excited over positive information about fitness – ripe, bulging fruits… other ripe, bulging things… is highly adaptive.
Jean-Sébastien Girard said,
October 21, 2010 @ 10:51 am
I think there might be another angle too: isn't there some research on proportions of praising vs. insulting synonyms that hinges toward that proportion?
I mean, it wouldn't be the first time journalists blatantly misrepresent language issues, and these I'm being particularly iffy about a magazine of which a recent cover story was called "a journalistic drive-by shooting".
Adam said,
October 21, 2010 @ 10:58 am
@Thomas: Couth is a word in my family.
Army1987 said,
October 21, 2010 @ 12:11 pm
@Lance: "non-trivial" is not so rare…
Chris said,
October 21, 2010 @ 12:25 pm
I wouldn't be surprised if the negative vs. positive claim was based on Pennebaker's LWIC data, based on human judgements. I have mixed feelings about LWIC.
Clayton Burns said,
October 21, 2010 @ 1:43 pm
I would be interested in how an analysis would work for a text sample (for example, here, from the end of Henry James's tale "The Beast in the Jungle"):
The escape would have been to love her; then, then he would have lived. She had lived—who could say now with what passion?—since she had loved him for himself; whereas he had never thought of her (ah how it hugely glared at him!) but in the chill of his egotism and the light of her use. Her spoken words came back to him—the chain stretched and stretched. The Beast had lurked indeed, and the Beast, at its hour, had sprung; it had sprung in that twilight of the cold April when, pale, ill, wasted, but all beautiful, and perhaps even then recoverable, she had risen from her chair to stand before him and let him imaginably guess. It had sprung as he didn’t guess; it had sprung as she hopelessly turned from him, and the mark, by the time he left her, had fallen where it was to fall. He had justified his fear and achieved his fate; he had failed, with the last exactitude, of all he was to fail of; and a moan now rose to his lips as he remembered she had prayed he mightn’t know. This horror of waking—this was knowledge, knowledge under the breath of which the very tears in his eyes seemed to freeze. Through them, none the less, he tried to fix it and hold it; he kept it there before him so that he might feel the pain. That at least, belated and bitter, had something of the taste of life. But the bitterness suddenly sickened him, and it was as if, horribly, he saw, in the truth, in the cruelty of his image, what had been appointed and done. He saw the Jungle of his life and saw the lurking Beast; then, while he looked, perceived it, as by a stir of the air, rise, huge and hideous, for the leap that was to settle him. His eyes darkened—it was close; and, instinctively turning, in his hallucination, to avoid it, he flung himself, face down, on the tomb. [End].
Knowledge is bitter, but better than delusion. How could it be meaningful to characterize words in this passage globally as positive or negative?
The entire tale, given that it is James's best, and certainly the best story in English, would be a good text for a quantitative measure–which would turn out to be a mirage of data. Even ignoring the manifest ambiguities of the text.
John Lawler said,
October 21, 2010 @ 1:55 pm
Possibly of interest — News Item: Record Negative Field Recorded on Internet. Though this is tongue in cheek, it's still true that if one speaks of semantactic negation, one can measure negativeness. But clearly the numbers given in the interview are rubbish from this standpoint, and it seems likely this is not what Nass was referring to.
Clayton Burns said,
October 21, 2010 @ 2:05 pm
I may have posted this comment at the Nass interview:
Brian Bethune has made Language Log:
[Excite your brain with negativity October 21, 2010 Filed by Geoffrey K. Pullum under Ignorance of Linguistics, Words words words. Darryl Shpak writes to me: I came across the following surprising claim this morning; you might find it interesting:
"50 per cent of all words are negative, and only 30 per cent are positive, in both English and Spanish. So we tend to have a much more colorful, rich, negative vocabulary, and it's all because our brains are built to be particularly excited by negative things."
Prof. Clifford Nass from Stanford… http://www2.macleans.ca/2010/09/30/clifford-nass-…
It seems that GKP is going to sack Prof. Clifford Nass. The professor and Brian Bethune had better get on the wire and fight back against this deliberate undermining of what must be valid research.
I should warn you that Pullum is more than capable of issuing a broadside. I have a private communication from him (which I am not going to release at this time) in which he decimates Marc Hauser.
Rod Johnson said,
October 21, 2010 @ 2:58 pm
You mean one out of every ten Marc Hauser is killed?!
Darryl Shpak said,
October 21, 2010 @ 4:48 pm
It does occur to me that the claim, as stated in the interview, might be missing some essential context. The previous question is about criticism and praise. So viewed in this context, it's possible that the statistic is described poorly, and instead is attempting to indicate (for example) that people use a more varied vocabulary when criticizing than when praising. (And perhaps then the remaining 20% of the vocabulary is made up of words that are used for both criticism and praise).
But all this is a complete guess, simply based on the assumption that Clifford Nass had some meaningful information that he was trying to communicate. I will not hazard a guess on whether he stated it poorly, or the reporter or editor managed to strip it of all meaning on the way to press.
Rosie Redfield said,
October 21, 2010 @ 4:50 pm
Rod, that would need to be nine out of every ten….
Clayton Burns said,
October 21, 2010 @ 5:47 pm
Rod Johnson: If you had Googled "decimate"–please, please, "Looked it up in Google," you would have found as the first entry:
[dec·i·mate/ˈdesəˌmāt/Verb
1. Kill, destroy, or remove a large percentage of.
2. Drastically reduce the strength or effectiveness of (something): "plant viruses that can decimate yields".]
That is what I meant. By analogy with a virus that can decimate yields. I hope that is settled.
You seem to have a negative cast of mind. As if nasty Mr. Nass were right.
Maria Wolters said,
October 21, 2010 @ 5:59 pm
Like Mark Liberman suggested, it might be worth asking Cliff Nass directly. I don't know him, but I have followed some of his work on Human-Computer Interaction. He strikes me as quite a careful experimenter, and I am wondering whether he was misquoted. As Rob Malouf pointed out, the figures might be based on a particular corpus. If Cliff used such a corpus for his research or for calibration purposes, that would explain a lot.
It would be interesting to see whether and in what context this assertion is repeated in his book; Wired for Speech, one of his previous offerings with Scott Brave, was extensively footnoted and annotated, and I hope the new one will be, as well.
Clayton Burns said,
October 21, 2010 @ 6:20 pm
The Montreal Gazette has an important caution: just because the word is "negative" does not mean it is negative.
[The word "negative" serves as a way to avoid some commercial four- letter words. For example, if you don't fancy the word "debt," try "negative cash"; if you want to avoid "loss," choose among "negative contribution," "negative impact on profits" and "negative cash-flow position."]
Even better:
[One of the most ghoulish and insensitive terms I've heard is "involuntary conversion" to refer to a plane crash… Some years ago, National Airlines stated in their annual report that the $1.7 million after tax insurance benefit was due to "the involuntary conversion of a 727."]
If you are in the plane, it is unpleasant. If you are a PR word doctor, you get a raise.
Clayton Burns said,
October 21, 2010 @ 6:39 pm
Rosie Redfield could be right:
[In Wonderful Life: The Burgess Shale and the Nature of History, Stephen Jay Gould uses "decimate" to indicate the taking of nine in ten, noting that the Oxford English Dictionary supports the "pedigree" of this "rare" meaning.]
I am willing to believe anything. If I had meant that one out of every ten Marc Hausers had been killed off by GKP, I probably would have said so.
You can do anything with language.
Clifford Nass said,
October 21, 2010 @ 7:22 pm
Thank you for the opportunity to address this issue. I do not think of this as "attack" or "defense"; my goal is to get the right answer.
Here's the full story:
1) The correct statement should be that "in English, 50% of words THAT DESCRIBE FEELINGS OR EMOTIONS are negative, 30% are positive, and 20% are neutral." In the context of my conversation with McLeans about emotions, I made this statement. I should have been clearer that I was not speaking about all words.
2) My reference to "English" was not to imply that English is special (like the Eskimos and snow" arguments. The paper I cited only mentioned English, and I was mindful of the fact that although I write in English, the book will be translated into other languages and the citation I relied on only discussed English and Spanish, both of which show this rough percentage. I have no idea about other languages.
3) The citation for this research is Schrauf, R. W., & Sanchez, J. (2004). The preponderance of negative emotion words in the emotion lexicon: A cross-generational and cross-linguistic study. Journal of Multilingual and Multicultural Development, 25(2), 266 – 284. I urge you to read the paper (which I did before citing it). The research is well done.
4) There is a big difference between the number of words and relative word use. It is very possible that even though half of all feeling words in English are negative, it's possible that people eschew those words and primarily use positive words. That's an interesting question and I do not know of data that addresses this, and I make no claims about that.
5) There are certainly non-feeling words that have positive, negative, or neutral connotations. I don't know the relative proportion of these words.
6) The dominance of negativity is well-established in many other domains of research, as discussed in Chapter 4 of my new book, "The Man Who Lied to His Laptop." Hence, the finding has face validity.
7) The literature on negativity and arousal is clear, large, and unambiguous to my mind. For a review, see Chapter 4 of my new book, "The Man Who Lied to His Laptop."
8) I agree that language seems to invite an unusually large number of myths and misstatements, for reasons that are not fully clear. I am cognizant of this, so when I cite research, I read it first and make my own judgments of its adequacy. I urge you to do the same.
9) Thanks again for your interest in my comments. I would certainly welcome people reading "The Man Who Lied to His Laptop" and providing feedback to me, either on the web or directly via email (nass@stanford.edu).
Clayton Burns said,
October 21, 2010 @ 7:23 pm
In relation to "…and it's all because our brains are built to be particularly excited by negative things" there is a powerful study that is accessible on the Internet in Emotion © American Psychological Association 2009, Vol. 9, No. 5, 609–618:
If the experiments were wrong, we would be hard put to it to account for the power of word choice in "Macbeth."
So Nass is onto something. He can keep his job after all. For the time being.
Darryl Shpak said,
October 21, 2010 @ 9:37 pm
The paper that Prof. Nass references in his third point above is available online here.
I haven't yet read the entire thing, though I intend to do so shortly. I'm probably not qualified to judge the quality of the research, but a casual scan suggests that the paper is well-grounded ("It is important to note here that we are not making an argument about languages themselves").
(And I'd like to thank Prof. Nass for his detailed reply!)
Julian Brooke said,
October 21, 2010 @ 10:25 pm
I'm part of a group that does lexicon-based automatic sentiment analysis. We built our dictionary primarily by pulling words from online review texts, and the results also support the general conclusion of Prof. Nass: off the top of my head, I recall that our (human tagged) dictionary is like something like 42% positive words and 58% negative words. However, if you count the appearance of words in sentiment-balanced corpora (i.e. equal numbers of positive and negative reviews), the skew is in the opposite direction (though the degree of positive bias is strongly domain dependent: in consumer product reviews, it's almost 2 to 1). That is, there are more negative words out there, but they are used less frequently than positive words, which might be because negative sentiment is often expressed by negating the positive, but positive sentiment is rarely expressed by negating the negative (except in certain fixed expressions like 'not bad').
Xmun said,
October 22, 2010 @ 1:45 am
Let me express the perhaps pedantic view that the term "negative" should only be applied to sentences with "no" or "not" or "none" (or such like) in them, or to words with prefixes like "in-", "un-", "non-", "dis-", etc.
"Bad", for example, is not a negative. Its negative is "not bad" (and let us not quibble about idiomatic subtleties).
maidhc said,
October 23, 2010 @ 1:36 am
I don't claim to know anything about brains, but it seems relatively obvious that people would find it more amusing to make up phrases of a negative nature, like "happy as a bastard on Father's Day" or "scarce as rocking horse manure", more so than "in the catbird seat". People would rather complain than say how great things are anyway, plus complaining gives more opportunity for verbal creativity.
Following Clayton Burns, it would be interesting to study Shakespeare with this in mind.
Graeme said,
October 23, 2010 @ 6:43 am
There's some intuitive appeal to this thesis. Everyday I hear people settling for the well worn 'that was nice/went well' but, as if there were only one way for things to go right but a million for them to go wrong, they are more voluble in describing the latter.
If true, it's as consistent with contentment as disgruntlement, so it implies nothing about whether humans are positive or negatively minded. (Computers I trust are neutrally minded).
Elise K M said,
October 26, 2010 @ 1:44 pm
Possibly irrelevant, definitely a little late, but I thought this might be of interest: http://www.wefeelfine.org/ is a project attempting to "[harvest] human feelings" by scouring blog posts and other online sources for the specific phrases "I feel" and "I am feeling." The results are quite interesting, though I admit I have not examined the site in great analytical depth.
They provide a list of 5000 feelings here, along with their corresponding colours: http://www.wefeelfine.org/data/files/feelings.txt
"The top 200 feelings were manually assigned colors that loosely correspond to the tone of the feeling. Happy positive feelings are bright yellow. Sad negative feelings are dark blue. Angry feelings are red. Calm feelings are green. And so on."
While they do not simply differentiate negative from positive feelings, it may be interesting to examine these top 200's colours in light of Professor Nass's assertions.
Arne H J said,
November 7, 2010 @ 10:01 am
It's often true that there is indeed only "one way for things to go right but a million for them to go wrong". For some reason my mind was drawn to the opening line of Anna Karenina:
Happy families are all alike; every unhappy family is unhappy in its own way.
If you think about a complex operation with a long series of steps it will only succeed if every step in it succeeds; there are many ways to fail but only one to succeed. I'm not so sure that I can agree completely with Tolstoy about families though; it seems to me I have seen many different types of families living that manage to be pretty happy :-)
A quick search found this more appropriate quotation:
It is possible to fail in many ways…while to succeed is possible only in one way.
Aristotle, http://www.quotationspage.com/quote/24233.html