December 17, 2003

Passive voice and bias in Reuter headlines about Israelis and Palestinians

The organization Honest Reporting recently released a study of bias in Reuters news agency headlines about events in Israel and Palestine. The part of the study on "Verb selection" claims that the choice between active and passive voice is being used to make Israeli violence more overt and apparent and Palestinian violence less so. The report says:

Violent acts by Palestinians are described with "active voice" verbs in 33% of the headlines.

Violent acts by Israelis are described with "active voice" verbs in 100% of the headlines.

Unfortunately, whatever the validity of the data on which the claims are based, the accuracy of their linguistic analysis is wrong two-thirds of the time in the examples that they give.

Here are their three examples:

Example 1:

"Israeli Troops Shoot Dead Palestinian in W.Bank" (July 3)
Israel named as perpetrator; Palestinian named as victim; described in active voice .

  vs.

"New West Bank Shooting Mars Truce" (July 1)
Palestinian not named as perpetrator; Israeli not named as victim; shooting described in passive voice.

Example 2:

"Israel Kills Three Militants; Gaza Deal Seen Close" (June 27)
Israel named as perpetrator; Palestinians ("Militants") named as victims; described in active voice.

  vs.

"Bus Blows Up in Central Jerusalem" (June 11)
Palestinian not named as perpetrator; Israelis not named as victims; described in p assive voice.

Example 3:

"Israeli Tank Kills 3 Militants in Gaza - Witnesses" (June 22)
Israel named as perpetrator; Palestinians ("Militants") named as victims; described in active voice.

  vs.

"Israeli Girl Killed, Fueling Cycle of Violence" (June 18)
Palestinian not named as perpetrator; killing described in passive voice.

The evidence of bias may seem clear enough (I won't be evaluating that here), but this is Language Log, and -- forgive me for being a pedant, but it is part of my job description -- I have to point out that only one out of the three examples here actually illustrates the passive voice.

Example 1. "New West Bank Shooting Mars Truce" is entirely active: the main verb is mars. The subject noun phrase is new West Bank shooting. The word shooting here is a nominalization -- a noun derived from a verb root (notice, you can talk about two shootings: it actually takes the plural marker -s like any other noun). Nominalization is one way to avoid reference to the agent of an action (here, who did the shooting), but it's not the same as using the passive voice.

Example 2. "Bus Blows Up" is indeed a strange way to describe an incident in which a human being straps explosives to himself, gets on a crowded bus in a city street, and kills 13 people by detonating his payload, clearly intending to murder as many Jews as possible at one go. However, there is no passive construction here. The predicate is active and intransitive. ("Bus is Blown Up" or "Bus Blown Up" would have been a passive.) What's weird is that a reference to the bus is used as the subject of this intransitive predicate. Reuters describes the event as if the bus had just exploded all on its own. But not with a passive.

Example 3. The third example is the only one with a passive verb: "Israeli Girl Killed" has the past participle killed used as the verb of a passive verb phrase: in a fuller (non-headline) form the sentence would be "Israeli girl is killed". There is no by-phrase following, so there is no reference to who did the killing; this is the point that Honest Reporting complains about in Reuters headline phrasing. The thing about the passive construction that makes it convenient for suppressing reference to perpetrators is that preposition phrases with by are almost always optional: you can leave them out without the result being ungrammatical. If you use a tensed active verb it's not so easy to suppress the identity of the actor, because subjects are obligatory in tensed clauses: Palestinian gunman kills Israeli girl would be grammatical, but *Kills Israeli girl would be ungrammatical. (A few newspapers do use subjectless tensed headlines -- I've seen it in the Chicago area -- but most do not.)

Honest Reporting is claiming that Reuters uses active and passive verb phrases differentially in its headlines, often suppressing facts of Palestinian agency in violent acts, but literally never suppressing the fact of agency when Israelis or the Israeli state are involved. If their analysis of the data is accurate, this deserves explanation. There ought to be no gross nationality difference in the frequency with which constructions making reference to the agents in acts of violence are used -- certainly not a difference as staggeringly large as 33% versus 100% according to whether Palestinian or Israeli violence is involved. But this sort of propaganda analysis would be best done by people who have a clear grasp of basic traditional grammar, so that when they refer to the use of passive voice they know what they are talking about and can give examples that do indeed show passive clauses.

Credibility is everything in studies of this kind. Honest Reporting cannot possibly claim to be non-partisan: they are avowedly devoted to the cause of righting what they see as a shocking anti-Israeli bias in the western media. So we can only trust that they are living up to the first word of their name if they are scrupulously accurate when they do their deliberately pro-Israel advocacy and analysis. When we find that they can only identify a passive verb 33% of the time, in an analysis that is explicitly about how many times the passive voice is used, it shakes our confidence in the accuracy of other aspects of their analysis too (perhaps quite wrongly).

Footnote added later: There could be other factors accounting for the numerical discrepancies, of course. Anthony Hope has pointed out to me that when the Israeli state does something the identity of the agent is known immediately, but Palestinian-initiated acts of violence are often hard to attribute to a specific person or group in the first few hours. Chris Potts points out a linguistic issue: the word "Palestinian(s)" is longer than the word "Israel(i)" by a factor that would be nontrivial in headline composition, where every millimeter of column width counts. Both these factors could be in play. I'm not suggesting otherwise when I observe that Honest Reporting's data needs explanation. By the way, they discuss many different kinds of bias on the part of Reuters, not just choice between actives and passives.

Posted by Geoffrey K. Pullum at December 17, 2003 12:30 AM