Skip to main content

Verified by Psychology Today


How Parts of Speech Could Improve Suicide Risk Prediction

Key findings on pronoun, verb, and adjective use.

Key points

  • Some predictors of suicide that have received little research attention are those related to language.
  • Suicidal thoughts may be predicted by linguistic features such as the use of more intensifiers and superlatives (e.g., "very", "never").
  • Suicidal behaviors may be predicted by greater use of nouns, pronouns, and prepositions, and reduced use of numerals and modifiers.
Source: SarahRichterArt/Pixabay

We know that people with suicidal tendencies are more inclined to think and talk about certain topics. For instance, a study of Reddit users—specifically, of the SuicideWatch subreddit—found these individuals made many references to loneliness (“no friend”), anxiety (“I’m afraid”), regret (“never again”), guilt (“I am sorry”), and frustration/hopelessness (“f***ing life”).

The Reddit study also found that the writings of suicidal people were characterized by certain linguistic features, including self-focused attention (“I am/not”), preoccupation with feelings (“make me feel”), use of negation (“no one”), and use of question marks (“Why is mankind afraid of death?”).

In what other ways might the linguistic features of suicidal thoughts and communication be unique (e.g., unusual vocabulary and word choice)? A recent, systematic review of 75 studies and nearly 280,000 participants provides an answer.

The research, by Homan and colleagues, published in the July issue of Clinical Psychology Review, is summarized below.

Analyzing the Language of Suicidal Thoughts and Behaviors

Study inclusion criteria:

“(1) Investigated individuals experiencing suicidal thoughts or who attempted suicide or died by suicide, (2) investigated linguistic features, (3) assessed non-lyrical written or spoken language, (4) cross-sectional and prospective studies, (5) full-text available, and (6) published in English.”

Of the over 700 potential investigations identified, 75 met the inclusion criteria. Roughly 85 percent were conducted in Western countries. Half used clinical populations and half other populations (e.g., students, social media users).

Sample characteristics: 279,032; average age of 30 years old; 35 percent female.

The studies reviewed evaluated the following linguistic characteristics: Prosody/phonetics (27 percent), lexicon (70 percent), and unspecified (3 percent).


Analysis of data showed suicidal thoughts were predicted by increased use of superlatives and intensifiers.

Suicidal behaviors were predicted by higher word count and first-person/second-person pronouns, changes in the number of verbs utilized, increased use of nouns, more prepositions, a larger number of prepend or multifunction words, and fewer numerals and modifiers.

Greater use of first-person pronouns, negative words (e.g., references to death), and negations (e.g., never, any) were also common but less predictive than the above because they were associated with both suicidal thoughts and behaviors.

Applications of Linguistic Features for Predicting Suicide

One application involves machine learning. Machine learning refers to developing computer algorithms that are able to learn and improve on their own (i.e., automatically, through experience).

In this case, the goal would be to build machine learning methods to help distinguish speech/writings with suicidal content from those with non-suicidal anxious or depressive content.

To predict suicide accurately, it is important to collect real-time, real-world data on people at risk for suicide.

In addition, it would be helpful to use voice recordings from therapy sessions, or voice diaries completed between sessions, to determine suicide risk. This can be done in real time, so that if the algorithm indicates a high suicide risk, it automatically activates an appropriate intervention, alerts the patient’s support system, and/or notifies the therapist.

Source: Awaix_Mughal/Pixabay


There are a number of static and dynamic risk factors for suicide, including male gender, Caucasian ethnicity, singlehood, mental illness diagnosis (particularly depression) or symptoms (e.g., sleep difficulties, impulsivity, hopelessness), suicidal ideation, psychiatric hospitalization, personal and family history of suicide attempts, self-injury, low education, physical abuse, sexual abuse, relationship conflict, money problems, and access to weapons.

There are also speech and language predictors of suicidal thoughts and behaviors. For instance, just prior to a suicide attempt, some people are said to sound flat, monotonous, toneless, mechanical, or hollow.

The review by Homan et al., which included 75 trials and nearly 280,000 individuals, showed there are other speech and language predictors of suicidal thoughts and behaviors as well.

Specifically, the investigation concluded that suicidal thoughts were predicted by:

  • A greater number of intensifiers (e.g., very, extremely) and superlatives (e.g., any, never).

Suicidal behaviors were predicted by:

  • Greater word count.
  • Changes in verb usage, such as using fewer future tense verbs.
  • Fewer modifiers and numerals.
  • More nouns and prepositions.
  • Increased use of first- and second-person pronouns.
  • More prepends and multifunctional words.

A number of features—e.g., greater use of first-person pronouns, negative words (such as death references), and negations (like "never", "any") were associated with both suicidal thoughts and behaviors.

Future research needs to examine why these linguistic features predict suicide. But just to speculate, consider a possible reason for the reduced use of modifiers:

Modifiers are used in logically complex sentences requiring higher-level thinking. The use of fewer modifiers might suggest mental rigidity and tunnel vision, which are often seen in people with a fixed goal in mind—in the case of people with suicidal tendencies, the fixed goal of ending their own life.

Facebook/LinkedIn image: Dmytro Zinkevych/Shutterstock