Gender Bias in Text: Origin, Taxonomy, and Implications

Gender inequality represents a considerable loss of human potential and perpetuates a culture of violence, higher gender wage gaps, and a lack of representation of women in higher and leadership positions. Applications powered by Artificial Intelligence (AI) are increasingly being used in the real world to provide critical decisions about who is going to be hired, granted a loan, admitted to college, etc. However, the main pillars of AI, Natural Language Processing (NLP) and Machine Learning (ML) have been shown to reflect and even amplify gender biases and stereotypes, which are mainly inherited from historical training data. In an effort to facilitate the identification and mitigation of gender bias in English text, we develop a comprehensive taxonomy that relies on the following gender bias types: Generic Pronouns, Sexism, Occupational Bias, Exclusionary Bias, and Semantics. We also provide a bottom-up overview of gender bias, from its societal origin to its spillover onto language. Finally, we link the societal implications of gender bias to their corresponding type(s) in the proposed taxonomy. The underlying motivation of our work is to help enable the technical community to identify and mitigate relevant biases from training corpora for improved fairness in NLP systems.


Introduction
Bias is prevalent in every aspect of our lives. We are hardwired to compartmentalize things we experience to form a plausible perception of the world around us. The process of forming these perceptions typically breeds prejudices, which allows for flagrant inequalities to shape across different demographics. The prevalence of certain biases in society, such as gender bias, can be attributed to social roles formed as a function of this compartmentalization process. According to the social role theory, the societal origin of gender stereotypes revolves around gender-typical social roles that mirror the sexual division of labor and gender hierarchy of the society (Bussey and Bandura, 1999).
The prevalence of gender bias in society is also spilled over onto language through the patriarchal worldview predominant among linguists prior to the prescriptive grammar movement in English. Bodine (1975) found that the generic use of he is derived from an androcentric worldview prevalent among 18th-century grammarians: "human beings were to be considered male unless proven otherwise" (Bodine, 1975). The perpetuation of bias onto language entails a negative feedback loop due to the direct impact of language on a person's perceptions (Boroditsky, 2011). Linguistic determinism, a hypothesis taken from the analytic branch of philosophy, posits that language "limits and determines human thought patterns and knowledge" (Hickmann, 2000). Hence, the recurring usage of bias in language consequently leads to a more biased perception which is fed back into our lexical (word) choice. This is even more amplified by the increased adoption of automated system based on AI, which exponentially expedites this feedback loop (as detailed in Section 2.4).
The linguistic spillover of gender bias has various direct and indirect implications on our society. The presence of gender bias in the language used by parents and in school text books causes children to develop sexist perceptions and behaviors towards other children of opposite gender and deepens the problematic outcomes of gender inequalities in society (Waxman, 2013). Additionally, sex-biased wording affects a person's perception of a career's attractiveness (Briere and Lanktree, 1983). Consequently, countries that adopt a gendered language tend to have disproportionate labor force participation (Gay et al., 2013). We also discuss the direct implication of hostile sexism on a person's physiological wellbeing, such as increased stress levels, anger, and elevated cardiovascular reactivity (Schneider et al., 2001). Finally, we examine the indirect implication of benevolent sexism in embedding gender inequality and intensifying its influence in the society by portraying the advantageous aspects of being a woman (deserving special treatment, care, protection, and love) (Hammond et al., 2014;Barreto and Ellemers, 2005).
Gender bias in NLP presents itself in many stages along the design and development process. It can be found in the training data, the pre-trained models, and the algorithms themselves. The propagation of bias from text to features and algorithms leads to real-world consequences when integrated into AI systems and are used in critical decision-making applications. In particular, discriminatory decisions occur when these systems assist humans in critical decisions (Dressel and Farid, 2018). These prejudiced decisions could entail allocational or representational harms (Blodgett et al., 2020b). As mentioned previously, discriminating algorithms accelerate the unavoidable feedback loop, which increases the degree and volume of bias against females and other gender minority groups, especially in online media content. Automated NLP-based decision-making algorithms will re-consume this increasingly growing biased content to update their models, and so on. This feedback loop contributes to an increased gender bias and further discrimination.
Several works in NLP revolving around bias focused on the projection of word embedding vectors on a gender direction (he -she) to detect and mitigate bias in a pre-trained model, without a clear link to the implications on society and their underlying applications (Blodgett et al., 2020b). There has been previous attempts that address bias at the sentence level and provide an initial categorization of gender bias types (Hitti et al., 2019). We build on their work and provide a more comprehensive understanding of the various forms of gender bias while linking to several real world implications on society.
In this paper, we develop a comprehensive taxonomy to identify various types of gender bias.
We also provide a bottom-up overview of gender bias, from its societal origin to its spillover onto language. We then link between the psycho-social implications of gender bias and the corresponding type(s) in the proposed taxonomy. Our underlying motivation is to enable the the technical community working on gender bias in NLP to focus on the identification and mitigation of relevant biases for improved fairness in NLP systems. We also hope that by addressing and linking the sources and implications of gender bias in text, we encourage the community to further push the research in this direction and raise more awareness on bias and discrimination in NLP systems.

Definition
We define gender bias in text as being an exclusionary, implicitly prejudicial, or generalized representation of a specific gender as a function of various societal stereotypes. The sections below provide a bottom-up overview of gender bias, from its societal origin to its spillover onto language while highlighting its perceptual and societal implications.

Social Role Theory
The social role theory posits that gender stereotypes are rooted in the distinct social roles designated to women and men (Bussey and Bandura, 1999). Historically, men and women have maintained diverse social roles: Men have been more likely to engage in tasks that require "speed, strength, and the possibility of being away from home for long periods of time", while women have been more likely to "stay home and engage in family tasks, such as child-rearing" (Eagly et al., 2000). This dispersion comes with various consequences. Firstly, men are perceived as, and expected to be agentic, particularly, active, independent, and resolute, whereas women are perceived as, and expected to be, communal, namely, kind, helpful, and benevolent (Eagly et al., 2000). Secondly, women and men become more inclined to acquire particular skills linked to successful role performance and by adapting their social behavior to role requirements (Eagly et al., 2000). Essentially, both actors and observers are inclined to inherit traits from observed behaviors in their specific social roles (Steffens et al., 2015). This creates an unavoidable negative feedback loop that continuously perpetuates gender bias in society by segregating each gender into a specific social role and actively promotes the divergence through the pursuit of successful role performance.

Linguistic Spillover
Gender stereotypes in society also found their way into language, tunneling through a patriarchal worldview adopted by grammarians prior to the prescriptive grammar movement. Bodine (1975) found that the generic use of he is derived from an androcentric worldview prevalent among 18thcentury grammarians: "human beings were to be considered male unless proven otherwise" (Bodine, 1975). This is also supported by the limited role of women in forming and shaping the English language (Kramarae, 1981). Feminist scholars maintain that the generic he and similar words "not only reflect a history of male domination" but also "actively encourage its perpetuation" (Sniezek and Jazwinski, 1986). The generic he has also intensified sexist behaviors and attitudes in a subtler psychological and perceptual manner. The foundation of this argument is in the Sapir-Whorf hypothesis: "our grammar shapes our thought" (Whorf, 1956). Blaubergs (1980) applies this hypothesis to sexist words and phrases in the English language, including the generic he. She maintains that regardless of its origins,"Sexist language by its existence reinforces and socializes sexist thinking and practices" (Blaubergs, 1980). Consequently, the recurring usage of biased language leads to a more biased perception which is fed back into our lexical (word) choice.

Bias in NLP
Natural Language Processing (NLP) and Machine Learning (ML) techniques, the main pillars of narrow or practical AI, are designed to learn from data and try to generalize the learned concepts to unseen data. However, they are prone to inherit, reflect, and amplify biases and stereotyped-associations that are present in historical data provided for training. Manifestations of different kinds of biases have been shown to exist in various components used to develop NLP and ML systems, from training data to pre-trained models to algorithms and resources (Olteanu et al., 2016;Tolan, 2018;Danks and London, 2017;Mehrabi et al., 2019;Sun et al., 2019;Blodgett et al., 2020b;Hovy et al., 2020;Hitti et al., 2019).
Word embeddings is a family of techniques that learn word representation from texts, such that words with similar meaning have a similar representation (Mikolov et al., 2013b,a). Since their inception, word embeddings have become the predominant representation of text features and an integral part of NLP applications. However, most research on gender bias in NLP has focused on the projection of word embedding vectors on a gender direction (he -she) to detect and mitigate bias in a pre-trained model. For example, occupational gender bias in word embedding models is typically measured by comparing the distances between gendered word vectors and occupational terms. The bias scores resulting from the manipulation of word vectors in a pre-trained word embedding are strictly dependent on the corpus utilized to train that model. Using such models to detect whether new sentences are biased will not only project the biases of the model but also misconstrue its origin (Blodgett et al., 2020b).
The key existing solutions to mitigate these biases focused on modifying the training data, imposing constraints on the word embeddings objective function, or applying post-processing techniques to reduce the bias in word embedding models including word2vec (Mikolov et al., 2013b,a), and GloVe (Bojanowski et al., 2017), and more recently in contextual word embedding models such as ELMO (Hoffman et al., 2010), BERT (Devlin et al., 2018), and ALBERT . Although several other papers discussed different methodologies to debias word embedding model, these techniques have been scrutinized on several occasions (Blodgett et al., 2020a). In addition, the majority of research did not focus on the impact of gender bias in real-word applications (Blodgett et al., 2020a). Automatic detection of gender bias beyond the word level requires an understanding of the semantics of written human language, which remains an open problem and successful approaches are restricted to specific domains and tasks. In an effort to redirect the focus to the linguistic forms of bias and their societal implications, Section 3 contains a comprehensive breakdown of the various gender bias types and their subsequent subtypes, while the next section will be geared towards their societal implications.

Implications
Gender bias leaks into some of the fundamental life aspects and tends to jeopardize the normal func-tioning of the affected gender group (Fraser, 2000). The sections below describe the negative implications of gender bias on children's mental imagery, career attractiveness, labor force participation, and human behavior.

Children's Mental Imagery
Gender bias can manifest itself at an early age in one's life and thus can have a more profound impact on one's attitude and behavior. Children and even infants can be exposed to gender bias presented in language and can also be affected by it, through the process of categorization (Waxman, 2013). The process of category learning begins early on in a person's life and is perceived as a building block for children's lexical acquisition (Waxman, 2013). However, this process could promote stereotypical beliefs and gender biases in children's cognition and perception about individuals, especially if the language used in this process is a gendered language (Bigler and Leaper, 2015). A gendered language which makes gender salient, tends to treat gender as a major attribute upon which children will rely on, to classify and make inferences about others (Hilliard and Liben, 2010). Therefore, the learnt categorizations will promote and perpetuate several forms of gender bias, such as in-group favoritism (Arthur et al., 2008;Bigler and Liben, 2006;Leaper and Bigler, 2004). In-group favoritism can be reflected in children's behavior where a child would prefer to play with another child of the same gender rather than a child with an opposite gender (Fagot et al., 1986).
Gender-generic noun statements, such as "Girls are good at activity X while boys are good at activity Y" that are usually stated by parents and found in school textbooks, influence how children think about themselves. These statements also undermine children's achievements in the relevant activities given their belonging to one of the gender categories (Bigler and Leaper, 2015;Cimpian et al., 2012). In their study, Cimpian et al. (2012) discovered that when children are exposed to gendergeneric statements that link their ability to perform a certain activity to a social group, they tend to perform worse on the given activity irrespective of whether the statement is positive or negative. Cimpian et al. (2012) study implies how threatening gendered generic statements can be in relation to the beliefs that children instantly create about their own capabilities and achievements.

Career Attractiveness
In a study to assess the contribution of biased language relating to the attractiveness of a career, Briere and Lanktree (1983) established that biased language significantly affects a subjects' perception of the attractiveness or employment in a psychology career for women (Briere and Lanktree, 1983). Generic pronouns (as detailed in Section 3.1) and masculine nouns were linked with a decline in the presumed attractiveness of a psychology career for women, with respect to a nonsexist condition (Briere and Lanktree, 1983). Consequently, the use of generic pronouns in texts could discriminatively inhibit female interest in fields they might alternatively seek out (Briere and Lanktree, 1983).
Additionally, a study conducted by Stout and Dasgupta (2011) reveals that gender-biased language in the professional field is associated with negative nonverbal emotional responses from women. Accordingly, women who are exposed to a gender exclusive language during a job interview tend to feel demotivated and socially and actively rejected by the workplace (Stout and Dasgupta, 2011). Other evidence by Vervecken et al. (2013) proposes how children's perceptions of stereotypically male jobs can be influenced by the linguistic form used to present an occupational title. For example, the generic use of masculine plural forms when describing occupations will most likely lead children to restrictive, male only associations and perceptions about stereotypically male occupations (Vervecken et al., 2013).

Labor Force Participation
The gender gaps between women and men in the labor market are almost present in every country, yet with varying degrees, given the cultural norms and values that play a crucial role in introducing or generating new stereotypical beliefs and resisting the existing ones as time passes and cultures change. Aside from the cultural system represented by the social norms and values, a country's adopted language system and the intensity to which it marks gender differences tends to be a very crucial variable in determining the extent to which women can participate in the socio-economic life (Gay et al., 2013). The idea that a country's language system affects women's socioeconomic participation sets off from the idea that language is a key vehicle of the cultural system (North et al., 1990). In their study, Gay et al. (2013) discovered that gendered language has a direct impact on women's socio-economic choices and outcomes. For example, female labor force participation for the year 2000 in countries following a gender binary linguistic system, such as France and Spain, was 16% lower as compared to countries which have no gender marking or have more than three genders in its most spoken language (Gay et al., 2013).

Human Behavior
As stated in the social role theory suggested by Eagly et al. (2000), the gendered roles construct the societal belief system that sets the expectations of men and women, and biased language is instrumentalized to maintain the genders' distinct responsibilities (Stahlberg et al., 2007). As a result, these stereotypical beliefs would be reflected in the everyday lexical choices that refer to men or women, including prejudice or stereotypes that are based on gender or, in other words, sexism (Menegatti et al., 2017). As detailed in Section 3.2, Glick and Fiske (1996) divided sexism into hostile sexism, the typical prejudice against women, and benevolent sexism, the seemingly 'positive' sexism that enforces masculine dominance in the society through viewing women as caring, delicate, emotional, and in need of men's protection (Glick and Fiske, 1996). Bosson et al. (2010) state that women suffer from the emotional impact of hostile sexism for a shorter period of time due to the direct anger expression that's linked to it. Moreover, the exposure to a hostile sexist language motivates women to participate in collective action to stop gender inequality, and it encourages them to socially compete with men in order to reclaim their righteous social status (Becker and Wright, 2011). Nevertheless, hostile sexist language may not have a direct impact on embedding further gendered stereotypes in society, but it has severe direct impact on women's physiological wellbeing, such as increased stress levels, anger, and elevated cardiovascular reactivity (Schneider et al., 2001).
On the other hand, there has been a research consensus on the impact of women's exposure to benevolent sexist language on embedding gender inequality and intensifying its influence in the society (Hammond et al., 2014; Barreto and Ellemers, 2005). For instance, Hammond et al. (2014) indicate that the positive attributes that benevolent sexism holds for women may impair women's opposition to the gendered stereotypes due to how this form of sexism portrays the advantageous aspects of being a woman (deserving special treat-ment, care, protection, and love). Another study shows that benevolent sexist language is often not identified as sexism for many people exposed to it (Barreto and Ellemers, 2005). Thus, this may keep this issue unrecognized and further maintain the acceptance of prejudicial gendered stereotypes, allowing for continuous promoting of sexism and their direct or indirect impact on women (Barreto and Ellemers, 2005).

Taxonomy
The first step of detecting biased language is to categorize the various forms of that bias while carefully maintaining a clear segregation between the resultant groups. The below sections develop a comprehensive taxonomy that includes a wide range of gender bias types and their subsequent subtypes. Each subsection includes the definition of a bias subtype and a couple of examples that illustrate its usage in a sentence. Table 2 provides an overview of the taxonomy, with one example pertaining to each subtype alongside its societal implication (discussed in Section 2.5).

Generic Pronouns
Given that the choice of a pronoun follows the sex of the referent, a problem arises when a pronoun is to be used with sex-indefinite antecedents (Ozieblowska, 1994). Pronouns which do not specify sex are traditionally called "generic", because generic statements about human referents discuss people in general, and therefore the sex of the referents is irrelevant (Ozieblowska, 1994). The most notable forms of generic pronouns are: generic he, generic she, and gendered generic man.

Generic He
The use of the pronoun he in circumstances of sexindefinite reference overly emphasizes men over women, thereby both "re-constituting and signifying males' micro-political hegemony" (Stringer and Hopper, 1998). Thus, generic he occurs when the pronoun he, his and him are used as referents to nouns of no specific gender. Among the gendered generic pronouns, his is the most recurring sexist antecedent to most nouns. Below are some example: • The client should receive his invoice in two weeks.
• A good employee knows that he should strive for excellence.
• A teacher is expected to be a good role model in all areas of his life.

Generic She
While the generic he is the most recurring form of generic pronouns, generic she is also excessively present in written discourse. Below are some example: • A nurse should ensure that she gets adequate rest.
• A dancer should watch her diet carefully.
• She presents us diverge ways, but she lets us choose our path.

Gendered Generic Man
Gendered generic man appears when man is utilized as a masculine noun representation both genders. It's used not only as a noun but also as a verb. Below are some example: • Good teachers know how to man the classroom.
• Effective teachers lead or man the students well.
• It is even more fulfilling when a teacher sees a once stubborn child who became a man of success and responsibilities crown with various achievements.
• All men are born for a reason.
• A teacher is an ordinary man with extraordinary roles.

Sexism
According to the ambivalent sexism theory, sexism against women is divided into an aggressive expression, or hostile sexism, and a positive (for men) expression, or benevolent sexism (Glick and Fiske, 1996). In this section, we will be discussing these two divergent forms of sexist language:

Hostile Sexism
Hostile sexism is the view of men as more powerful and competent than women (Becker and Wright, 2011). It views women as a threat to men's dominance through their violation to traditional gendered roles in the society (Becker and Wright, 2011;Mastari et al., 2019). In general, hostile sexism reflects men's hatred towards women (or misogyny), and it is expressed in aggressive and blatant manner (Connor et al., 2017). Men with hostile sexist mentality view women as manipulative, unintelligent, and incompetent (Jain et al., 2019). Below are some examples of hostile sexist statements: • The people at work are childish. It's run by women and when women don't agree to something, oh man.
• Women always get more upset than men.
• Women are incompetent at work.

Benevolent Sexism
Benevolent sexism is a softer form of sexism that expresses male dominance in a more chivalrous tone (Becker and Wright, 2011). It expresses affection and care for women in return for their acceptance to their limited gendered roles (Becker and Wright, 2011;Mastari et al., 2019). Benevolent sexism describes women as caring, innocent, and in need of men's protection, and these stereotypical notions are used to reinforce women's subordinate position (Connor et al., 2017). This form of sexism explains how women complete men's chivalry, power, and intelligence with their delicate characteristics (Cross and Overall, 2018). Below are some examples of benevolent sexist statements: • They're probably surprised at how smart you are, for a girl.
• No man succeeds without a good woman besides him. Wife or mother. If it is both, he is twice as blessed.
• I am not exploiting women: I love, protect, and care for them.

Occupational Bias
As discussed in Section 2.2, the societal origin of gender stereotypes revolves around gender-typical social roles and thus reflect the sexual division of labor and gender hierarchy of the society (Eagly et al., 2000). The resultant social roles lead to gendered occupational bias, which is a form of generalization that occurs when an occupation or role/duty is generalized onto a specific gender. This section will illustrate both the gendered division of labor and gendered roles/duties.

Gendered Division of Labor
Below are some examples that illustrate how certain jobs are seen as only appropriately and exclusively held by either women or men: • Professors are men and elementary teachers are women.
• Politicians are men and women are wives.
• Housework is the duty of women and an option or out of question for men.
• Scientists are men and secretaries are women.
• Doctors are men and nurses are women.

Gendered Roles/Duties
In the first example below, the speaker's sales assistant is referred to as a girl, which diminishes the status of the role. In the second, the sales assistant is referred to by job title, which indicates that gender is not an important prerequisite for the role that the sales assistant plays.
1. I'll have my girl get you a cup of coffee.
2. I'll ask my assistant to get you a cup of coffee.

Explicit Marking of Sex
Explicit marking of sex occurs when an unknown gender-neutral entity is referred to using genderexclusive term(s).

Gender-based Neologisms
Neologisms are newly coined words/expressions that may be in the process of mainstream adoption, but have not yet been fully accepted. Gender-based neologisms are gendered coinages that could have underlying stereotypical tendencies (Foubert and Lemmens, 2018). Below are some examples: • Man-bread: bread that is baked so big that it will take a grown man a whole week to eat it, having 4 slices a day.
• Man-sip: a man sized sip of a beer or drink, one can finish a beer in 4 or 5 Man-sips. For a female or light weight, it borders on chugging the drink, but for a man it is merely a sip.
• Mantini: a martini or alcoholic beverage that appeals to a man's palate. "My boyfriend prefers his mantini straight up which is just too strong for my tastes."

Gendered Word Ordering
Gendered word ordering is tendency for the male version to come first in binomials such as "men and women", "brothers and sisters", "boys and girls", or "Mr and Mrs". Many words that incorporate the word "man", such as "man-made", "mankind", "manpower", have perfectly acceptable gender-neutral alternatives: for example, "artificial" or "synthetic", "humankind", and "workforce".

Semantics
Gender bias in semantics appears when utilizing words and sentences that are demeaning in their semantic meaning (Umera-Okeke, 2012). The implicit meaning behind sexist jokes, proverbs, or even using specific non-human terms to refer to women, consciously or unconsciously, deepens the existing bias and projects it onto new generations (Umera-Okeke, 2012). The current study suggests three types of semantic gender bias: metaphors, gendered attributes, and old sayings.

Metaphors
People tend to express a part of the world's reality through metaphors, which contributes to ingraining their culture and beliefs. By looking into the window of metaphors, several biases of society are revealed (Rodriguez, 2009). Masculinity and bias against females are represented in metaphoric words that describe women as a non-human comparing females to food, animals, plants (Martín, 2011;Lan and Jingxia, 2019). Below are some examples of English metaphoric words that describe woman as food and animal: • "Cookie": lovely woman • "Old Hen": middle aged women who love to talk to each other

Bias Propagation
Gendered Attributes An unmarried male (bachelor) is a "personal choice". An unmarried female (spinster) is derogatorily an "old maid".

Semantics
Old Sayings A woman's tongue three inches long can kill a man six feet high.

Gendered Attributes
Societal ideologies revolving around each gender role, preferences, interests, and characteristics were originally created due to many historical conditions and various lifestyles, and are conveyed to language in which reflects sexist stereotypes, which might presents invisible limitations for women. Lan and Jingxia (2019) suggest that placing men in a leading position and women as subordinates is the main cause of creating gendered stereotypes (Lan and Jingxia, 2019). Researchers noted that commendatory or complementary terms are used as male words while the corresponding female words are derogatory (e.g. wizard/ witch, spinster / bachelor , governor / governess) (Lan and Jingxia, 2019). Associating positive meaning with male and negative meaning with female represents semantic derogation and disparagement. Here are sentences show the derogatory meaning of some female words: • An unmarried male (bachelor) is a "personal choice". An unmarried female (spinster) is derogatorily an "old maid".
• A "strict male manager" is described as a responsibility taker. A "strict female manager" is described as hard to work with.

Old Sayings
Biased old sayings come in various forms including: proverbs, set-phrases, and formulaic expressions that present a source of stereotype against women. Those sayings are culturally seen as axioms and absolute truth, which affect people behavior to adapt them as moral standards (Martín, 2011). Below are sentences exemplifying implicit sexism in proverbs: • A woman's tongue three inches long can kill a man six feet high • Bad words make a woman worse • When you see an old man, sit down and take a lesson; when you see an old woman, throw a stone

Conclusion
In this paper, we propose a comprehensive gender bias taxonomy that distinguishes between the various forms of gender biases in English text. The taxonomy includes various exclusionary, implicitly prejudicial, and generalized forms of biased gender representations in text. Our work also provides a bottom-up understanding of gender bias, highlighting the social role theory and its impact on gender stereotypes in society. We also explain how societal gender bias spilled over onto language while being fed back into our perceptions as stated in the Sapir-Whorf hypothesis. We hope that our comprehensive taxonomy of gender bias enables the technical community working on gender bias in NLP to focus on the identification and mitigation of relevant biases in text for improved fairness in NLP systems. We also hope that by addressing and connecting the sources and implications of gender bias in text from a linguistic, sociological, and real-life perspective, we would encourage the community to further push the research in this direction and raise more awareness on bias and discrimination in NLP systems. In future work, we will work on expanding the taxonomy to include other languages and address other forms of bias such as racial and ethnic biases.