This paper presents preliminary corpus-based research of special words of the Russian language - parametric adjectives. Parametric adjectives are adjectives whose semantics are associated with the denotation of physical parameters such as size, length, width. Characteristics of parametric adjectives seem to be poorly investigated from the perspective of Sentiment analysis. Therefore, our research is the first step in this direction. We identify and classify the most common contexts for adjectives belonging to four semantic classes such as size, height, weight, and strength.
In recent years, the Internet is rapidly developing, including its Russian-speaking segment. In our daily life we face a variety of opinions every day: we read reviews before we buy anything, click “like” button, write comments, read news. We are surrounded by the world of opinions, the world of rating. Society in the modern world is highly exposed to the evaluation activity both in culture and discourse. This phenomenon has become so widespread and has made essential the need for its careful exploration, using the analysis of the sentiment-related words of the Russian language. Different sciences are engaged in research on sentiment analysis: from philosophy and axiology to psychology, political science and linguistics. Linguistics studies the means for opinion and sentiment expression in text and speech at all levels of the language system: phonetics, morphology, lexicology, and syntax.
Opinions and sentiment can be expressed most clearly and fully on the lexicology level. Sentiment-related words can explicitly express positive or negative sentiment in a text or speech.
Sentiment analysis is one example of the practical use of sentiment-related
words [
One of the most numerous classes of sentiment-related words is adjectives. Among evaluative adjectives, parametric adjectives occupy a special place. Parametric adjectives are adjectives whose semantics are associated with the denotation of physical parameters such as size, length, width. Parametric adjectives in Russian are qualitative adjectives with a nominative size value.
This class of words deserves a separate study, since they form a lexico-semantic class of words, which has distinctive features: 1. They form antonymic pairs, expressing opposite values on the same mental scale.
For example, bolshoy (big) - malenkiy (small), vysokiy (high) - nizkiy (low), uzkiy (narrow) - shirokiy (wide). 2. Parametric adjectives can reverse their polarity depending on the context. As for the first point, the semantics of parametric adjectives are based on the idea of the parameter of objects as a quantity whose values serve to distinguish between objects of a certain subclass.
Words marking the extreme elements of the mental scale can be defined as
follows: for example, large – “one that is larger than normal”, low – “one that is lower
than normal” [
Very often, the use of one or another parametric adjective is subjective, since there is no measurement standard. A person, depending on the situation, can choose his own standard. Thus, different persons can use words with the opposite meanings, describing the same objects in the same situations. For one person, the volume level may seem high, and for another person it may be low.
The second feature is related to the fact that parametric adjectives are able to change their polarity to the opposite, depending on which aspect they relate to. For example, the adjective “small” has a positive polarity in combination with the “price” aspect and a negative polarity in combination with the “memory capacity”. In this case, subjectivity cannot be avoided too. For example, two people can consider an object small (for example, a tea-pot), but for one it will be a positive moment, and for another it will be negative.
In our study, we will focus on the study of the reverse polarity. The relevance of the study is determined, on the one hand, by its belonging to the field of cognitive linguistics. And on the other hand, by the applied aspect, namely the possibility of using the results of research in Sentiment analysis systems in order to improve the quality of analysis. The purpose of the article is to analyze and systematically represent the most frequent contexts of parametric adjectives, indicating their most probable polarity in these contexts on the material of product reviews. All of the above makes the task of the parametric adjectives contexts study very relevant now. 2
There are many studies of parametric vocabulary, and basically most of them have been conducted in terms of comparative linguistics, cognitive linguistics, linguoculturology or psycholinguistic.
Mikheeva S.L. in her study [
In the work [
The article [
The paper [
M.S. Achaeva in her work [
The article [
We were able to find only a small number of works studying parametric vocabulary from the perspective of applied linguistics.
The study [
In the paper [
The research results show that parametric words can expresses the opinion implicitly. Some of the parametric vocabulary may be assigned to one of the main classes: positive or negative. Such classification is specific to the given subject sphere. And some of the parametric vocabulary refers to the auxiliary classes (increments, decrements, modifiers).
Increments are words that enhance the polarity of other words in a sentence (for instance, ‘very’). Polarity modifiers are words that reverse the polarity of other words in a sentence (for instance, ‘not’). Decrements are words that cancel the change in polarity, despite the presence of polarity modifiers in the sentence (for example, the word ‘so’ in the sentence “I’ve never been so deceived”).
Parametric adjectives sense disambiguation task is similar to the well-known word polarity disambiguation task which aims to resolve polarity of the sentiment ambiguous words in a specific context.
One of the first appearances of the disambiguating word polarity task was in
competition SemEval-2010 [
This problem is still under an active investigation. Xia, Y., Cambria, E.,
Hussain, A., & Zhao, H. [
As can be seen from the above, while investigations of adjectives polarity disambiguation were conducted in the past, very few systematic and fine-grained studies of the parametric adjectives contexts are available at the moment, especially for Russian language. 3
For the study in this paper, we use the set of reviews for products placed on the Internet resource Yandex.Market. For automatic reviews extraction the program in python was developed that uses the API Yandex.Market.
In this way, we’ve constructed a corpus of 41913 reviews (4 739 010 usage) on 32 categories of products.
For representativeness, we have chosen a variety of product groups that are not like each other. So, we have selected several groups of products from the categories: household appliances, electronics, health and beauty goods, goods for children, pet goods, goods for hobbies and leisure. The distribution of the number of reviews by categories is presented in Table 1. We have chosen to investigate the most frequent parametric adjectives and have extracted for them all the contexts in which they appeared in the corpus: previous and next word.
Word большой (big) )маленький (little) ) лёгкий (light) высокий (tall) )небольшой (small) тяжёлый (heavy) низкий (low) слабый (weak) сильный (strong) 41913
The category “size”: ‘большой’(large) – ‘небольшой’/ ‘маленький’ (little/small) ‘Большой’(large) 1. The first context for the word ‘большой’(large) is significant in physical size. This adjective often expresses a positive assessment in combination with the component parts of the object. For example, ‘большой экран’ (large screen) ‘+’. But it also can be used in a negative context. For example, ‘большие габариты велотренажера’ (the huge size of an exercise bike) ‘-’. There is a connotation “takes a lot of space”. 2. The second context of the use of the adjective ‘большой’(large) is the expression of the characteristics of the physical volume. For example, ‘большой объём резервуара’ (large volume of the tank) ‘+’. In this context, this word usually has only positive polarity. 3. The adjective ‘большой’(large) is often found in a combination with the words expressing that one can choose from something or assortment/variety. For example, ‘большой выбор цветов’ (a large selection of colors) ‘+’, ‘большой набор функций’ (a large set of functions) ‘+’. In this context, this word also has only positive polarity. 4. The fourth context is the use of an adjective with nouns characterizing advantages and disadvantages. In this context the adjective ‘большой’(large) has an amplifying meaning and can have both positive and negative polarities. For example, ‘большой плюс’ (the great advantage) ‘+’, ‘самый большой минус’ (the biggest disadvantage) ‘-’. 5. The last context includes all other uses, most often expressing technical and time characteristics. The common meaning of the word ‘большой’(large) in this case is significant in strength, intensity or duration. For instance, one can find phrases ‘большое время работы от аккумулятора’ (long battery life) ‘+’, ‘большая мощность’ (high power) ‘+’, ‘большой объем памяти’ (large amount of memory) ‘+’, ‘большой шум’ (a lot of noise) ‘-’, ‘большой люфт’ (a big backlash) ‘-’. ‘Небольшой’/ ‘маленький’ (little/small) For words ‘небольшой’/ ‘маленький’ (little/small), the opposite tendency is observed: in almost all cases when large will have a positive polarity, ‘небольшой’ and ‘маленький’ (little and small) will have a negative one. For example, ‘маленькая цена’ (small price) has a positive polarity.
Both the positive contexts and the negative contexts for adjectives ‘большой’ (large) – ‘небольшой’/ ‘маленький’ (little/small), are available, see Table 3. Type of context
Context
Example Positive contexts Negative contexts Positive contexts Negative contexts
большой (large) The size of the component экран (screen), горлышко(neck), parts of the object колеса(wheels) Volume объем загрузки(loading volume), вме
стительность(capacity) Time and technical charac- время работы от аккумулятора (operteristics ating time from batteries), время работы в автономном режиме (work offline) громкость звука (sound volume) Assortment/variety выбор цветов (choice of colors),
функционал(functional) Pros and cons плюс (plus),преимущество (advantages) The size of the component габариты автокресла (car seat dimenparts of the object
sions) Time and technical charac- вес велосипеда (bike weight), вес teristics смартфона (smartphone weight) расход воды (water consumption) шум при работе (noise at work) Pros and cons минус (minus), недостаток (
disadvantage) небольшой/маленький(little/small) Time and technical charac- вес ноутбука (laptop weight) teristics потребление электроэнергии (electrici
ty consumption)
parts of the object The size of the component кнопки (buttons), морозилка(freezer) parts of the object Volume объем резервуара (tank volume), объем чаши (bowl volume) Time and technical charac- время работы без подзарядки (work teristics without recharging)
мощность (power)
Assortment/variety набор программ (set of programs) 4.2
The category “height”: ‘высокий’ (high) – ‘низкий’ (low) ‘Высокий’ (high) 1. The most frequent context of using the adjective ‘высокий’ (high) is the intention to emphasize the intensity of the sentiment feature. It usually has a positive polarity. For example, ‘высокое качество’ (high quality) ‘+’, ‘высокая износостойкость’ (high wear resistance) ‘+’. 2. Also quite often the adjective ‘высокий’ (high) evaluates the level of technical characteristics. It can have both positive and negative polarities: ‘высокое разрешение экрана’ (high screen resolution) ‘+’, ‘высокая светочувствительность’ (high light sensitivity) ‘+’, ‘высокий уровень шума при работе’ (high noise during operation) ‘-’. 3. The next context is the price characteristic: ‘высокая цена’ (high price) ‘-’. It always has only negative polarity. 4. In its main meaning - long extension from bottom to top – the adjective ‘высокий’ (high) is used much less frequently in reviews. For example, ‘высокие бортики коляски’ (high sides of a stroller) ‘+’. 5. In reviews on food or nutritional supplements, animal feed, stable combinations with the adjective ‘высокий’ (high) are used, indicating the composition of the product. For example, ‘высокое содержание белка’ (high protein content) ‘+’, ‘высокое содержание химикатов’ (high chemical content) ‘-’. ‘Низкий’ (low) The adjective ‘низкий’ (low) in almost all cases will have opposing contexts of use. In a positive context, it will most often denote a price, and in a negative context, it will indicate a low degree of positive tonal sign, for example, ‘низкое качество сборки’ (low build quality).
Both the positive contexts and the negative contexts for adjectives ‘высокий’ (high) – ‘низкий’ (low), are presented in Table 4. поддон (pallet) содержание минералов(content of minerals)
The category “weight”: ‘тяжелый’ (heavy) – ‘легкий’ (light) ‘Тяжелый’ (heavy) 1. The most frequent for the word ‘тяжёлый’ (heavy) is weight of the object. This adjective can express a positive sentiment. For example, ‘тяжелый металлический корпус смартфона’ (heavy metal case of the smartphone) ‘+’. In such cases, there is a “strong”, “reliable” or “steady” connotation. It also can express negative sentiment: ‘тяжелое зарядное устройство’ (heavy power bank) ‘-’. 2. This adjective may also be used in negative contexts expressing difficulty in implementing a process or the need to put a lot of effort. ‘Легкий’ (light) This adjective can occur in all three contexts described above as the adjective ‘тяжёлый’ (heavy).
But also, it can express the intensity of the negative sentiment feature: ‘легкий скрежет’ (light rattle) ‘-’, ’легкий запах пластика’ (light plastic smell) ‘-’, ‘легкое дребезжание’ (light chatter) ‘-’.
Both the positive contexts and the negative contexts for adjectives ‘тяжелый’ (heavy) – ‘легкий’ (light), are available in Table 5.
Difficulty in implementing a настройка(adjustment), process очистка(cleaning) лёгкий (light) Simplicity in implementing a навигация по меню(menu navigaprocess tion), эксплуатация(exploitation) Weight ноутбук(laptop) It takes a little of effort
ход(baby carriage run) Intensity of the negative sentiment feature скрежет(rattle), запах пластика(plastic smell)
The category “strength”: ‘сильный’ (strong) – ‘слабый’ (weak) 1. The first context for the word ‘сильный’ (strong) is significant in physical strength, powerful. For example, ‘сильный адаптер’ (strong adapter) ‘+’. In this context it always expresses a positive sentiment. 2. The second context of the use of the adjective ‘сильный’ (strong) is impressive.
For example, ‘сильный дизайн’ (strong design) ‘+’. In this context it also usually expresses a positive sentiment. 3. The adjective ‘сильный’ (strong) is used in context of the intensity of the sentiment feature. Most often this is a negative context: ‘сильное искажение’ (strong distortion) ‘-’, ‘сильный нагрев’ (strong heat) ‘-’. ‘Cлабый’ (weak) We did not reveal explicit positive contexts for the adjective ‘слабый’ (weak). In negative contexts, the adjective is most often used to describe the construction, assembly or for expression the intensity of the negative sentiment feature.
Both the positive contexts and the negative contexts for adjectives ‘сильный’ (strong) – ‘слабый’ (weak) are shown in Table 6. Several observations can be made on the basis of the results presented above. First, some contexts of the given adjectives can carry varying polarities: positive or negative. Other contexts have only one polarity – either positive or negative. Second, disambiguation contexts of the same words provide information about the semantics of these words. Besides, it partially solves the polarity disambiguation problem for those classes that carry only one polarity. 5
We set the task to automatically determine the contexts the parametric adjective «big». It was chosen as the most frequently used parametric word in our corpus. We formed a dataset of 750 sentences in which the word "big" is used across the 29 domains. The dataset was divided into training and test. There were 375 sentences in each dataset. Both datasets were manually tagged with the target cluster.
We tokenized the sentences and extracted aspect terms for word «big» for each
sentence using UDPipe [
This study is the first step in the study of parametric vocabulary of customer reviews in the perspective Sentiment analysis. Parametric adjectives can reverse their polarity depending on the context, which makes it difficult to automatically determine their tonality.
In this paper, we highlighted the most common contexts for adjectives of four semantic classes such as size, height, weight, and strength. Our experiment demonstrates that it is possible to determine the context automatically using machine learning with the sufficiently high precision.
As future work, we can conduct a more detailed quantitative study of the use of parametric adjectives in different contexts and the development of rules or other methods for automatic determining their polarity depending on the context.