Experiment in psycholinguistics.

Part I. Experiment in psycholinguistics

0. Introduction

At present, it seems, we can talk about the desire of psycholinguistics to expand their space: the scientific direction of psycholinguistics turns out to be oriented both to linguistics, and psychology, and to physiology, and to “cognitive science”, and to many other related sciences. man. " At the same time, the often interdisciplinary nature of psycholinguistics leads to the development of conceptual and methodological contradictions.

At the same time, right now we are witnessing the process of active introduction of psycholinguistics into general educational programs of universities, the emergence of a large number of specializations and programs of “Psycholinguistics” at universities and pedagogical universities of Russia. Such a process puts the task of explaining and finding ways to overcome these contradictions among the priorities.

Experiment is the cornerstone of psycholinguistics: “Psycholinguistics is, first of all, experimental science” (Sugar Academy 1989) or “In essence, psycholinguistics is nothing more than experimental linguistics” (Leontiev 1976). However, there are still significant disagreements over what is an experiment in the general scientific space and an experiment in the psycholinguistic space. If even among psycholinguists with experience there is no consensus about the experiment, then what about those who only recently joined the ranks of psycholinguists and teachers of psycholinguistics?

1. Experiment: the controversy between "traditionalists" and experimenters

Let us turn to the recognized authorities in this question: linguistic encyclopedic dictionary and R.М. Frumkina as the author of the relevant dictionary entry.

"Experimental methods in linguistics - methods that allow to study the facts of the language in conditions controlled and controlled by the researcher" (active experiment, or an objective according to R.Frumkina). However, the nature of the facts and phenomena under consideration is so complex and diverse that not everything can and should be “managed and controlled,” the extent and method of such management and control are determined by the goals and objectives of the study. “... the term“ EM ”is not clear; linguists often speak of experiment where observation takes place, above all observation of texts ... <> Object Em. is a person who is a native speaker, generates texts, perceives texts and acts as an informant for a researcher ” (Frumkina 2002).

Experiment in the broad sense is traditionally contrasted with either naked theory or the results of equally naked computer simulation, in which the researcher does not ask himself whether the procedures embodied in computer algorithms and the procedures of language and human activity are appropriate.

We believe that if the opposition of an experiment (active experiment) to observation (a passive experiment) exists, it can be considered in some way as gradual - depending on the number of parameters and the degree of control and monitoring.

Consider this question in more detail. Each native speaker reflects the peculiarities of human speech activity in general, the carrier of a given national language, a certain social, age, gender class, has individual language and textual picture of the world (Kasevich 1989; Kasevich 2004), speech experience and numerous acquired speech strategies, individual psychophysiological characteristics ( eg, hemispheric dominance, memory, attention, etc.), many of which may depend on fatigue, external influences, year and day, etc.

In the course of human speech activity, a large number of signs — phonetic and nonphonetic (visual and non-visual), linguistic and extralinguistic — are set by decision-making, defined both by the actual speech stimulus and the communicative situation. At the moment we do not even know the exact set of these signs, not to mention the decision-making procedures in which these signs are involved.

Consequently, the complexity of the phenomena under study, on the one hand, does not allow one to be satisfied with the results of pure observation (one would have to have an incredibly long observation over a huge number of people in various linguistic and communicative situations). On the other hand, the diversity of parameters and features that influence human language and speech activity does not allow one to imagine an objective experiment in which all parameters and conditions can be monitored and controlled. Conditions are one of the most significant moments, since Different communicative (including experimental) conditions can activate different speech strategies. We may well assume that during our experiment, those speech strategies that will never work in natural communication will be activated. And even in different situations of natural communication, different strategies may be involved.

Why is the question of the difference in strategies often found to be crucial when planning an experiment, and even more so when interpreting experimental data? First of all, because the results of any experiment can be considered only in the context of modeling certain speech strategies. It is only in this context that we receive data on how the characteristics of a speech stimulus are connected into a single whole with the responses received from the subjects. Outside of this context, the units, properties, signs of the answers of the subjects will be uninterpreted, for example, the same answers outside the context in question may be evidence of the work of various mechanisms and procedures.

We think everybody is familiar with an example of a classroom learning a foreign language: as a rule, in the audience and in natural communication with foreign friends we speak differently (different number and type of mistakes, different texts, different “sensations” from the conversation). The various tests used in teaching practice can be correlated to a certain extent with the experimental conditions, and the results of the tests are necessarily related to the teaching methods used previously. In the course of performing various tasks, we often find ourselves catching (and generating) texts we are offered in different ways.

As an illustration, let us imagine the difference of goals and procedures in the work with the text of the reader, editor and teacher; a person who perceives the well-known (for example, “Chicken-ryaba”) and an unknown text (for example, a new work by the avant-garde writer); easily perceived text (on a well-known topic, within the framework of the developed domain) and poorly understood text [1].

We believe that at the moment, psycholinguists can claim, mainly, a model description of speech strategies, the role of different linguistic, verbal, communicative, cognitive, psychophysiological and other signs and parameters, decision-making procedures. We will not reach a strict model of speech activity very soon. In this respect, the use of experiment and observation (passive experiment) is similar.

Apparently, an integrated (or integrative) approach, including a set of various experimental techniques, including observations, and instrumental analysis of speech products (received and / or generated texts), can be considered one of the most successful ways to solve the problems posed.

In general, it can be said that the “ideal and complete” psycholinguistic research contains a combination of observation and experiment methods, based on the results of which a model (theory) is built, including a functional computer-implemented model (program). In the case of a deliberately incomplete research (which applies to almost all of our work, theories and models), the purpose of the research determines the choice of certain methods and / or their combinations.

2. What objections is caused by an experiment in psycholinguistics [2]

Based on my experience of communicating with colleagues, representatives of other sciences, reports at conferences and managing research qualifying works of students, undergraduates, working with graduate students, I will try to imitate the following dialogue. You, probably, at conferences, on the protection of your own and others' qualifying works, just in the course of scientific conversations you will have to hear such questions more than once.

“You competently conducted an experiment, got reliable results. So, what is next? Nothing follows from all this, because it is nothing more than the results of your experiment. ” Such a statement can be heard from both the “naked theorist” and the advocate of pure observation. However, this statement already contains a contradiction: a competent experiment (especially a series of experiments) should give a competent answer to the question posed, which is often associated with the requirement of repeatability of results in comparable conditions. Otherwise, we are dealing with either an illiterate experiment or an unfinished experimental series.

The second question seems completely natural: “You conducted an experiment, got some results. So, what is next? Nothing follows from all this, because to obtain objective results it is necessary to take into account some more parameters and conditions (for example, an increase in the number of subjects) . ” Well, the experimental approach requires a lot of time, effort and luck. It is possible that at this stage of your research work you will only be able to conduct a preliminary study, test the methodology and set new tasks for yourself and others. This is also the result.

The third question relates to the ability not to distract from the goals and objectives of the study, either at the planning stage of the experiment, or at the stage of statistical processing of results, to the ability to see behind the numbers a model description of linguistic and psychological facts and phenomena (that is, this question also applies to works constructed based on observation and instrumental analysis). The third question is especially relevant at the stage of processing the results, because an experienced manager passes the planning stage, holding the beginning researcher “by the hand”. “Elena Viktorovna, here are the numbers. And what next? " . And this question does not cause a grin. First, the issue of processing the results is extremely complicated, requires a variety of knowledge (linguistic, psychological and statistical) and skills, and secondly, the numbers often act hypnotically on an unprepared person: it seems that the research process is to collect numbers, and then they get lost.

If the researcher made an inaccuracy at the planning and preparation stage of the experiment, these could be issues of incorrect selection of stimuli, inaccurate instructions, unsuccessful selection of subjects and control of the experimental conditions as a whole — all work can be thrown away. This is the main danger that awaits the careless experimenter.

Not without reason many “traditionalists” at best treat the experimental work with distrust: unfortunately, a fair number of diplomas, dissertations, reports, articles give them grounds for this [3].

We should also mention the need to use an adequate statistical method for analyzing the data obtained and the accuracy of its application (see Sidorenko 2004).

3. Experiment in psycholinguistics and its main types

3.0 The traditional scheme of psycholinguistic experiment:

S → █ → R, where S is a stimulus, R is the reaction of the subject, and the connection between them is through a “black box”, which symbolizes the incomprehensibility of the person being studied, an active participant in speech and thought activity. Under the stimulus can be understood as directly stimulus material, and stimulus material in combination with the conditions and methods of the experiment (for example, instructions).

The proposed scheme corresponds to any psycholinguistic experiment. It is impossible to directly trace the most complicated procedures of human's cogitative thinking: they take place both on a conscious and an unconscious level, a person is not able to realize what is happening and how when he performs speech activity of any level of complexity. Not able to give a clear answer to this question and the various methods of measuring the physiological and neurophysiological changes that occur with the subject. That is why for the study of human speech activity we can only use the “black box” method, when we can judge the procedures, units and features by comparing the structure of output values ​​(reaction) with the corresponding structure of input (stimulus), and, as we said earlier, The only way to do this is to simulate the transition from stimulus to reaction in the context of the intended speech strategy.

The language and speech experience of a person allows him to easily cope with the most difficult tasks of speech activity. What does the researcher do?

3.1. Experiment in the study of speech perception

3.1.0. General requirements

Problems of speech perception are worthy of separate textbooks, textbooks and monographs. Apparently, the solution of such problems is especially vital for creating an experimental methodology, since it is obvious that it is with perception that any activity begins. However, in the framework of this textbook, we allow ourselves only briefly to focus on an experiment in this area, focusing on the perception of the voiced speech.

When carrying out any experimental work, it is advisable to remember the basic principles of the psychology of speech, for which we recommend that every aspiring psycholinguist read Alla Solomonovna Stern's textbook “Introduction to Psychology” (Stern 2003), which includes very simply outlined the main provisions of the psychology of speech.

At the moment, there are many theories of speech perception, so far none of them can claim to be complete, but progress in solving the questions posed is very significant. Until recently, there were discussions about the passive or active nature of perception, and supporters of the passive nature of perception (as opposed to the active nature of generation) had considerable weight. At the same time, passive character means passive unidirectional movement from sound to phonemes and further to meaning, and active character consists in active pursuit of meaning, which (along with sound) is included in the work already at the initial stage of perception and allows, on the one hand, to cope with substantial incompleteness and variability of acoustic information (sound), and on the other hand, significantly increase the speed of information processing.) [4].

Nowadays, ideas about unidirectionality and obligatory ponemone perception of a sounding text have irrevocably faded into the past, now we are confidently talking about the coexistence of two directions of perception “from the bottom up” (from the sound) and “from top to bottom” (from the meaning). The ratio of these two directions depends on the chosen strategy of perception, and the choice of a particular strategy is determined by a variety of parameters determined by both individual characteristics and the stimulus, along with the conditions of the communicative situation.

When constructing any descriptions of speech perception processes, it is extremely important to take into account such an important characteristic of its mechanisms as leveling. It would seem that this level level of mechanisms and the hierarchical organization of any system of units and attributes naturally follow from all our knowledge of both psychology and linguistics. However, this requirement forces the researcher at every moment to look for answers to questions such as what level is the subject of study for us, what level of units we have to compare the answers of the subjects, what operational units were chosen by the subject within the framework of his chosen perception strategies . Despite the fact that different levels of perception are used simultaneously, their ratio may vary depending on the goal and the possibilities of achieving it. For example, writing down sounding text (dictation), the subject can write down precisely meaningful text (if he manages to use text level attributes), meaningful words (if he manages to use units and word level attributes, that is, identify words) or words using phonetic records (if the words cannot be identified, in particular, by writing meaningless or unfamiliar words). Note that with an increase in the leading level, the significance of lower levels decreases (although it is never reduced to zero).

When perceiving a sounding text, the number of important features used by a person turns out to be quite large (their set, hierarchy and interaction are still unknown), the processing of these signs, as neuro-linguistic studies show, is carried out largely parallel, through different channels.

The redundancy inherent in the language, the mechanisms of contextual predictability and probabilistic forecasting make it easy to cope with possible difficulties in the perception of the text. Thus, everything that certainly helps us as a native speaker in every possible way makes the work of a researcher difficult.

What remains for the researcher? Mainly, acting on the stimulus, try to distort the speech signal, removing, weakening some signs or organizing a situation of conflict between different signs.

When considering various methods, we suggest you to pay attention to the choice of one or another method, depending on what level is the subject of study for us. We note here that prosodic signs, as well as semantic ones, belong to levels from word to text (the syllable units and the morpheme will not be considered for simplicity).

3.1.1. Some traditional techniques

Prosodic and semantic features are increasingly becoming the subject of study in studies of speech perception. Naturally, in the course of such a study, the problem of removing (weakening) of such symptoms most often arises.

• Often used noise - full or partial, for example, the whole text or its fragments, isolated words or their segments are noisy to a greater or lesser extent. What to use as noise is solved by the researcher in relation to the research task, but white noise is most often used. , в нем присутствуют частоты всего речевого диапазона с одинаковым энергетическим вкладом. Соотношение между полезным речевым сигналом и шумом задает исследователь (часто для его определения надо проводить предварительные эксперименты), чем выше уровень речевого сигнала, тем слабее искажение. Эта методика часто используется при оценке помехоустойчивости тех или иных признаков. Особенно плодотворной она кажется при исследовании просодических структур (просодических характеристик), как наиболее помехоустойчивых и с точки зрения восприятия (результатов эксперимента), и с точки зрения инструментального анализа. Частота основного тона (а также форманты гласных) в речевом сигнале обладает максимальной интенсивностью, т.е. можно условно сказать, что она искажается при зашумлении значительно меньше, чем шумовые фонетические характеристики глухих согласных. С помощью компьютерных программ работы со звуковым материалом решение задач зашумления оказывается очень простым. Например, в стандартной программе Cool Edit Pro шум синтезируется с помощью опции “Generate”, а накладывается на речевой сигнал с помощью “Mix Paste”. При наличии у Вас компьютера Вы сможете опробовать разнообразные способы зашумления речевых сигналов. Проверку того, что получается в результате зашумления, Вы можете осуществить, используя не только слуховой, но и инструментальный (спектральный) анализ, для чего Вам надо переключиться верхней кнопкой “Toggle between Spectral and Waveform views”.
• Использование фильтров с указанной полосой пропускания в определенном частотном диапазоне позволяет создавать речевые сигналы, в которых содержится лишь та фонетическая информация, которая укладывается в задаваемую полосу частот. Подобные фильтры теперь заложены в стандартные программы работы со звуковым материалом, для осуществления же процедуры фильтрации надо лишь нажать несколько клавиш. Гораздо сложнее попытаться определить, какая полоса частот Вас интересует. Продуктивно работать с фильтрацией, например, при исследовании просодических характеристик (структур), при исследовании особенностей восприятия различных характеристик гласных и т.д., поскольку они расположены в сравнительно компактных нижних полосах частот. Самостоятельно попробовать разобраться в тех возможностях, которые несет фильтрация речевого сигнала, Вы сможете на примере все той же программы Cool Edit Pro. Для примера алгоритм действий по определению интересующей Вас полосы частот: в верхнем падающем меню выбираете “Transform”, “Filters”, “Scientific filter”, тип полосы “Low Pass” (остаются нижние частоты) или “High Pass” (остаются верхние частоты), наконец, в “Cut off” проставляете максимальную (или минимальную) пропускаемую частоту (т.е. частоту среза). Для проверки того, что получается в результате фильтрации, предлагаем воспользоваться не только слуховым анализом, но и инструментальным (спектральным), для чего Вы можете переключиться верхней кнопкой “Toggle between Spectral and Waveform views”.
• Лакунарные тексты представляют собой исключение из текста тех или иных фрагментов, напр., слов, словосочетаний, слогов, морфем[5]. При этом может сохраняться большая или меньшая информация о пропущенном фрагменте, напр., место, число букв или слогов и т.д. Подобный вариант искажения возможен как для звучащего текста, так и для письменного. С письменными лакунарными текстами многие из вас сталкивались при обучении иностранному языку. При подготовке звучащего лакунарного текста придется поработать, вырезая исключаемые единицы (фрагменты) или заменяя их, например, на шум.
• Десемантизация текста может быть представлена по-разному, главная задача – ослабить возможность извлекать из текста значение (напр., лексическое и/или грамматическое). С помощью этой методики мы, напр., можем попытаться разрушить целостность такой единицы как текст (переставляя предложения) или такой единицы как высказывание (переставляя словосочетания или конструкции любой величины). И этот вариант искажения возможен как для звучащего текста, так и для письменного. Для подготовки звучащего десемантизованного текста необходимо учитывать особенности просодического оформления рассматриваемых единиц и их компонентов.
• Монотонизация текста разрушает или ослабляет просодическую целостность текста и его компонентов. Подобная монотонизация может включать только выравнивание основного тона (одинаковое и ровное положение основного тона по всему тексту) или сопровождаться еще и выравниванием по интенсивности (одинаковая интенсивность по всему тексту). Осуществляется такая процедура с помощью некоторых специальных программ работы с речевым материалом. Как известно, просодические характеристики бывают трех типов: связанные с движением основного тона (мелодикой), связанные с интенсивностью (динамические) и длительностью (включая паузы). Рассматриваемые виды монотонизации исключают просодические характеристики первого типа или первого и второго типов из оформления просодических структур в тексте. Искажение долготных характеристик собственно говорения используется довольно редко из-за сложности проведения подобных модификаций. Исключение составляют паузы: их удаление легко осуществить с помощью любой программы работы со звуковым материалом. Депаузация искажает просодическую целостность структур в тексте (см. напр., Венцов, Касевич 2004).

На самом деле разнообразных экспериментальных методик исследования восприятия речи очень много. Часть из них требует не только длительной подготовки, но и сложной экспериментально-технической базы. Нейролингвистические эксперименты часто предполагают дихотическое прослушивание, когда через наушники на разные уши подается разный материал (напр., разные тексты, фразы, наборы слов). Несмотря на обилие экспериментальных методик, исследователи продолжают создавать новые и видоизменять старые.

3.2. Эксперимент в исследовании ментального лексикона

3.2.0. Общие требования

Проблемам исследования единиц и системных связей ментального лексикона посвящено немало монографий, статей, докладов. Безусловно, предлагаемое Вашему вниманию учебное пособие не ставит себе недостижимую цель полно представить концептуальные и методологические основы этих исследований. Заинтересовавшимся данными вопросами предлагаем обратиться к соответствующей литературе (в том числе, указанной в списке рекомендованной). Наиболее общий метод ассоциативного эксперимента представлен во второй части учебного пособия. В этом разделе Вашему вниманию предлагается краткое описание некоторых других, тоже важных и полезных методик. Но прежде несколько слов о том, что же такое ментальный лексикон.

На настоящий момент не существует сколько-нибудь полных модельных представлений о ментальном лексиконе. Хотя этот термин широко используется в публикациях последних лет, тем не менее единое толкование содержания соответствующего понятия фактически отсутствует.

Для того, чтобы не быть голословными приведем представления об этом термине некоторых авторов (при том, что употребляющие его авторы далеко не всегда конкретизируют собственное понимание термина). Под лексиконом понимается индивидуальный словарный запас (Баранов, Добровольский 1996: 343), «совокупность номинаций, упорядочивающих знания человека о мире» (Овчинникова 1994: 24), репрезентации слов в долговременной памяти человека (Carroll 1994: 102), хранилище слов в памяти человека и т.п. (отметим отсутствие расшифровки того, что именно представляет собой слово в языковом/речевом механизме индивида) . Под ментальным лексиконом понимается отражение в сознании человека значения лексических единиц национального языка, представляющее систему кодов и кодовых переходов от образа к слову, от картины мира к национальному языку (Залевская, 1990), т.е. ментальный лексикон входит в идиолект (индивидуальный язык личности).

Существуют определения, которые являются функционально более полными, зависящими от того, для какой цели они предназначаются, какие аспекты структуры и функционирования ментального лексикона при этом учитываются. Например, его определяют как такой компонент грамматики, который содержит всю информацию – фонологическую, морфологическую, семантическую и синтаксическую – т.е. все, что говорящие знают об отдельных словах и/или морфемах (Emmorey, Fromkin 1989: 124); как словарь значимых элементов языка, входящий в базовый компонент грамматики (Слобин 1976); как систему единиц, которые могут быть либо простыми (содержащими одну (отдельную) морфему) или составными (сложными) (содержащими несколько морфем) (Jarema et al. 1999: 1).

We will proceed from the simplest and “rectilinear” interpretation of the notion of mental lexicon: the mental lexicon is a declarative component of the language . Если все языковые и речевые знания человека разделить на декларативные и процедурные, то первые и составят то, что называют ментальным лексиконом. Иначе говоря, те единицы, которые хранятся в памяти, не будучи порождаемы в индивидуальных речевых актах, принадлежат к лексикону независимо от их формата, степени сложности внутреннего устройства и т.д. Разумеется, установление границ между тем, что “хранится”, и тем, что “порождается” – это особая сложная проблема, прежде всего, потому что эти границы могут быть подвижны, изменяясь в соответствии с избранной речевой стратегией. В качестве своеобразной иллюстрации подобной изменчивости мы можем обратиться к опыту изучения иностранного языка: в зависимости от условий коммуникации сложная для нас словоформа, по-видимому, может извлекаться непосредственно из памяти или порождаться (причем не всегда правильно).

The organization of the dictionary in a certain sense should reflect the procedures by which the transition from the text to the results of its perception and / or understanding is carried out, and, conversely, from the intention of the text to its verbal material embodiment, since the dictionary serves the processes of generation and perception of the text. It is quite obvious that in order for these processes to succeed, the dictionary must have the most “convenient” internal organization. Consequently, knowledge of the internal organization of the dictionary “cannot but be largely knowledge of how vocabulary units are used in speech activity, i.e. knowledge of the procedural type "(Ventsov, Kasevich 1994: 135). Subjects should be put in such an experimental situation that would allow them to identify the signs of words and the connections between words that a native speaker uses when identifying perceived speech and when searching for units (words and constructions) in speaking processes.

Separately, it makes sense to dwell on what we understand by language , defining the mental lexicon as the declarative component of the language. Are psycholinguistic studies of the structure and functioning of the mental lexicon central to its entry into the idiolect (the individual language of the individual) or the data obtained on the set of subjects (in other words, the set of idiolects) can reveal certain general patterns? Apparently, the answer to the proposed question depends on the goals and objectives of the study: from the particular features that characterize specific individuals to the characteristics responsible for the success of the speech-thinking activity in speakers of certain national languages.

Experimental data on the units of the human mental lexicon and the nature of the connections between them can be obtained using a variety of methods: psychophysiological, associative, experiments on free reproduction, using subjective scaling, classification, priming, etc.

3.2.1. Some traditional techniques

The subjective scaling method , borrowed from classical psychophysics, is used in this latter as one of the simplest methods for measuring the subjective distances between sensations. There are various modifications of this method: for example, a ranking method, a method of successive intervals, a method of pairwise comparisons, etc. You may need these variations on the subject of subjective scaling if you want to compare the stimuli of interest to you (for example, words or phrases) according to a given attribute , for example, based on the signs of “subjective frequency”, “use in the specified communication environment”, “development of new words”, “emotional coloring”, “similarity of meanings” and many others.

The essence of the ranking method is that the experiment participants are invited to order a certain set of stimuli according to some common trait, assigning the lowest number (rank) to the stimulus possessing the given attribute to the greatest extent. The peculiarity of this method is that the subjects work at once with the whole set of stimuli and give estimates based on a comparison of the stimuli between themselves, therefore the assessment of any stimulus depends on the assessments assigned to the other stimuli. However, the prohibition of attributing the same ranks to different stimuli forces the subjects to evaluate as different even those stimuli that appear to them to be the same. In such an experiment, the scope of the set of stimuli is substantially limited, so that the subject can work with him, keeping in his head all the assessments made earlier.

The method of consecutive intervals is that the subjects are presented with a set of stimuli and it is proposed to divide them according to a certain attribute into a given number of categories. For example, the subject is presented with a set of 100 words, which should be divided according to the frequency of occurrence into 7 categories: from “ never” to “ at every step” . The peculiarity of the method of consecutive intervals is that the subjects do not work with the whole set of stimuli at the same time, but give an assessment of each stimulus separately. The volume of the set does not impose such restrictions as in the previous case.

The method of pairwise comparisons is one of the most simple and direct methods for obtaining a matrix of semantic similarity, and, perhaps, therefore, it is very often used in psycholinguistic studies. Subjects are asked to rate the “similarity of values” (“semantic proximity”) using a certain gradual scale. The scale can be, for example, five-digit, where "0" corresponded to the minimum degree of similarity (ie, difference), and "4" - the highest; or ten. The scaling method is very labor intensive.

3.3. Some requirements for stimulus material and test instructions

It would seem that you and I have already considered some experimental techniques that are actively used in psycholinguistics. However, we almost did not dwell on questions about the stimulus material and instructions to the subjects. And they are of great importance, forming experimental conditions, which in the overwhelming number of cases are more or less significant, but differ from natural speech communication.

An important feature of the active nature of perception, including speech perception, is the clearly detectable ability of the subject to learn (the presence of the orientation phase). Therefore, during the experiment, the subject develops rules and decision criteria that allow him to most effectively cope with the task set before him in this experience. To try as much as possible to reduce and / or control such an “adjustment” to the given experimental conditions, one must pay close attention to the stimulus material and instructions in terms of how they can influence self-study.

One of the most reliable and natural results is given by the “dictation” method , when the subject is asked to write down everything that hears, and, possibly, restore distorted fragments with the maximum degree of accuracy and completeness. In the case of such an instruction, there is a high probability that the subject uses those signs and procedures (one of those) that he uses in the process of the natural speech situation. In any case, the experimenter to a lesser extent imposes others on him ..

Much more caution is to refer to the method of forced choice [6], when the experimenter imposes a set of classes (the alphabet of answers) to the subject, to which the stimuli should be referred. The difficulty lies in the fact that it is extremely difficult to correlate the number and nature of these classes (categories) with those that the subject has and are used in a natural situation.

Separately, it should be noted that some types of experiment may try to take into account the possibilities of self-learning. Thus, the behavior strategies of the subject will vary greatly depending on, for example, whether the characteristics of the stimuli, which characteristic, etc., are indicated in the instructions. Examples:

• Now you will hear the words belonging to the same text, write them down, please;
• Now you will hear the words of the Russian language, write them down, please;
• Now you will hear the words of an unknown language, write them down, please;
• Now you will hear meaningless pseudowords, write them down, please ..

These examples of fragments from instructions together with the same stimulus material can lead to activation of different speech strategies and, accordingly, to obtaining different results. Therefore, the requirements for instructions are very strict. In the event that your subjects are adults (reading) subjects, the instruction should be presented necessarily in writing, but, in addition, it is also necessary to pronounce it out loud in front of the subject. Practice shows that the answers of the subjects depend on all of this.

Experiments with distorted texts as incentives include the interaction of the process of self-learning with the natural for the perception of text process of increasing the so-called left context (already perceived piece of text), which in itself leads to greater possibilities of contextual prediction.

Requirements for stimulus material seem both obvious and difficult to meet. Obvious, because it is clear that if in a set of incentives (eg, words) some incentives or their types will be repeated more often, then they should be better predicted; a certain choice of incentives will determine a certain set of possible answers; the type of text, its subject matter suggests a more likely detection of certain words and constructions, etc. For example, the results of identifying the same English word in a set of words 1) Russian, 2) English, 3) English and Russian will be different. The probability of finding the word cognitive will be different in the text for children and for psycholinguistic students.

Requirements turn out to be difficult, because there is no ideal stimulus material, but we can strive for the most adequate in terms of the objectives of our study, the one that will provoke the student’s self-study to the least extent, the parameters that interest us, and so on. Thus, at a minimum, the following rules should be followed:

• if a set of incentives provokes a certain choice of answers, and usually it happens, in the table the incentives of interest to you should be supplemented with background ones;
• a set of incentives (in psychophysics it is called a table of incentives) must be balanced according to a certain set of features (see, for example, Stern 2001);
• incentives must be randomized.

We proceed from the assumption that there are a number of different speech strategies already in situations of natural speech communication, which, in turn, determine a number of speech signal processing modes; under experimental conditions, speech strategies and, accordingly, processing modes are added to an even less holistic model.

“So that the results obtained in the experiment could be transferred to the model, either the experimental conditions should correspond exactly to the conditions of natural perception and / or speech generation, or the rules of transition from artificial experimental to natural conditions should be known in advance” (Ventsov, Kasevich 2004). Let me add that the presence of a multiplicity of speech strategies and processing modes in natural communication situations makes the task of the researcher even more difficult.

The main task of the experimenter is the reliability of the results. The reliability of the results obtained is determined primarily by the adequacy of the methods used for the purposes of the study, the representativeness of the sample and the correctness of statistical data analysis methods. Some general recommendations are offered to your attention in the next section; their illustration using the example of an associative experiment is presented in Part 2.

4. Some general recommendations for data analysis

4.0 Some terminology issues

During the transition from data collection to analysis, we often begin to call the reactions of the subjects (in the aggregate) observations, so that in analyzing the results of both observations and experiments, we understand all the considered units (phenomena) as observations. Such terminology reflects the use of statistical methods of analysis (and a partial transition to the terminology of statistics).

The dimension of the unit under consideration is determined by the objectives of the study and the methodology of the experiment, for example, the realization of phoneme / syllable, morpheme, word, sentence, text.

Using the term sampling, we combine all the obtained and considered observations that appear as random variables in statistical analysis. The sample is a part of the so-called general population , uniting all possible implementations of this phenomenon in speech activity (available and not, existing and potential, etc., etc.). It is clear that access to the general population is fundamentally impossible for the researcher, therefore, it is necessary that the analysis of the sample allows you to get closer to the virtual analysis of the general population (understanding the realization of this phenomenon in speech activity). From here we come to the requirement of representativeness of the sample.

The representativeness (representativeness) of the sample is determined by the structure and size of the sample of observations.

The structure and volume of the sample are determined in accordance with the objectives of the study and the methodology of the experiment; they are interdependent characteristics: the more complex the sample structure, the greater the requirements for its size, so that the smallest class is represented by the permissible number of observations; conversely, the smaller the sample size, the simpler its structure should be.

The sample size is determined by the number of subjects and the number of incentives:

N = Ss * n, where Ss is the number of subjects, n is the number of stimuli,

for example, N = 20 * 40 = 800, i.e. a sample size of 800 observations was obtained at the expense of both 20 subjects and 40 stimuli.

The sample structure is set by the controlled parameters under consideration, and it can be changed in accordance with the resulting parameters. The experimenter sets the controlled parameters in advance; already at the planning stage of the experiment, the sampling structure specified by the controlled parameters, for example, the characteristics of the stimulus material and / or different groups of subjects is known. The resulting parameters may not be known in advance, and, in any case, it is not known how the sample will be divided according to the resulting parameters, since they are already determined as a result of the experiment, these are characteristics of the reaction or pairs of stimulus-response. For example, even in the case when, in the course of planning an associative experiment, we know what classification of associative links we choose, it is impossible to determine in advance their exact correlation.

In determining the structure and volume, it is necessary to take into account the requirement of sample homogeneity .

In fact, the requirement of homogeneity of the sample must be taken into account twice: first, at the stage of planning and conducting an experiment and / or observation, and, second, at the stage of data analysis.

This requirement assumes that the data that are more or less homogeneous from the point of view of both the stimulus material and the composition of the subjects are subjected to the basic analysis, for example, in many cases it would be strange to combine the data obtained from a group of subjects aged from 2 to 80 years.

4.1. Possible general recommendations

Very often we hear the question of how much sample size is permissible for a given structure (if possible), what number of subjects and what number of stimuli is the minimum possible for a given experiment. We believe that a clear answer to this question is difficult to expect. However, some considerations on this subject exist.

It is known that studies aiming at studying individual and group variability imply a large sample size, mainly due to an increase in the number of subjects; studies examining general language patterns may allow a smaller sample size. Why is this happening? What can we determine at the planning stage of an experiment, and what is not?

The answer lies in the nature of the distribution of observations and in the magnitude of the spread, i.e. consistency of the answers of the subjects. If we have reason to assume a large variation in the answers of the subjects, we are forced to recognize the need for a large sample, if not, we can try to limit ourselves to a smaller one.

Part II . Associative experiment

1. General provisions

The method of conducting an associative experiment may most vividly illustrate the traditional scheme of the psycholinguistic experiment: S → █ → R, where S is the stimulus, R is the test person's reaction, and the connection between them is through a “black box”, which symbolizes the incomprehensibility of the person being studied, the active participant - mental activity. It is not by chance that in describing this experiment, the terms “stimulus” and “reaction” become frequent and obligatory.

“A stimulus word in an associative experiment acts as an impulse that“ triggers ”the activation process in the network and results in a significant portion of it in pre-speech readiness, with the result that the subject verbalizes one of the nodes that are in pre-speech readiness (Karaulov 1993: 250). When a stimulus word is presented, the subject, as we can assume, finds himself in a situation similar to the one in which the listener (reading) is in perception of the first word of a new utterance not related to the previous (left) context and not caused by the communicative situation. . However, in both cases it is impossible to avoid a certain degree of predictability, which is determined by the left context and the communicative situation; one can only try to minimize or control it.

The stage of identification of the perceived word is obligatory in all cases, it precedes further actions of a person. В ассоциативном эксперименте данный этап как бы вычленен, а реакции, полученные от испытуемых, позволяют судить о том, какие признаки слов-стимулов оказались для них наиболее актуальными и вызвали активацию соответствующих связей в лексиконе.

Не вызывает сомнений, что разные типы реакций возникают в результате разных стратегий ассоциирования. Выбор той или иной стратегии происходит спонтанно, неосознанно. Самоконтроль и саморедактирование возможны лишь на более поздних этапах, когда отсеиваются нежелательные реакции.

Таким образом, можно считать доказанным то, что вербальные ассоциации менее всего зависят от контекста и в то же время существенно связаны со значением; что совокупность множества ассоциативных реакций, полученная в массовом эксперименте, отражает и множество возможных стратегий вербального ассоциирования; объединенные по разным основаниям в группы и подгруппы, ответы-ассоциации могут дать представление о существующих типах, а также этапах и ступенях самого процесса ассоциирования; что ассоциативный эксперимент является одной из наиболее разработанных и эффективных методик для анализа единиц ментального лексикона и принципов его организации.

Многообразие психолингвистических задач диктует необходимость использования разнообразных методик ассоциативного эксперимента. В соответствии с предлагаемой испытуемому инструкцией различают следующие виды ассоциативных экспериментов:

• Свободный ассоциативный эксперимент, в котором испытуемым предлагают отвечать словом, первым пришедшим в голову при предъявлении слова-стимула;
• Направленный ассоциативный эксперимент, в котором накладываются некоторые ограничения на выбор реакции с помощью