Introduction

Direction of the Paper

Strawson [1], as one of the main proponents of cognitive phenomenology (CP), recalls that phenomenology is understood as: “the general study, the -ology, of appearances, of the experiential character of experiences—the experiential or qualitative or what-it’s-likeness character that experiences have for those who have them as they have them.” But, with this, he considers a controversial philosophical issue. In addition to a commonplace sense/feeling character of experience, Strawson argues that there is something it is like to be/feel/experience when we think, reason, or understand, and that is called cognitive phenomenology (CP). The primary issue is that this does not have a sensory character. Strawson refers to “understanding experience”, “meaning experience”, and “thought experience” as the major elements of CP. It is important to note that Strawson treats CP as mutually exclusive, but jointly exhaustive with sensory phenomenology (SP). This indicates that CP and SP must both be considered as part of the same overall phenomenological experience. We assert that being aware of the understanding character of CP is important for researchers working on linguistic aspects of cognitive computation.

There is, however, a controversy as to whether CP actually exists as an essential element of the phenomenological state. Some have argued against this existence The Nature of Cognitive Phenomenology and Its Denial. In this paper, we introduce additional arguments for the existence of CP (particularly the understanding element) including one from neuroscientific experiments Cognitive Phenomenology: Neurolinguistic Issues. We argue that the understanding of natural language is an area that heavily depends on both sensory and putatively cognitive aspects of phenomenology.

In The Nature of Cognitive Phenomenology and Its Denial, we discuss some significant examples and additional literature on CP and its denial before adding new arguments for its existence. We refer to a study by Carruthers and Veillet [6] to highlight the different views of proponents and opponents to CP. Chudnoff [5] is cited as it relates to the irreducibility of CP.

In Cognitive Phenomenology: Neurolinguistic Issues, results in neuroscience (neurolinguistics) are introduced to demonstrate possible existence of the understanding experience, and a neurophysiological numerical measure is proposed to quantify CP. An example of sentence comprehension and its physiological markers is used to make as a concrete illustration. This implies that a concept in philosophy (CP) that relates to the special phenomenal nature of understanding can find support in existing experimental results in linguistics which may suggest a new basis for future experimental work.

While this is not a paper on computational modeling, in Computational Issues, the computational importance of the presence of CP in living systems is considered. In the concluding Conclusions, possible future developments are discussed including the possibility of measuring not only understanding experience but also thought experience and the experience of adding meaning to sensory input.

Neighboring Work

As this paper suggest that there are logical and neurological arguments relating to cognitive phenomenology, it becomes important to look at neighboring work and relate it to that in 1.1 above. First, we distinguish this paper from the concept of Cognitive Consciousness due to Bringsjord and Sundar [2]. Our stance asks whether there is something it is like to understand, find meaning, or think, while Bringsjord and Sundar [2] develop a logical theory based on formally stated axioms of what constitutes being conscious, in humans and machines. The measure of consciousness mentioned in Bringsjord and Sundar [2] is based on an analysis of the axiomatic model, whereas in the current paper, it is based on neurological measurements.

A further analytic concern with phenomenology can be found in Chrisley [3]. Referring to “Synthetic Phenomenology”, the author addresses the sensory nature of phenomenology and argues that, in an artificial system, communication could be established with a human user which is not linguistic but pictorial, based on the visual experience of the machine. For example, a visual state within a system could make sense to a human observer when displayed on a screen. CP in the current paper relates to phenomenological states that have no sensory content, as described in 1.1 above, and so create a computational/communication challenge beyond the visual.

The Nature of Cognitive Phenomenology and Its Denial

A Contrast Argument

Strawson [1] gives a contrast-based illustration of the understanding (CP) experience. Jacques, a monoglot French speaker, and Jack, a monoglot English speaker, are listening to the news in French. Jacques’ and Jack’s experiences are not the same even though they are exposed to the same stimulus. Their reported understanding experiences are different. Jacques understands what is being said, whereas Jack does not. This suggests that the CP component of their overall phenomenological state is distinct and different. Proponents of the “cognitive-experience-view” state that this difference is the understanding cognitive experience—an element of CP and not SP. Proponents of the “no-cognitive-experience-view” believe that there is, indeed, a difference in experience; however, it could be claimed that it is of a sensory phenomenological type, due to differing sensory matching properties with previous auditory memories. Experiences between individuals cannot be compared, but the responses of understanding and not, supports the presence of CP.

To contrast the above CP against sensory phenomenology (SP), consider looking at a picture, say the Mona Lisa, or listening to a piece of music, say the Beatles’ “Yesterday”. Sensory phenomenology indicates that the central subject of our experience (the “what it’s like”) is respectively the painting itself or the music itself. Such experiences include sensory and/or affective sensations evoked by the painting or the music. There may, also, be associations with why one is there looking at the painting or a memory of a previous occasion that the music was heard. It is evident that for CP the something it is like to understand, assign meaning, or be thinking are unique experiences, qualitatively different from experiencing the modality-specific phenomenology of SP.

The Continuing Debate

Montague [4] approaches CP with the purpose of having strong definitions of both CP and SP and their roles. She defines low- and high-level properties where low-level properties are used to denote properties like color, pitch, and smell. High-level properties are properties such as the ones found in natural kinds (lions and palm trees) or functional kinds (tables, beds). Accordingly, she posits a “rough” dichotomy of sensory phenomenology and cognitive phenomenology as follows:

“Sensory phenomenology is the phenomenology associated with the representation of low-level properties. Cognitive Phenomenology (CP) is the phenomenology associated with the representation of high-level properties (including the conceptual activity this representation requires or the activity associated with the deployment of concepts and abstractions)”.

She identifies three possible beliefs about the phenomenal state: (i) the “cognitive phenomenology” view which includes both sensory and cognitive experience, but differentiates between them; (ii) the “thick sensory” view which argues that the high level and low level are of the same sensory kind; and (iii) the “thin sensory” view which argues that some seeming high level is of the same kind as the low. With these beliefs, she assesses three kinds of test phenomena: first, “seeing x as x”; second, perceiving words in a language one understands; and third, visual/auditory imagery as occurs in inner speech. She concludes that only (i), involving both CP and SP, favors all three tests.

As an aside, we glance at a neurological issue. In the visual modality for instance, sensory phenomenology is known to be supported by the collaborative action of various areas of the visual cortex. In more detail, the signal is relayed from the retina through the lateral geniculate nucleus (LGN) to the primary visual cortex and to the other visual areas. It is also known that the retina contains many cells, specifically layers of photoreceptors that are specialized for different colors. So color, a low level property, is already “present” in the retina, even ahead of subcortical areas. However, if we take a “word”, the experience of seeing the letters’ shape in vision is partly low level. The high-level (or CP) experience is more difficult to position neurally. Thus, experiencing the meaning of words is a high-level or CP experience for which neurological evidence is not easily available. Hence, the example in Cognitive Phenomenology: Neurolinguistic Issues is of some import.

In further support of CP, an “irreducibility” argument is set out by Chundoff [5]. He highlights the salient irreducible nature of cognitive phenomenology, irreducibility being defined as “Some cognitive states put one in phenomenal states for which no wholly sensory states suffice”. Our arguments, herein, confirm the insufficient explanatory power of sensory phenomenology (SP) in the case of “Pure Phenomenal Contrast Arguments” such as Jacques and Jack’s experiences adding evidence for the reality of cognitive phenomenology.

The Counter-Argument

Before returning to arguments in favor of the existence of CP, a strong, and oft-quoted argument that questions cognitive phenomenology needs to be considered. It is about concepts. We note that concepts, understanding, and thought fall under cognitive phenomenology in the opinion of CP proponents. Carruthers and Veillet [6, 7], however, argue that cognitive content (the content that is part of our concepts and thoughts, not the one particular to our senses and emotions), doesn’t contribute to our “felt or subjective experience”. That is, they are not part of the overall phenomenal experience. They cite Tye [8, 9] and argue that our mental lives are only “invaded” by our perceptual (senses and emotive experiences and that those constitute our full phenomenal consciousness. They hypothesize that a “confusion” arises from the fact that we do not make the difference between the causal contributions and the constitutive contributions that concepts do to our phenomenological experience. They take bird-watching as an example. The argument is that when we learn the categories or bird types, the new experience that one feels is not CP. They claim that the change comes from the visual experience that is different when one attends to different features peculiar to each bird type. This, they contend, is not due to any “new/different” phenomenology that pervades our mental life but follows from the fact that attention, allocated to different parts of the visual scene, “causes” us to experience a change or a different visual experience. The point is about a change in overt attention,they also give another example about covert attention in the auditory modality just like the visual modality above. We could divert our attention overtly or covertly and hence undergo different verbal/auditory experiences. While this debate continues, we assume the Strawson view that CP is mutually exclusive, but jointly exhaustive with sensory phenomenology (SP) and add arguments to this view.

New Arguments in Favor

Next, two new arguments are given that corroborate the point of view that the understanding experience that exists in reading, hearing, or thinking is of a non-sensory and non-emotive type. Reflecting on the sensory experiences of sight, sound, touch, taste, and smell, we can fairly tell that they “feel” different when experienced through our sensory “input ports” compared to when we imagine them. It is noted that the difference lies in how easy it is to “imagine” in each of the cited modalities. For example, “looking at something red” feels different from when we “imagine redness”. In other words, looking at red is more “vivid” than imagining red. If we get exposed to a sound, it feels different from when we imagine this sound; it is, also, a question of vividness. However, if we understand a sentence, written or heard, the understanding experience “feels” the same if we imagine it. It is easy to check by a self-experiment that the experience is the same both ways (written/heard or imagined). Consequently, since understanding feels the same when it first occurs or through imagination, it is of a different nature from sensory phenomenology.

Next, we note that the time affects SP and CP differently. If something is not understood, a solution to a problem in physics let us say, unless the understander gathers more knowledge, the problem can remain not understood with time. Of course, this assumes that the state of knowledge of the understander remains constant and no new learning has taken place. On the other hand, a problem resolved or understood, largely stays understood, given no deficit in the understander’s mechanisms. Often quoted in the literature is that someone with conventional knowledge of algebra, faced with something like (2A = A implies A = 0), after a little effort, may enter a CP state of understanding which is unlikely to change with time. But something like A@9 = P* causes entry into a CP state of not understanding and will never be understood as the rules with which it is written are not known. SP experiences have a greater tendency to change with time. Experience of a traumatic event, a friend’s death, for instance, works differently. Even if we imagine some unhappy past event, it will feel relatively different and not as dramatic as it has been felt once, when it actually happened. Hence, flow of time impacts, differently, the understanding experience and feeling experiences; this is another reason to say that the understanding experience is not of an emotive/sensory type.

Cognitive Phenomenology: Neurolinguistic Issues

Based on our arguments so far, cognitive phenomenology refers to real experiences which are examined further here in the context of experiments on sentence comprehension carried out by a measurement on the scalp of the electrical effect of mass neuron action as a function of exposure to speech.

Sentence Comprehension: The N400 Link

Sentence comprehension at the neurophysiological level is chosen here as a suitable vehicle for corroborating the existence of CP by showing that it has a physical impact on event-related potentials neural measurements. When we hear or read a sentence, we either understand it or we do not. How does the “feeling of understanding” manifest itself in the neurophysiology of sentence comprehension? Key to our discussion is that the ERP has a strong differentiating character for speech that is understood and not understood giving credence to the reality of CP.

ERP brain imaging is achieved by the averaging of signals received by many sensors placed with a cap on points of the scalp. The trace of this average is a function of time, and specific points in time are indicated as follows. N400 means the potential is negative at 400 ms after the presentation of some external sensory event (a word of a spoken sentence in the current case). An example appears below in Fig. 1. P600 means that the trace is positive at 600 ms following stimulus onset (word). So for sentence comprehension, two ERPs, the N400 (negative signal at 400 ms) and the P600 (positive signal at 600 ms), are most important. The N400 is measured in the time-window 300–500 ms after each word of the sentence is input; its peak activity occurs around 400 ms, hence its name. The word under study is named (CW) for “critical word”.

Fig. 1
figure 1

The EEG (N400) while reading the three sentences [10] reprinted with permission

Full size image

The N400 and the P600 are measured for each word (constituting a certain sentence) as it is heard or read. Classical literature has reported the significance of the N400 ERP in revealing brain activity peculiar to semantics or meaning [10,11,12,13,14]. The N400 and P600 measurements have been performed with people who are recruited specifically to conduct such studies/experiments. Simply stated, the participants wear the EEG apparatus on their heads and are exposed to sentences, either to speech (hearing) or to text (reading); the EEG measurements, the N400 value (for each word/CW), is picked up at specific locations or electrodes peculiar to each modality.

A simple example is considered in the following two sentences; the critical word is underlined (in this case, it is the last word):

  1. a.

    The boy ate the cake.

  2. b.

    The boy ate the fear.

The N400 value of the CW (cake and fear in our examples) reflects two activities; in the terminology of Kutas and Federmeier [14], these activities are named as follows:

  1. 1.

    Pre-lexical, which is about word access. The N400 value is larger when access to the CW is, relatively, difficult (larger in absolute value, so a drop compared to a baseline activity, the control, representing another CW of easier access). The more difficult the lexical access is, the bigger is the value of the deflection of the N400 [11, 12, 14, 15].

  2. 2.

    Post-lexical, which is about integration. The process of integration means construction of the meaning of the sentence to which the new word (CW) has been added. Similarly, the value of the N400 (the negative deflection) is bigger or smaller when integration is more or less difficult [12, 14,15,16,17). In other words, the bigger/smaller the value of the N400 deflection, the less/more meaningful is the sentence; hence, the sentence is less understandable or more understandable as a function of the CW’s N400 value.

Examples of sentences from Fig. 1 in Hagoort et al. [10] are used in Fig. 1 below.

Figure 1 is used here to illustrate the mentioned measurements about the N400. Taking the case about semantic plausibility, case 3 (in red) constitutes a semantic violation (the red dashed line). Critical words (CW) are underlined. Accordingly, when, in case 3, the CW is input (sour), the N400 is more negative (implausible sentence) than in case 1, and the black line constitutes the plausible, hence, understandable sentence.

To summarize, we introduced the post-lexical “dimension” of the N400 as a potential neurophysiological indicator and numerical value for CP. That is, through reflecting the integration process, the post-lexical N400 value relates to meaning construction, i.e., the “understanding experience”. This suggests that CP is real in the sense of being measurable. This work is one of the clearest indications of a neural experiential measure (the N400 ERP) of CP. Future studies could profitably focus on experimentation and show how helpful is this suggestion.

The present investigation adds to other attempts at finding the neural correlates and brain processing regions for language and meaning construction: Fedorenko et al. [18] and Lau et al. [19] are two examples among many. Cortical localization of auditory sentence comprehension as in Friederici [20] and story comprehension traces captured from the brain as in Wehbe et al. [21] are also valuable attempts aiming at constructing the anatomical pathways for language comprehension.

The P600 Option

In addition to the consideration of the N400 ERP, a recent line of research takes specific cases of detected N400 and P600 as a basis to build a new functional interpretation of the N400 and the P600 ERP’s in language processing.

Brouwer et al. [22] and Aurnhammer et al. [23] state that the N400, long considered as being related to semantics and supposed to mirror two brain processing functions of lexical access (pre-lexical) and integration (post-lexical), reflects only one process. According to this relatively new interpretation, N400 is said to refer only to lexical access, while the P600 (long considered to reflect syntactic processing) is said to be linked to integration. Many previous research articles had alluded to such special effects (P600 effects); however, Brouwer et al. [22] have a remarkable work that proposes a new interpretation for both ERPs,it is called the “RI” model, where “R” is for retrieval, related to the N400, and “I” is for integration, related to the P600. Whether their claim or point of view is valid or not, it is noted that a change of model (function of N400 and P600) does not affect the key issue of the current paper: neuroscience favors the argument for a concreteness of CP understanding.

Computational Issues

The debate over the existence of CP is mainly sustained among philosophers. The object of this paper, however, has been to involve a wider audience in the discussion including those engaged in designing artificial language understanding systems. Aleksander [24] has proposed that, in a dynamic neural network language learner, the distinction between understanding and not reflects the “felt” state structure in the network. Learning is assumed to create depictive states and state structures. Then, SP leads to a concern for the content of states in a structure and CP to a concern for the shape of the trajectories within a structure. This still requires considerable work but begins to make a case for separate theoretical and practical functions of SP and CP.

Of course, natural language “understanding” in artificially intelligent systems is often successful in a practical sense and much used. Also neural networks have been used to model N400/N600 responses [25] but the possible interplay of SP and CP does not appear in this research. For example, the role and measure of CP experiential states in learning situations are a ripe topic for research.

The key advantage of being aware of CP in its understanding mode is in the design of cognitive computing systems. A system might use “understood or not understood” as a measure at a particular point in time which determines further behavior in a way that is close to the processes of a human agent. For example, a vector of symptoms in a medical advice generator may be assigned an understood or not value which opens a measured approach to accurate action advice. Similar parameters of the quality of thought or meaning assignment given by an artificial CP may be beneficial in system design.

Conclusions

In this paper, we have reviewed the philosophical concept of cognitive phenomenology as the “what’s it like” in cognitive processes such as understanding, meaning assignment, or thought. This stands in contrast to sensory phenomenology where “what’s it like” refers to current or past sensory experience. We have also introduced the debate within philosophy where a faction argues that phenomenal experiences of the sensory kind suffice for an understanding of cognition and that CP does not exist. We have presented several arguments that support the existence and function of CP. Specifically, we described work on event-related potentials N400 and P600 in sentence comprehension which indicates measured trace differences for input that is understood and that which is not. This is an example of an understanding experience which shows that CP has a physical confirmation.

We further note that a consideration of the influence that phenomenology can have for designers of intelligent systems (e.g., Chrisley [3] could be augmented by considering an interplay between SP and CP. We suggest there is interesting research in further understanding the way in which SP and CP work together in human cognitive systems and what benefit might accrue in having similar arrangements in artificial cognitive systems. That is, as most computational work on phenomenology is in the SP domain, we here encourage computational modelers to add CP to their considerations.

Finally, in this study, we have proposed a novel way to quantify or measure understanding through measurement of neural activity. This encourages extending the research to other elements of CP, such as thought, to further bridge the gap between philosophy, linguistics, math, and computation.