Module u1: Speech in the Interface 6: Human Communication Jacques Terken SAI User-System Interaction U1, Speech in the interface: 6. Human communication 1 Motivation provide knowledge about basic concepts of human communication relevant to the design of speech interfaces explore use of knowledge about human communication as model for man-machine interaction SAI User-System Interaction U1, Speech in the interface: 6. Human communication 2 Contents basic scheme for communication the structure of messages the structure of communication “how to do things with words”1: communication as process implications for man-machine interaction key points human-human communication as model for human-computer interaction 1 J.L. Austin: “How to do things with words”, 1962 SAI User-System Interaction U1, Speech in the interface: 6. Human communication 3 basic scheme context sender channel context receiver feedback Context – Situational context situated communication: time, place, objects – Social context – Cultural context – Dialogue context SAI User-System Interaction U1, Speech in the interface: 6. Human communication 4 intentions and effects Sender: intention to convey a message message needs to be encoded Receiver: decoding effect Ideal communication: intention = effect Distortions underway: intention effect’ Furthermore: effect is more than intention (sender also communicates unintentional information) complex messages require sequence of steps communication as structured action / collaboration SAI User-System Interaction U1, Speech in the interface: 6. Human communication 5 Contents basic scheme for communication the structure of messages the structure of communication communication as process implications for man-machine interaction key points human-human communication as model for human-computer interaction SAI User-System Interaction U1, Speech in the interface: 6. Human communication 6 formal aspects (1) Simple case: messages containing a single dialogue act (inform, request, confirmation etc) Sentence/utterance – put the red block before the yellow block next to the ball – I herewith pronounce you man and wife – What time does the next flight to Boston leave? Sentences made up of word groups Word groups made up of words Words made up of syllables and phonemes (~40) or graphemes (~25) / @, e, a, r, … / ; { e, e, a, r, … } SAI User-System Interaction U1, Speech in the interface: 6. Human communication 7 formal aspects (2) rules determine well-formedness of representations at different levels: – pragmatics: sentence meaning in context (speech act, dialogue act) contextual meanings *the king of France is bald (but may be okay in a story) – semantics: sentence meanings possible meanings *the 50 year old bachelor surprised his wife – syntax: sentence structure possible sentences – morphology: word structure possible words – Dutch: loopje; *loopetje; kommetje; *kompje – Phonology: sound structure possible sound sequences Dutch: herfst; Polish: przewalski SAI User-System Interaction U1, Speech in the interface: 6. Human communication 8 formal aspects (3) mapping between levels not fully specified (not one - to - one): ambiguity Mapping sounds onto words: /Aiscri:m/ {I scream} or {Ice cream} Mapping words onto sentence structure: Syntactic ambiguities i saw the girl with the binoculars SAI User-System Interaction U1, Speech in the interface: 6. Human communication 9 put the red block before the yellow block next to the ball (1) put the red block before the yellow block [ that is ] next to the ball (2) put the red block before the yellow block, next to the ball SAI User-System Interaction U1, Speech in the interface: 6. Human communication 10 Contents basic scheme for communication the structure of messages the structure of communication communication as process implications for man-machine interaction key points human-human communication as model for human-computer interaction SAI User-System Interaction U1, Speech in the interface: 6. Human communication 11 levels of analysis topic … topic … topic transaction exchange exchange exchange turn turn turn move move move SAI User-System Interaction U1, Speech in the interface: 6. Human communication 12 transaction (part of) conversation dealing with a single topic consists of one or more exchanges SAI User-System Interaction U1, Speech in the interface: 6. Human communication 13 exchange smallest interactive unit consists minimally of two turns produced by two different speakers involves the negotiation of a single piece of information examples: request - inform - confirm presentation - acceptance SAI User-System Interaction U1, Speech in the interface: 6. Human communication 14 adjacency pairs: question answer request accept / turn down invitation accept / decline greeting greeting apology smoother Interactive pressure: opening move induces obligation to answer with closing move (conversational obligation) SAI User-System Interaction U1, Speech in the interface: 6. Human communication 15 nesting of adjacency pairs: s: question X what time is the next train to amsterdam a: question Y amsterdam central? s: answer Y yes a: answer X in seven minutes at platform 5 mixed initiative vs. system initiative SAI User-System Interaction U1, Speech in the interface: 6. Human communication 16 turn Turn: everything the speaker says before the next speaker takes over Rules for turn-taking behaviour 1. Speaker selects next speaker 2. Next speaker selects self Window of opportunity for turn switching (minimizes overlap and interruption; interruptions only permissible in particular circumstances Turn consists of one or more moves SAI User-System Interaction U1, Speech in the interface: 6. Human communication 17 temporal aspects of turn-taking dialog not as a sequence of well-formed non-overlapping turns: – time window for turn taking – overlap, interruptions SAI User-System Interaction U1, Speech in the interface: 6. Human communication 18 move/dialog act no fixed inventory, but dependent on application Common types: Dialogue control acts Primary dialogue acts – acknowledge – request for action – check – inform – confirm – question – greeting – thanks Game theory conceives of communication as a sequence of moves Expresses the assumption that communication is governed by a set of rules SAI User-System Interaction U1, Speech in the interface: 6. Human communication 19 Indirect communication No one-to-one mapping between form and function: indirect communication (politeness forms) Questions: – Can you tell me what time it is (yes/no question) – I would like to know what time it is (statement) – Do you have a watch (yes/no question with indirect communication ) Direct communication: – What time is it (wh-question) – Please tell me the time (request) SAI User-System Interaction U1, Speech in the interface: 6. Human communication 20 spoken dialog systems Identify dialog act performed by user in order to decide what dialog act to produce as an an answer Machine learning techniques in order to learn best continuation in context x (and to learn best path) SAI User-System Interaction U1, Speech in the interface: 6. Human communication 21 Contents basic scheme for communication the structure of messages the structure of communication communication as process implications for man-machine interaction key points human-human communication as model for human-computer interaction SAI User-System Interaction U1, Speech in the interface: 6. Human communication 22 spoken communication as joint action/cooperation Speech is volatile Mechanisms to cope with volatile nature of speech co-operation receiver must be in receiving mode, i.e., be open to message sender must adjust to channel and receiver characteristics SAI User-System Interaction U1, Speech in the interface: 6. Human communication 23 spoken communication as joint action/cooperation (2) Goal achievement is usually accomplished through a number of steps involving both participants: dialogue Monitoring of transmission success by sender on the basis of conscious and unconscious feedback from receiver Immediate action in case of problems either by user or sender – Role of eye gaze in staying in touch, monitoring and turn taking SAI User-System Interaction U1, Speech in the interface: 6. Human communication 24 speech generation two simultaneous processes: – conceptualisation (“what to say”: content) – formulation (“how to say it”: converting content into utterance) simultaneous, but formulation lagging behind ( spontaneous speech phenomena) – restarts, self-corrections, hesitations (silence, ehm), disfluencies – grammar of spoken language SAI User-System Interaction U1, Speech in the interface: 6. Human communication 25 message formulation: issues continuity, coherence: what we say usually links to what was said before: context provides search space for interpretation and allows for efficient expression – pronouns, anaphora – prosody – bridging: establishing a relation between expressions in consecutive utterances can you take the picnic supplies out of the van? the beer is probably warm, so please put it in the fridge. SAI User-System Interaction U1, Speech in the interface: 6. Human communication 26 grice’s cooperativity principle human communication (supposed to be) governed by grice’s cooperativity principle, consisting of the following maxims (guidelines for communication): quantity: make your contribution as informative as is required I want a flight; I want a flight from A to B because …. quality: do not say what you believe is false or for which you lack adequate evidence relation: be relevant manner: be perspicuous, brief and orderly; avoid obscurity and ambiguity SAI User-System Interaction U1, Speech in the interface: 6. Human communication 27 speech comprehension Stages in comprehension: – Perception: perceiving sounds and mapping sounds onto words – Interpretation – Evaluation – From a processing point of view perception and interpretation are closely interlinked keywords: on-line processing, hypothesis testing and integrating SAI User-System Interaction U1, Speech in the interface: 6. Human communication 28 backchannelling “utterances” and other communicative signals guiding the progress of the communication, i.e., serving a control function – nodding – yes, hm-hm – “madrid?” SAI User-System Interaction U1, Speech in the interface: 6. Human communication 29 error recovery on-line property allows fast and efficient recovery: – problems noticed immediately – cause of problem usually easily identified “what did you say?” “did you say …” “what do you mean” SAI User-System Interaction U1, Speech in the interface: 6. Human communication 30 common ground Grounding agreeing about current state of the dialogue Important role of feedback Shared knowledge + knowledge that speaker knows is available to hearer + world knowledge (“living memory”) + knowledge concerning context of communication and what has been established by previous part of discourse SAI User-System Interaction U1, Speech in the interface: 6. Human communication 31 Contents basic scheme for communication the structure of messages the structure of communication communication as process implications for man-machine interaction key points human-human communication as model for human-computer interaction SAI User-System Interaction U1, Speech in the interface: 6. Human communication 32 human communication no neat succession of turns: overlap in time: provision for speech recognition needed Indirect communication: message function and form no 1:1 relation interpretation dependent on domain knowledge: requires domain modelling: successful for restricted domains interpretation dependent on context: requires discourse model SAI User-System Interaction U1, Speech in the interface: 6. Human communication 33 supporting the interaction Prompting: system initiative Zooming: increase amount of feedforward in system prompt – How may I help you – Which connection – Are you interested in connections or prices – Please respond with yes or no Cooperative replies – User: “Do you have flights between 7:15 and 7:45?” – S: “No, but there is a flight at 7:55” SAI User-System Interaction U1, Speech in the interface: 6. Human communication 34 robustness issue Lack of robustness creates need for verifcation Explicit verification: – Did you say “amsterdam”? Implicit verification – U: “I want to travel from .. to rotterdam” – S: “when do you want to travel to rotterdam?” – Cf: S: “when do you want to travel to amsterdam?” SAI User-System Interaction U1, Speech in the interface: 6. Human communication 35 Verification is costly when you got it right Feedback (positive feedback or implicit verification) is misleading when you good it wrong Use verfication when uncertain Use feedback when confident SAI User-System Interaction U1, Speech in the interface: 6. Human communication 36 confidence-based verification High confidence: just accept or implicit verification Medium confidence: explicit verification Low confidence: explicit request for repetition Precision/recall issue for correctness and confidence measure: – Confidence measure may not be right: either too high or too low SAI User-System Interaction U1, Speech in the interface: 6. Human communication 37 Contents basic scheme for communication the structure of messages the structure of communication communication as process implications for man-machine interaction key points human-human communication as model for human-computer interaction SAI User-System Interaction U1, Speech in the interface: 6. Human communication 38 communication as joint action (cooperation) – taking into consideration listener needs – listening for intention rule-governed behaviour at many different levels – structure of messages – structure of conversations (e.g. turn-taking behaviour) grounding feedback (back-channelling) SAI User-System Interaction U1, Speech in the interface: 6. Human communication 39 Contents basic scheme for communication the structure of messages the structure of communication communication as process implications for man-machine interaction key points human-human communication as model for human-computer interaction SAI User-System Interaction U1, Speech in the interface: 6. Human communication 40 cooperative assistant Characteristics of cooperative assistant – Interprets messages in terms of intentions not in terms of form (“c.a. is not a politician or lawyer”) – Cooperative replies (not blunt “no” but offers alternatives, e.g. through constraint relaxation) – Context-sensitive replies and help No global help manual – Proactive, triggered by implicit cues Does not await explicit request for help but anticipates SAI User-System Interaction U1, Speech in the interface: 6. Human communication 41 discussion Is this a good model for human computer interaction? Considerations: – Schneiderman c.s.: user should stay in control – Maes/Sidner: • computers should be able to understand our intentions so that we don’t have to specify all the actions explicitly ourselves • Computers should monitor our actions so that they can learn how to anticipate our needs and preferences SAI User-System Interaction U1, Speech in the interface: 6. Human communication 42